A machine learning researcher is claiming to have knowledge of kinky drug-fueled orgies in Silicon Valley's storied hacker houses and appears to be linking those parties, and the culture surrounding them, to OpenAI.

"The thing about being active in the hacker house scene is you are accidentally signing up for a career as a shadow politician in the Silicon Valley startup scene," begins the lengthy X-formerly-Twitter post by Sonia Joseph, a former Princeton ML researcher who's now affiliated with the deep learning institute Mila Quebec.

What follows is a vague and anecdotal diatribe about the "dark side" of startup culture made particularly explosive by Joseph's reference to so-called "consensual non-consent" sex parties that she says took place within the artificial general intelligence (AGI) enthusiast community in the valley.

The jumping off point, as far as we can tell, stems from a thread announcing that OpenAI superalignment chief Jan Leike was leaving the company as it dissolved his team that was meant to prevent advanced AI from going rogue.

At the end of his X thread, Leike encouraged remaining employees to "feel the AGI," a phrase that was also ascribed to newly-exited OpenAI cofounder Ilya Sutskever during seemingly cultish rituals revealed in an Atlantic expos last year but nothing in that piece, nor the superalignment chief's tweets, suggests anything having to do with sex, drugs, or kink.

Still, Joseph addressed her second viral memo-length tweet "to the journalists contacting me about the AGI consensual non-consensual (cnc) sex parties." And in the post, said she'd witnessed "some troubling things" in Silicon Valley's "community house scene" when she was in her early 20s and new to the tech industry.

"It is not my place to speak as to why Jan Leike and the superalignment team resigned. I have no idea why and cannot make any claims," wrote the researcher, who is not affiliated with OpenAI. "However, I do believe my cultural observations of the SF AI scene are more broadly relevant to the AI industry."

"I don't think events like the consensual non-consensual (cnc) sex parties and heavy LSD use of some elite AI researchers have been good for women," Joseph continued. "They create a climate that can be very bad for female AI researchers... I believe they are somewhat emblematic of broader problems: a coercive climate that normalizes recklessness and crossing boundaries, which we are seeing playing out more broadly in the industry today. Move fast and break things, applied to people."

While she said she doesn't think there's anything generally wrong with "sex parties and heavy LSD use," she also charged that the culture surrounding these alleged parties "leads to some of the most coercive and fucked up social dynamics that I have ever seen."

"I have seen people repeatedly get shut down for pointing out these problems," Joseph wrote. "Once, when trying to point out these problems, I had three OpenAI and Anthropic researchers debate whether I was mentally ill on a Google document. I have no history of mental illness; and this incident stuck with me as an example of blindspots/groupthink."

"Its likely these problems are not really on OpenAI but symptomatic of a much deeper rot in the Valley," she added. "I wish I could say more, but probably shouldnt."

Overall, it's hard to make heads or tails of these claims.We've reached out to Joseph and OpenAI for more info.

"I'm not under an NDA. I never worked for OpenAI," Joseph wrote. "I just observed the surrounding AI culture through the community house scene in SF, as a fly-on-the-wall, hearing insider information and backroom deals, befriending dozens of women and allies and well-meaning parties, and watching many them get burned."

More on OpenAI: Sam Altman Clearly Freaked Out by Reaction to News of OpenAI Silencing Former Employees

Read more here:
Machine Learning Researcher Links OpenAI to Drug-Fueled Sex Parties - Futurism

The advent of AI has been hailed by many as a societal game-changer, as it opens a universe of possibilities to improve nearly every aspect of our lives.

Astronomers are now using AI, quite literally, to measure the expansion of our universe.

Two recent studies led by Maria Dainotti, a visiting professor with UNLVs Nevada Center for Astrophysics and assistant professor at the National Astronomical Observatory of Japan (NAOJ), incorporated multiple machine learning models to add a new level of precision to distance measurements for gamma-ray bursts (GRBs) the most luminous and violent explosions in the universe.

In just a few seconds, GRBs release the same amount of energy our sun releases in its entire lifetime. Because they are so bright, GRBs can be observed at multiple distances including at the edge of the visible universe and aid astronomers in their quest to chase the oldest and most distant stars. But, due to the limits of current technology, only a small percentage of known GRBs have all of the observational characteristics needed to aid astronomers in calculating how far away they occurred.

Dainotti and her teams combined GRB data from NASAs Neil Gehrels Swift Observatory with multiple machine learning models to overcome the limitations of current observational technology and, more precisely, estimate the proximity of GRBs for which the distance is unknown. Because GRBs can be observed both far away and at relatively close distances, knowing where they occurred can help scientists understand how stars evolve over time and how many GRBs can occur in a given space and time.

This research pushes forward the frontier in both gamma-ray astronomy and machine learning, said Dainotti. Follow-up research and innovation will help us achieve even more reliable results and enable us to answer some of the most pressing cosmological questions, including the earliest processes of our universe and how it has evolved over time.

In one study, Dainotti and Aditya Narendra, a final-year doctoral student at Polands Jagiellonian University, used several machine learning methods to precisely measure the distance of GRBs observed by the space Swift UltraViolet/Optical Telescope (UVOT) and ground-based telescopes, including the Subaru Telescope. The measurements were based solely on other, non distance-related GRB properties. The research was published May 23 in the Astrophysical Journal Letters.

The outcome of this study is so precise that we can determine using predicted distance the number of GRBs in a given volume and time (called the rate), which is very close to the actual observed estimates, said Narendra.

Another study led by Dainotti and international collaborators has been successful in measuring GRB distance with machine learning using data from NASAs Swift X-ray Telescope (XRT) afterglows from what are known as long GRBs. GRBs are believed to occur in different ways. Long GRBs happen when a massive star reaches the end of its life and explodes in a spectacular supernova. Another type, known as short GRBs, happens when the remnants of dead stars, such as neutron stars, merge gravitationally and collide with each other.

Dainotti says the novelty of this approach comes from using several machine-learning methods together to improve their collective predictive power. This method, called Superlearner, assigns each algorithm a weight whose values range from 0 to 1, with each weight corresponding to the predictive power of that singular method.

The advantage of the Superlearner is that the final prediction is always more performant than the singular models, said Dainotti. Superlearner is also used to discard the algorithms which are the least predictive.

This study, which was published Feb. 26 in The Astrophysical Journal, Supplement Series, reliably estimates the distance of 154 long GRBs for which the distance is unknown and significantly boosts the population of known distances among this type of burst.

A third study, published Feb. 21 in the Astrophysical Journal Letters and led by Stanford University astrophysicist Vah Petrosian and Dainotti, used Swift X-ray data to answer puzzling questions by showing that the GRB rate at least at small relative distances does not follow the rate of star formation.

This opens the possibility that long GRBs at small distances may be generated not by a collapse of massive stars but rather by the fusion of very dense objects like neutron stars, said Petrosian.

With support from NASAs Swift Observatory Guest Investigator program (Cycle 19), Dainotti and her colleagues are now working to make the machine learning tools publicly available through an interactive web application.

Go here to read the rest:
Cosmic Leap: NASA Swift Satellite and AI Unravel the Distance of the Farthest Gamma-Ray Bursts - UNLV NewsCenter

VanLehn, K. Cognitive skill acquisition. Ann. Rev. Psychol. 47, 513539 (1996).

Article Google Scholar

DeKeyser, R. in Skill Acquisition Theory 83104 (Routledge, 2020).

Tabibian, B. et al. Enhancing human learning via spaced repetition optimization. Proc. Natl Acad. Sci. USA 116, 39883993 (2019).

Article MathSciNet Google Scholar

Evans, N. J., Brown, S. D., Mewhort, D. J. & Heathcote, A. Refining the law of practice. Psychol. Rev. 125, 592 (2018).

Article Google Scholar

Heathcote, A., Brown, S. & Mewhort, D. J. The power law repealed: the case for an exponential law of practice. Psychon. Bull. Rev. 7, 185207 (2000).

Article Google Scholar

Shrager, J., Hogg, T. & Huberman, B. A. A graph-dynamic model of the power law of practice and the problem-solving fan-effect. Science 242, 414416 (1988).

Article Google Scholar

Wixted, J. T. The enigma of forgetting. Proc. Natl Acad. Sci. USA 119, e2201332119 (2022).

Article Google Scholar

Averell, L. & Heathcote, A. The form of the forgetting curve and the fate of memories. J. Math. Psychol. 55, 2535 (2011).

Article MathSciNet Google Scholar

Roediger III, H. L. Relativity of remembering: why the laws of memory vanished. Annu. Rev. Psychol. 59, 225254 (2008).

Article Google Scholar

Chiaburu, D. S. & Marinova, S. V. What predicts skill transfer? An exploratory study of goal orientation, training self-efficacy and organizational supports. Int. J. Train. Dev. 9, 110123 (2005).

Article Google Scholar

Sturm, L. P. et al. A systematic review of skills transfer after surgical simulation training. Ann. Surgery 248, 166179 (2008).

Article Google Scholar

Logan, G. D. Toward an instance theory of automatization. Psycho. Rev. 95, 492 (1988).

Article Google Scholar

Logan, G. D. Shapes of reaction-time distributions and shapes of learning curves: a test of the instance theory of automaticity. J. Exp. Psychol. 18, 883 (1992).

Google Scholar

Anderson, J. R. Acquisition of cognitive skill. Psychol. Rev. 89, 369 (1982).

Article Google Scholar

Tenison, C. & Anderson, J. R. Modeling the distinct phases of skill acquisition. J. Exp. Psychol. 42, 749 (2016).

Google Scholar

Tenison, C., Fincham, J. M. & Anderson, J. R. Phases of learning: how skill acquisition impacts cognitive processing. Cognitive Psychol. 87, 128 (2016).

Article Google Scholar

Jordan, M. I. Serial order: a parallel distributed processing approach. Adv. Pyschol. 121, 471495 (1997).

McClelland, J. L. et al. Parallel Distributed Processing Vol. 2 (MIT Press, 1986).

Young, R. M. & Lewis, R. L. in Models of Working Memory: Mechanisms of Active Maintenance and Executive Control 224256 (Cambridge Univ. Press, 1999).

Anderson, J. R., Matessa, M. & Lebiere, C. in HumanComputer Interaction Vol. 12, 439462 (Lawrence Erlbaum Associates, 1997).

Ritter, F. E., Tehranchi, F. & Oury, J. D. ACT-R: a cognitive architecture for modeling cognition. WIREs Cogn. Sci. 10, e1488 (2019).

Article Google Scholar

Goldstone, R. L. & Lupyan, G. Discovering psychological principles by mining naturally occurring data sets. Topics Cogn. Sci. 8, 548568 (2016).

Article Google Scholar

Jenkins, J. J., Cermak, L. & Craik, F. in Levels of Processing in Human Memory 429446 (Pyschology, 1979).

Udrescu, S.-M. & Tegmark, M. AI Feynman: a physics-inspired method for symbolic regression. Sci. Adv. 6, eaay2631 (2020).

Article Google Scholar

Wang, H. et al. Scientific discovery in the age of artificial intelligence. Nature 620, 4760 (2023).

Article Google Scholar

Schmidt, M. & Lipson, H. Distilling free-form natural laws from experimental data. Science 324, 8185 (2009).

Article Google Scholar

Cranmer, M. et al. Discovering symbolic models from deep learning with inductive biases. In 34th Conference on Neural Information Processing Systems 1742917442 (NeurIPS, 2020).

Rudy, S. H., Brunton, S. L., Proctor, J. L. & Kutz, J. N. Data-driven discovery of partial differential equations. Sci. Adv. 3, e1602614 (2017).

Article Google Scholar

Chen, Z., Liu, Y. & Sun, H. Physics-informed learning of governing equations from scarce data. Nat. Commun. 12, 6136 (2021).

Article Google Scholar

Margraf, J. T., Jung, H., Scheurer, C. & Reuter, K. Exploring catalytic reaction networks with machine learning. Nat. Catal. 6, 112121 (2023).

Article Google Scholar

Han, Z.-K. et al. Single-atom alloy catalysts designed by first-principles calculations and artificial intelligence. Nat. Commun. 12, 1833 (2021).

Article Google Scholar

Wang, Y., Wagner, N. & Rondinelli, J. M. Symbolic regression in materials science. MRS Commun. 9, 793805 (2019).

Article Google Scholar

He, M. & Zhang, L. Machine learning and symbolic regression investigation on stability of mxene materials. Comput. Mater. Sci. 196, 110578 (2021).

Article Google Scholar

LeCun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436444 (2015).

Article Google Scholar

Vaswani, A. et al. Attention is all you need. In Advances in Neural Information Processing Systems 30 (NIPS, 2017).

Baker, F. B. The Basics of Item Response Theory (ERIC, 2001).

Hambleton, R. K., Swaminathan, H. & Rogers, H. J. Fundamentals of Item Response Theory Vol. 2 (Sage, 1991).

Swaminathan, H. & Gifford, J. A. Bayesian estimation in the three-parameter logistic model. Psychometrika 51, 589601 (1986).

Article MathSciNet Google Scholar

Maris, G. & Bechger, T. On interpreting the model parameters for the three parameter logistic model. Measurement 7, 7588 (2009).

Google Scholar

Hornik, K. Approximation capabilities of multilayer feedforward networks. Neural Netw. 4, 251257 (1991).

Article Google Scholar

Raghu, M., Poole, B., Kleinberg, J., Ganguli, S. & Sohl-Dickstein, J. On the expressive power of deep neural networks. In Proc. 34th International Conference on Machine Learning 28472854 (PMLR, 2017).

Cybenko, G. Approximation by superpositions of a sigmoidal function. Math. Control Signals Syst. 2, 303314 (1989).

Article MathSciNet Google Scholar

Spearman, C. General intelligence, objectively determined and measured. Am. J. Psychol. 15, 201293 (1904).

Article Google Scholar

Eid, M., Geiser, C., Koch, T. & Heene, M. Anomalous results in g-factor models: explanations and alternatives. Psychol. Methods 22, 541 (2017).

Article Google Scholar

Steyvers, M. & Schafer, R. J. Inferring latent learning factors in large-scale cognitive training data. Nat. Hum. Behav. 4, 11451155 (2020).

Article Google Scholar

Simons, D. J. et al. Do brain-training programs work?. Psychol. Sci. Public Interest 17, 103186 (2016).

Article Google Scholar

Kievit, R. A. et al. Mutualistic coupling between vocabulary and reasoning supports cognitive development during late adolescence and early adulthood. Psychol. Sci. 28, 14191431 (2017).

Article Google Scholar

Kievit, R. A., Hofman, A. D. & Nation, K. Mutualistic coupling between vocabulary and reasoning in young children: a replication and extension of the study by Kievit et al. (2017). Psychol. Sci. 30, 12451252 (2019).

Article Google Scholar

Fisher, R. A. Design of experiments. Brit. Med. J. 1, 554 (1936).

Article Google Scholar

Kumar, A., Benjamin, A. S., Heathcote, A. & Steyvers, M. Comparing models of learning and relearning in large-scale cognitive training data sets. NPJ Sci. Learn. 7, 24 (2022).

Article Google Scholar

Liu, R. & Koedinger, K. R. Towards reliable and valid measurement of individualized student parameters. In Proc.10th International Conference on Educational Data Mining 135142 (International Educational Data Mining Society, 2017).

Koedinger, K. R., Carvalho, P. F., Liu, R. & McLaughlin, E. A. An astonishing regularity in student learning rate. Proc. Natl Acad. Sci. USA 120, e2221311120 (2023).

Article Google Scholar

Neath, A. A. & Cavanaugh, J. E. The bayesian information criterion: background, derivation, and applications. WIREs Comput. Stats 4, 199203 (2012).

Article Google Scholar

Vrieze, S. I. Model selection and psychological theory: a discussion of the differences between the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). Psychol. Methods 17, 228 (2012).

Article Google Scholar

Paszke, A. et al. PyTorch: an imperative style, high-performance deep learning library. In Proc. 33rd International Conference on Neural Information Processing Systems 80268037 (Curran Associates Inc., 2019).

Kingma, D. P. & Ba, J. Adam: a method for stochastic optimization. Preprint at https://arxiv.org/abs/1412.6980 (2015).

Steyvers, M. & Benjamin, A. S. The joint contribution of participation and performance to learning functions: exploring the effects of age in large-scale data sets. Behav. Res. Methods 51, 15311543 (2019).

Article Google Scholar

The rest is here:
Automated discovery of symbolic laws governing skill acquisition from naturally occurring data - Nature.com

Summary

Machine learning is sometimes considered a subset of overarching AI. But in the context of digital security, it may be better understood as a driving force, the fuel powering the engine.

The terms AI and machine learning are often used interchangeably by professionals outside the technology, managed IT and cybersecurity trades. But, truth be told, they are separate and distinct tools that can be coupled to power digital defense systems and frustrate hackers.

Artificial iIntelligence has emerged as an almost ubiquitous part of modern life. We experience its presence in everyday household robots and the familiar Alexa voice that always seems to be listening. Practical uses of AI mimic and take human behavior one step further. In cybersecurity, it can deliver 24/7 monitoring, eliminating the need for a weary flesh-and-blood guardian to stand a post.

Machine learning is sometimes considered a subset of overarching AI. But in the context of digital security, it may be better understood as a driving force, the fuel powering the engine. Using programmable algorithms, it recognizes sometimes subtle patterns. This proves useful when deployed to follow the way employees and other legitimate network users navigate systems. Although even discussions regarding AI and machine learning feel redundant, to some degree, they are a powerful one-two punch in terms of automating security decisions.

Integrating AI calls for a comprehensive understanding of mathematics, logical reasoning, cognitive sciencesand a working knowledge of business networks. The professionals who implement AI for security purposes must also possess high-level expertise and protection planning skills. Used as a problem-solving tool, AI can provide real-time alerts and take pre-programmed actions. But it cannot effectively stem the tide of bad actors without support. Enter machine learning.

In this context, machine learning emphasizes software solutions driven by data analysis. Unlike human information processing limitations, machine learning can handle massive swaths of data. What machine learning learns, for lack of a better word, translates into actionable security intel for the overarching AI umbrella.

Some people think about machine learning as a subcategory of AI, which it is. Others comprehend it in a functional way,i.e., two sides to the same coin. But for cybersecurity experts determined to deter, detectand repel threat actors, machine learning is the gasoline that powers AI engines.

Its now essential to leverage machine learning capabilities to develop a so-called intelligent computer that can defend itself, to some degree. Although the relationship between AI and machine learning is diverse and complex, an expert can integrate them into a cybersecurity posture with relative ease. Its simply a matter of repetition and the following steps.

When properly orchestrated and refined to detect user patterns and subtle anomalies, the AI-machine learning relationship helps cybersecurity professionals keep valuable and sensitive digital assets away from prying eyes and greedy digital hands.

First and foremost, its crucial to put AI and machine learning benefits in context. Studies consistently conclude that more than 80% of all cybersecurity failures are caused by human error. Using automated technologies removes many mistake-prone employees and other network users from the equation. Along with minimizing risk, these are benefits of onboarding these automated next-generation technologies.

Improved cybersecurity efficiency. According to the 2023 Global Security Operations Center Study, cybersecurity professionals spend one-third of their workday chasing downfalse positives. This waste of time negatively impacts their ability to respond to legitimate threats, leaving a business at higher than necessary risk. The strategic application of AI and machine learning can be deployed to recognize harmless anomalies and alert a CISO or vCISO only when authentic threats are present.

Increased threat hunting capabilities.Without proactive, automated security measures like MDR (managed detection and response), organizations are too often following an outdated break-and-fix model. Hackers breach systems or deposit malware, and then the IT department spends the remainder of their day, or week, trying to purge the threat and repair the damage. Cybersecurity experts have widely adopted the philosophy that the best defense is a good offense. A thoughtful AI-machine learning strategy can engage in threat hunting without ever needing a coffee break.

Cure business network vulnerabilities.Vulnerability management approaches generally employ technologies that provide proactive automation. They close cybersecurity gaps and cure inherent vulnerabilities by identifying these weaknesses and alerting human decision-makers. Unlike scheduling a routine annual risk assessment, these cutting-edge technologies deliver ongoing analytics and constant vigilance.

Resolve cybersecurity skills gap.Its something of an open secret that there are not enough trained, certified cybersecurity experts to fill corporate positions. Thats one of the reasons why industry leaders tend to outsource managed IT and cybersecurity to third-party firms. Outsourcing helps to onboard the high-level knowledge and skills required to protect valuable digital assets and sensitive information. Without enough cybersecurity experts to safeguard businesses, automation allows the resources available to companies to drill down and identify true threats. Without these advanced technologies being used to bolster network security, its likely the number of debilitating cyberattacks would grow exponentially.

The type of predictive analytics and swift decision-making capabilities this two-prong approach delivers has seemingly endless industry applications. Banking and financial sector organizations can not only use AI and machine learning to repel hackers but also ferret out fraud. Healthcare organizations have a unique opportunity to exceed Health Insurance Portability and Accountability Act (HIPAA) requirements due to the advanced personal identity record protections it affords. Companies conducting business in the global marketplace can also get a leg-up in meeting the EUs General Data Protection Regulation (GDPR) designed to further informational privacy.

Perhaps the greatest benefit organizations garner from AI and machine learning security automation is the ability to detect, respondand expel threat actors and malicious applications. Managed IT cybersecurity experts can help companies close the skills gap by integrating these and other advanced security strategies.

John Funk is a Creative Consultant at SevenAtoms. A lifelong writer and storyteller, he has a passion for tech and cybersecurity. When hes not found enjoying craft beer or playing Dungeons & Dragons, John can be often found spending time with his cats

Check out our free e-newsletters to read more great articles..

Visit link:
How Machine Learning Revolutionizes Automation Security with AI-Powered Defense - Automation.com

Fastweb has announced it will leverage Amazon Web Services (AWS) generative AI and machine learning services to make its LLM, natively trained in Italian, available to third parties. This builds on Fastwebs work with AWS to help accelerate the digital transformation of Italian businesses and public sector organizations.

Fastweb is constructing a comprehensive Italian language dataset by combining public sources and licensed data from publishers and media outlets. Using this data, Fastweb has fine-tuned the Mistral 7B model using Amazon SageMaker, achieving performance improvements of 20-50% on Italian language benchmarks.

The new models will be made available on Hugging Face, allowing customers to deploy them via Amazon SageMaker. In the future, Fastweb plans to run its model on Amazon Bedrock using Custom Model Import, so it can easily build and scale new generative AI solutions for its customers using a broad set of capabilities available on Amazon Bedrock.

Walter Renna, CEO, Fastweb

Current AI models primarily rely on English data, a nuanced understanding of Italian culture can be harnessed by training on carefully chosen high-quality Italian datasets. This strategic initiative will help propel digital transformation for Italian organizations using technologies at the forefront of innovation. By making these models and applications available not only at a national level but also at a global level through AWSs comprehensive portfolio of generative AI services, were able to more easily build and scale our own generative AI offering, bringing new innovations to market faster.

Fabio Cerone, General Manager, Telco Industry, EMEA, AWS

We are committed to democratizing access of generative AI technology and applications to customers all over the world. The availability of LLMs natively trained on more languages is a critical piece in realizing that mission. Fastwebs effort to create an Italian-language LLM and generative AI is an important step in making the transformative power of generative AI more accessible to Italian businesses and government agencies. Through this work, Fastweb will make it easier for Italian customers to use generative AI to help resolve business challenges more quickly, tackle operational hurdles, and accelerate growth via digital transformation.

Originally posted here:
Natively Trained Italian LLM by Fastweb to Leverage AWS GenAI and Machine Learning Capabilities - The Fast Mode

It used to be that if you wanted to track down a rare frog, youd have to go to a likely place and wait until you heard its call. The rarer the frog, the less likely it was youd hear one.

Now, there are better tools for that.

The technologies that we work with are designed mostly to give you a better chance of detecting things that are hard to detect, said Justin Kitzes, an assistant professor of biological sciences in the Kenneth P. Dietrich School of Arts and Sciences.

Kitzes makes use of tools for bioacoustics the study of sounds made by animals which, along with satellite imaging and DNA methods, is part of a new generation of conservation technologies that allow researchers to search more broadly and efficiently than ever before.

At the beginning of a project, researchers in his lab place up to hundreds of sensors that listen in on an area of interest. Then researchers bring those recordings into the lab, where they sort the signal from the noise. And theres plenty of noise.

Each recorder can track about 150 hours of sound, and when the team deploys 50 sensors, as they did recently when searching for frogs in Panama, those hours add up.

7,500 is pretty small for us, because 50 recorders is actually a small deployment, Kitzes said. In our bird work, its more like 75,000 hours.

Theres no use in collecting eight continuous years of audio if you dont have time to listen to it, though. The labs research owes thanks to two technologies made available in 2017: an inexpensive audio recorder that allows the team to deploy hundreds of sensors and an open-source platform that gave scientists the ability to develop machine learning tools to sort through the data.

That was really what kicked everything off, said Kitzes. Because that gave us an explosion of field data along with the ability to train deep learning models to analyze it.

Tracking birds using this technology is one main focus for the team.

Another is its amphibian research, a collaboration with the lab of Biological Sciences Professor Corinne Richards-Zawacki as part of the RIBBITR program. That work, including biological sciences graduate student Sam Lapp and staff researcher Alexandra Syunkova, has the team focusing on sites in Pennsylvania, California, Panama and Brazil.

In one recent instance, audio recordings helped the researchers track down an elusive variable harlequin toad (pictured above) in an unlikely site in Panama that was only just beginning to recover from an outbreak of the deadly chytrid fungus. And just this year, the team published a study led by Lapp where they listened in on the underwater behavior of the endangered Sierra Nevada yellow-legged frog.

Studies like the latter rely on training what's called convolutional neural network models related to the ones used by tech companies use to recognize features in photos to categorize different types of sounds when presented with a visual representation of the audio recordings.

Were using the same kinds of models as Google and Amazon, where in your vacation photo albums they might be able to recognize a palm tree by a beach, Kitzes said.

But as high-tech as the work is, theres no replacement for the eye of a trained human. Members of the lab always check some of the algorithms work to ensure that its looking for the right calls. Its similar, Kitzes explains, to how he sees other uses of machine learning and artificial intelligence: Not as a replacement for the work of humans, but as a way to augment it.

"The reason our lab exists is that were trying to make conservation biologists and ecologists more effective at their job, said Kitzes. So they can get out there, find more species, learn better about whats impacting those species and, ultimately, take the actions that are necessary to conserve those species and protect biodiversity.

Patrick Monahan, photography by Corinne Richards-Zawacki

Read more from the original source:
To find rare frogs and birds, Pitt researchers are using machine learning algorithms and hundreds of microphones - University of Pittsburgh

A strength of using machine learning (ML) in oncology is its potential to extract data from unstructured documents, explained Will Shapiro, vice president of Data Science at Flatiron Health, during a session at the Association of Cancer Care Centers (ACCC) Annual Meeting & Cancer Center Business Summit (AMCCBS) in Washington DC. According to Shapiro, the ML team at Flatiron Health is focused on this endeavor in relation to oncology data and literature.

There's a ton of really rich information that's only in unstructured documents, Shapiro said during the session. We build models to extract things like metastatic status or diagnosis state, which are often not captured in any kind of regular structured way.

Image credit: ipopba | stock.adobe.com

Shapiro explained further that more recently, his ML team has started working with large language models (LLMs). He noted this space has significant potential within health care.

[At Flatiron Health] we built out a tool at the point of care that matches practice-authored regimens to NCCN guidelines, Shapiro said. That's something that we're really excited about.

Notably, Shapiro explained that his background is in fact not in health care, as he worked for many years at Spotify, where he built personalized recommendation engines using artificial intelligence (AI) and ML.

I really got excited about machine learning and AI in the context of building personalized recommendation engines [at Spotify], Shapiro explained. While personalizing music for a place like Spotify is radically different from personalizing medicine, I think there's actually some core things that really connect them, and I believe strongly that ML and AI have a key role to play in making truly personalized medicine a reality.

Shapiro noted that terminology can pose challenges for professionals in health care as they begin to dive into terms that contain a wealth of knowledge based on decades of research and thousands of dissertations. Terms such as LLM, natural language processing (NLP), generative AI, AI, and ML each represent an abundance of information that have helped us understand their potential today. Specifically, Shapiro noted that this collection of terms is distinct from workflow automation, which is another term in the same field that is often grouped together. Shapiro noted that workflow automation is distinct from these other terms in that currently there are well-known ways in which we evaluate quality for workflow automation.

With something like generative AIwhich is, I think, one of the most hyped things out in the world right nowit's so new that there really aren't ways that we can think about quality, Shapiro said. That's why I think it's really important to get educated and understand what's going on [around these terms].

According to Shapiro, a lot of these terms get used interchangeably, which can lead to additional confusion.

I think that there's a good reason for that, which is that there's a lot of overlap, Shapiro said. The same algorithm can be a deep learning algorithm and an NLP algorithm, and a lot of the applications are also the same.

Shapiro noted that one way of structuring these terms is to think of AI as a very broad category that encompasses ML, deep learning, and generative AI as nested subcategories. NLP, however, contains some differences.

There is an enormous amount of overlap between NLP and AI. A lot of the major advances in ML and AI stemmed from questions from NLP. But then there are also parts of NLP that are really distinct. [For example,] rules-based methods of parsing text are not something that I will think about with AI, and I will caveat this by saying that this is contentious, Shapiro said. If you google this, there will be 20 different ways that people try to structure this. My guidance is to not get too bogged down in the labels, but really try to focus on what the algorithm is or the product is that you're trying to understand.

According to Shapiro, one reason that oncologists should care about these terms is that ChatGPT, the most famous LLM currently in use today, is used by 1 in 10 doctors in their practice, according to a survey conducted over the summer of 2023. Shapiro noted that by the time of the presentation at the ACCC AMCCBS meeting in February 2024, that number has likely increased.

LLMs, which are large language models, are also a type of language model. According to Shapiro, the technical definition of a language model is a probability distribution over a sequence of words.

So, basically, given a chunk of text, what is the probability that any word will follow the chunk that you're looking at, Shapiro said. LLMs are essentially language models that are trained on the internet, so they're enormous.

According to Shapiro, language models can also be used to generate text. For instance, in the example My best friend and I are so close, we finish each other's ___ it is not difficult for humans to finish this with the appropriate word in the blank, which in this case would be sentences. Shapiro explained that is very much how language models work.

Probabilistically, sentence is the missing word [in that example], which is very much at the core of what's happening with a language model, Shapiro said. In fact, autocomplete, which you probably don't even think about as you see it every day, is generative AI that's an example [of a language model], and it's one of the motivating examples of generative AI.

To be clear in terms of definition, Shapiro noted that generative AI are AI models that generate new content. Specifically, the GPT in ChatGPT (which is both an LLM and generative AI) stands for generative pre-trained transformer. According to Shapiro, pre-trained models can be understood as having a foundational knowledge, which is in contrast to other kinds of models that just do one task.

I mentioned my team works on building models that will extract metastatic status from documents, and that's all they do, Shapiro said. In contrast, pre-trained models can do a lot of different kinds of things. They can classify the sentiment of reviews, they can flag abusive messages, and they probably are going to write the next 10 Harry Potter novels. They can extract adverse events from charts, and they can also do things that extract metastatic status. So, that's a big part of the appealone model can do a lot of different things.

However, this capacity of one model being capable of doing many different things can also have a trade off in terms of quality. Shapiro explained that that is something his team at Flatiron Health has found to be true in their work.

What we've found at Flatiron Health is that generally, purpose-built models can be much better at actually predicting or doing one task. But one thing that's become really exciting, and kind of gets into the weeds of LLMs, is this concept of taking a pre-trained model and fine-tuning it on labeled examples, which is a way to really increase the performance of a pre-trained model.

Further, the T in ChatGPT stands for transformer, which is a type of deep learning architecture that was developed at Google in 2017, explained Shapiro. It was originally described in a paper called Attention is All You Need.

Transformers are actually kind of simple, Shapiro said. If you read about the history of deep learning, model architectures tended to get more and more complex, and the transformer actually stripped away a fair amount of this complexity. But what's been really game changing is how big they are, as they're trained on the internet. So things like Wikipedia, Redditthese huge corpuses of texthave billions of grammars, and they're really, really expensive to train.

Yet, the size of them is what has led to these incredible breakthroughs in performance and benchmarks that have caused quite a bit of buzz recently, explained Shapiro. With this buzz and attention raises the importance of becoming more educated in what these models are and how they work, especially in areas such as health care.

With 10% of doctors using ChatGPT, it is something that everyone really needs to get educated about pretty quickly. I also just think there are so many exciting ways that ML and AI have a role to play in the future of oncology, Shapiro said.

Shapiro explained further that using these models, there is the potential in oncology to conduct research that is pulled from enormous patient populations, which can made available at scale. Additionally, there is the potential to summarize visit notes from audio recordings, to predict patient response to a treatment, and to discover new drug targets.

There are huge opportunities in ML and AI, but there are also a lot of challenges and a lot of open questions. When you see someone like Sam Altman, the CEO of OpenAI, going to Congress and asking it to be regulated, you know that there's something to pay attention to, Shapiro said. That's because there's some real problems.

Such problems include hallucinations, which consists of models inventing answers. Shapiro explained what makes hallucinations by AI models even more pernicious is that they come from a place of technological authority.

There's an inherent inclination to trust them, Shapiro said. There's a lot of traditional considerations for any type of ML or AI algorithm around whether they are biased, whether they are perpetuating inequity, and whether data shifts affect their quality. For this reason, I think it's more important than ever to really think closely about how you're validating the quality of models. High quality ground truth data, I think, is essential for using any of these types of ML or AI algorithms.

Reference

Shapiro W. Deep Dive 6. Artificial and Business Intelligence Technology. Presented at: ACCC AMCCBS; February 28-March 1, 2024; Washington, DC.

Read the original here:
Unlocking the Potential of Machine Learning and Large Language Models in Oncology - Pharmacy Times

Jadon A, Kumar S. Leveraging Generative AI Models for Synthetic Data Generation in Healthcare: Balancing Research and Privacy. arXiv. 2023;2305.05247

Umer F, Khan M. A call to action: concerns related to artificial intelligence. Oral Surg Oral Med Oral Pathol Oral Radiol. 2021;132:255.

Article PubMed Google Scholar

Adnan N, Umer F. Understanding deep learningchallenges and prospects. J Pak Med Assoc. 2022;72:S5963.

Google Scholar

Umer F. Could AI offer practical solutions for dentistry in the future? BDJ Team. 2022;9:268.

Article Google Scholar

Bttner M, Schwendicke F. Natural language processing in dentistry. Br Dent J. 2023;234:753.

Article PubMed Google Scholar

Adnan N, Khalid WB, Umer F. An artificial intelligence model for teeth segmentation and numbering on orthopantomograms. Int J Comput Dent. 2023;0:0.

Google Scholar

DAmico S, DallOlio D, Sala C, DallOlio L, Sauta E, Zampini M, et al. Synthetic data generation by artificial intelligence to accelerate research and precision medicine in hematology. JCO Clin Cancer Inf. 2023;7:e2300021.

Article Google Scholar

Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019;366:44753.

Article CAS PubMed Google Scholar

Sengupta E, Garg D, Choudhury T, Aggarwal A, editors. Techniques to eliminate human bias in machine learning. In: Proceedings of international conference on System Modeling & Advancement in Research Trends (SMART). (IEEE, 2018).

Futoma J, Simons M, Panch T, Doshi-Velez F, Celi LA. The myth of generalisability in clinical research and machine learning in health care. Lancet Digit Health. 2020;2:e48992.

Article PubMed PubMed Central Google Scholar

Adnan N, Umer F, Malik S. Implementation of transfer learning for the segmentation of human mesenchymal stem cellsa validation study. Tissue Cell. 2023;83:102149.

Article PubMed Google Scholar

Goodfellow I, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, Ozair S, et al. Generative adversarial nets. Adv. Neural Inf. Process. Syst. 2014;27:267280.

Adams LC, Busch F, Truhn D, Makowski MR, Aerts H, Bressem KK. What does DALL-E 2 know about radiology? J Med Internet Res. 2023;25:e43110.

Article PubMed PubMed Central Google Scholar

Kenig N, Monton Echeverria J, Muntaner Vives A. Human Beauty according to Artificial Intelligence. Plast Reconstr Surg Glob Open. 2023;11:e5153.

Article PubMed PubMed Central Google Scholar

Gonzales A, Guruswamy G, Smith SR. Synthetic data in health care: a narrative review. PLoS Digit Health. 2023;2:e0000082.

Article PubMed PubMed Central Google Scholar

Seyyed-Kalantari L, Liu G, McDermott M, Chen IY & Ghassemi M. CheXclusion: fairness gaps in deep chest X-ray classifiers. In BIOCOMPUTING 2021: Proc. Pacific Symposium 232243; World Scientific. 2020.

Schwendicke F, Samek W, Krois J. Artificial intelligence in dentistry: chances and challenges. J Dent Res. 2020;99:76974.

Article CAS PubMed Google Scholar

Celi LA, Cellini J, Charpignon M-L, Dee EC, Dernoncourt F, Eber R, et al. Sources of bias in artificial intelligence that perpetuate healthcare disparitiesa global review. PLoS Digit Health. 2022;1:e0000022.

Article PubMed PubMed Central Google Scholar

Abdel-Jaber H, Devassy D, Al Salam A, Hidaytallah L, El-Amir M. A review of deep learning algorithms and their applications in healthcare. Algorithms. 2022;15:71.

Article Google Scholar

Nikolentzos G, Vazirgiannis M, Xypolopoulos C, Lingman M, Brandt EG. Synthetic electronic health records generated with variational graph autoencoders. NPJ Digit Med. 2023;6:83.

Article PubMed PubMed Central Google Scholar

Joo S, Jung W, Oh SE. Variational autoencoder-based estimation of chronological age and changes in morphological features of teeth. Sci Rep. 2023;13:704.

Article CAS PubMed PubMed Central Google Scholar

Kingma DP, Welling M. An introduction to variational autoencoders. Found. Trends Mach Learn. 2019;12:30792.

Article Google Scholar

Yoon J, Drumright LN, Van Der Schaar M. Anonymization through data synthesis using generative adversarial networks (ads-gan). IEEE J Biomed Health Inf. 2020;24:237888.

Article Google Scholar

Goodfellow I, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, Ozair S, et al. Generative adversarial networks. Commun ACM. 2020;63:13944.

Article Google Scholar

Huang Y, Fan F, Syben C, Roser P, Mill L, Maier A. Cephalogram synthesis and landmark detection in dental cone-beam CT systems. Med Image Anal. 2021;70:102028.

Article PubMed Google Scholar

Kokomoto K, Okawa R, Nakano K, Nozaki K. Intraoral image generation by progressive growing of generative adversarial network and evaluation of generated image quality by dentists. Sci Rep. 2021;11:18517.

Article CAS PubMed PubMed Central Google Scholar

He H, Zhao S, Xi Y, Ho JC. MedDiff: Generating Electronic Health Records using Accelerated Denoising Diffusion Model. arXiv preprint arXiv:2302.04355. 2023.

Kazerouni A, Aghdam EK, Heidari M, Azad R, Fayyaz M, Hacihaliloglu I, et al. Diffusion models in medical imaging: a comprehensive survey. Med Image Anal. 2023;88:102846.

Article PubMed Google Scholar

Frid-Adar M, Diamant I, Klang E, Amitai M, Goldberger J, Greenspan H. GAN-based synthetic medical image augmentation for increased CNN performance in liver lesion classification. Neurocomputing. 2018,321:32131.

Bian Y, Xie XQ. Generative chemistry: drug discovery with deep learning generative models. J Mol Model. 2021;27:71.

Article CAS PubMed Google Scholar

Elreedy D, Atiya AF. A comprehensive analysis of synthetic minority oversampling technique (SMOTE) for handling class imbalance. Information Sciences. 2019;505:3264.

Liu X, Glocker B, McCradden MM, Ghassemi M, Denniston AK, Oakden-Rayner L. The medical algorithmic audit. Lancet Digit Health. 2022;4:e38497.

Article CAS PubMed Google Scholar

Arora A. Artificial intelligence: a new frontier for anaesthesiology training. Br J Anaesth. 2020;125:e4078.

Article PubMed Google Scholar

Chen RJ, Lu MY, Chen TY, Williamson DFK, Mahmood F. Synthetic data in machine learning for medicine and healthcare. Nat Biomed Eng. 2021;5:4937.

Article PubMed PubMed Central Google Scholar

Budd S, Robinson EC, Kainz B. A survey on active learning and human-in-the-loop deep learning for medical image analysis. Med Image Anal. 2021;71:102062.

Article PubMed Google Scholar

Abrmoff MD, Tarver ME, Loyo-Berrios N, Trujillo S, Char D, Obermeyer Z, et al. Considerations for addressing bias in artificial intelligence for health equity. NPJ Digit Med. 2023;6:170.

Article PubMed PubMed Central Google Scholar

Arora A, Arora A. Generative adversarial networks and synthetic patient data: current challenges and future perspectives. Future Healthc J. 2022;9:190.

Article PubMed PubMed Central Google Scholar

Liu M, Ning Y, Teixayavong S, Mertens M, Xu J, Ting DSW, et al. A translational perspective towards clinical AI fairness. NPJ Digit Med. 2023;6:172.

Article PubMed PubMed Central Google Scholar

Akhtar Z. Deepfakes generation and detection: a short survey. J Imaging. 2023;9:18.

Article PubMed PubMed Central Google Scholar

Taeihagh A. Governance of artificial intelligence. Policy Soc. 2021;40:13757.

Article Google Scholar

View original post here:
Generative artificial intelligence: synthetic datasets in dentistry | BDJ Open - Nature.com

Photo by Vince Fleming on Unsplash

By now, Im sure most of you have heard the news about Googles new LLM*, Gemini, generating pictures of racially diverse people in Nazi uniforms. This little news blip reminded me of something that Ive been meaning to discuss, which is when models have blind spots, so we apply expert rules to the predictions they generate to avoid returning something wildly outlandish to the user.

This sort of thing is not that uncommon in machine learning, in my experience, especially when you have flawed or limited training data. A good example of this that I remember from my own work was predicting when a package was going to be delivered to a business office. Mathematically, our model would be very good at estimating exactly when the package would get physically near the office, but sometimes, truck drivers arrive at destinations late at night and then rest in their truck or in a hotel until morning. Why? Because no ones in the office to receive/sign for the package outside of business hours.

Teaching a model about the idea of business hours can be very difficult, and the much easier solution was just to say, If the model says the delivery will arrive outside business hours, add enough time to the prediction that it changes to the next hour the office is listed as open. Simple! It solves the problem and it reflects the actual circumstances on the ground. Were just giving the model a little boost to help its results work better.

However, this does cause some issues. For one thing, now we have two different model predictions to manage. We cant just throw away the original model prediction, because thats what we use for model performance monitoring and metrics. You cant assess a model on predictions after humans got their paws in there, thats not mathematically sound. But to get a clear sense of the real world model impact, you do want to look at the post-rule prediction, because thats what the customer actually experienced/saw in your application. In ML, were used to a very simple framing, where every time you run a model you get one result or set of results, and thats that, but when you start tweaking the results before you let them go, then you need to think at a different scale.

I kind of suspect that this is a form of whats going on with LLMs like Gemini. However, instead of a post-prediction rule, it appears that the smart money says Gemini and other models are applying secret prompt augmentations to try and change the results the LLMs produce.

In essence, without this nudging, the model will produce results that are reflective of the content it has been trained on. That is to say, the content produced by real people. Our social media posts, our history books, our museum paintings, our popular songs, our Hollywood movies, etc. The model takes in all that stuff, and it learns the underlying patterns in it, whether they are things were proud of or not. A model given all the media available in our contemporary society is going to get a whole lot of exposure to racism, sexism, and myriad other forms of discrimination and inequality, to say nothing of violence, war, and other horrors. While the model is learning what people look like, and how they sound, and what they say, and how they move, its learning the warts-and-all version.

Our social media posts, our history books, our museum paintings, our popular songs, our Hollywood movies, etc. The model takes in all that stuff, and it learns the underlying patterns in it, whether they are things were proud of or not.

This means that if you ask the underlying model to show you a doctor, its going to probably be a white guy in a lab coat. This isnt just random, its because in our modern society white men have disproportionate access to high status professions like being doctors, because they on average have access to more and better education, financial resources, mentorship, social privilege, and so on. The model is reflecting back at us an image that may make us uncomfortable because we dont like to think about that reality.

The obvious argument is, Well, we dont want the model to reinforce the biases our society already has, we want it to improve representation of underrepresented populations. I sympathize with this argument, quite a lot, and I care about representation in our media. However, theres a problem.

Its very unlikely that applying these tweaks is going to be a sustainable solution. Recall back to the story I started with about Gemini. Its like playing whac-a-mole, because the work never stops now weve got people of color being shown in Nazi uniforms, and this is understandably deeply offensive to lots of folks. So, maybe where we started by randomly applying as a black person or as an indigenous person to our prompts, we have to add something more to make it exclude cases where its inappropriate but how do you phrase that, in a way an LLM can understand? We probably have to go back to the beginning, and think about how the original fix works, and revisit the whole approach. In the best case, applying a tweak like this fixes one narrow issue with outputs, while potentially creating more.

Lets play out another very real example. What if we add to the prompt, Never use explicit or profane language in your replies, including [list of bad words here]. Maybe that works for a lot of cases, and the model will refuse to say bad words that a 13 year old boy is requesting to be funny. But sooner or later, this has unexpected additional side effects. What about if someones looking for the history of Sussex, England? Alternately, someones going to come up with a bad word you left out of the list, so thats going to be constant work to maintain. What about bad words in other languages? Who judges what goes on the list? I have a headache just thinking about it.

This is just two examples, and Im sure you can think of more such scenarios. Its like putting band aid patches on a leaky pipe, and every time you patch one spot another leak springs up.

So, what is it we actually want from LLMs? Do we want them to generate a highly realistic mirror image of what human beings are actually like and how our human society actually looks from the perspective of our media? Or do we want a sanitized version that cleans up the edges?

Honestly, I think we probably need something in the middle, and we have to continue to renegotiate the boundaries, even though its hard. We dont want LLMs to reflect the real horrors and sewers of violence, hate, and more that human society contains, that is a part of our world that should not be amplified even slightly. Zero content moderation is not the answer. Fortunately, this motivation aligns with the desires of large corporate entities running these models to be popular with the public and make lots of money.

we have to continue to renegotiate the boundaries, even though its hard. We dont want LLMs to reflect the real horrors and sewers of violence, hate, and more that human society contains, that is a part of our world that should not be amplified even slightly. Zero content moderation is not the answer.

However, I do want to continue to make a gentle case for the fact that we can also learn something from this dilemma in the world of LLMs. Instead of simply being offended and blaming the technology when a model generates a bunch of pictures of a white male doctor, we should pause to understand why thats what we received from the model. And then we should debate thoughtfully about whether the response from the model should be allowed, and make a decision that is founded in our values and principles, and try to carry it out to the best of our ability.

As Ive said before, an LLM isnt an alien from another universe, its us. Its trained on the things we wrote/said/filmed/recorded/did. If we want our model to show us doctors of various sexes, genders, races, etc, we need to make a society that enables all those different kinds of people to have access to that profession and the education it requires. If were worrying about how the model mirrors us, but not taking to heart the fact that its us that needs to be better, not just the model, then were missing the point.

If we want our model to show us doctors of various sexes, genders, races, etc, we need to make a society that enables all those different kinds of people to have access to that profession and the education it requires.

See more here:
Seeing Our Reflection in LLMs. When LLMs give us outputs that reveal | by Stephanie Kirmer | Mar, 2024 - Towards Data Science

Deep distilling is an automated method that learns relationships in data using essence neural networks. It then condenses the neural representation of these relationships into human-understandable rules, usually in the form of executable computer code that is much more concise than the neural network. (Illustration credit: Anda Kim)

DALLAS Feb. 29, 2024 UTSouthwestern Medical Center researchers have developed an artificial intelligence (AI) method that writes its own algorithms and may one day operate as an automated scientist to extract the meaning behind complex datasets.

Milo Lin, Ph.D., is Assistant Professor in the Lyda Hill Department of Bioinformatics, Biophysics, and the Center for Alzheimer's and Neurodegenerative Diseases at UTSouthwestern.

Researchers are increasingly employing AI and machine learning models in their work, but with the huge caveat that these high-performing models provide limited new direct insights into the data, saidMilo Lin, Ph.D., Assistant Professor intheLyda Hill Department of Bioinformatics,Biophysics,and theCenter for Alzheimers and Neurodegenerative Diseasesat UTSouthwestern.Our work is the first step in allowing researchers to use AI to directly convert complex data into new human-understandable insights.

Dr. Lin co-led the study, published inNature Computational Science,with first author Paul J. Blazek, M.D., Ph.D.,who worked on this project as part of his thesis work while he was at UTSW.

In the past several years, the field of AI has seen enormous growth, with significant crossover from basic and applied scientific discovery to popular use. One commonly used branch of AI, known as neural networks, emulates the structure of the human brain by mimicking the way biological neurons signal one another. Neural networks are a form of machine learning, which creates outputs based on input data after learning on a training dataset.

Although this tool has found significant use in applications such as image and speech recognition, conventional neural networks have significant drawbacks, Dr. Lin said. Most notably, they often dont generalizefarbeyond the data they train on, and the rationale for their output is a black box, meaning theres no way for researchers to understand how a neural network algorithm reached its conclusion.This study was supported by UTSWs High Impact Grant Program, which was initiated in 2001 and supports high-risk research offering high potential impact in basic science or medicine.

Seeking to address both issues, the UTSW researchers developed a method they call deep distilling. Using limited training data datasets used to train machine learning models deep distilling automatically discovers algorithms, or the rules to explain observed input-output patterns in the data. This is done by training an essence neural network (ENN), previously developed in the Lin Lab, on input-output data. The parameters of the ENN that encode the learned algorithm are then translated into succinct computer codes so users can read them.

The researchers tested deep distilling in a variety of scenarios in which traditional neural networks cannot produce human-comprehensible rules and have poor performance in generalizing to very different data. These included cellular automata, in which grids contain hypothetical cells in distinct states that evolve over time according to a set of rules often used as model systems for emergent behavior in the physical, life, and computer sciences. Although the grid used by the researchers had 256 possible sets of rules, deep distilling was able to learn the rules for accurately predicting the hypothetical cells behavior for every set of rules after seeing only grids from 16 rule sets, summarizing all 256 rule sets in a single algorithm.

In another test, the researchers trained deep distilling to accurately classify a shapes orientation as vertical or horizontal. Although only a few training images of perfectly horizontal or vertical lines were required, this method was able to apply the succinct algorithm it discovered to accurately solve much more ambiguous cases, such as patterns with multiple lines or gradients and shapes made of boxes as well as zigzag, diagonal, or dotted lines.

Eventually, Dr. Lin said, deep distilling could be applied to the vast datasets generated by high-throughput scientific studies, such as those used for drug discovery, and act as an automated scientist capturing patterns in results not easily discernible to the human brain, such as how DNA sequences encode functional rules of biomolecular interactions. Deep distilling also could potentially serve as a decision-making aid to doctors, offering insights on its thought process through the generated algorithms, he added.

This study was supported by UTSWs High Impact Grant Program, which was initiated in 2001 and supports high-risk research offering high potential impact in basic science or medicine.

About UTSouthwestern Medical Center

UTSouthwestern, one of the nations premier academic medical centers, integrates pioneering biomedical research with exceptional clinical care and education. The institutions faculty members have received six Nobel Prizes and include 25 members of the National Academy of Sciences, 21 members of the National Academy of Medicine, and 13 Howard Hughes Medical Institute Investigators. The full-time faculty of more than 3,100 is responsible for groundbreaking medical advances and is committed to translating science-driven research quickly to new clinical treatments. UTSouthwestern physicians provide care in more than 80 specialties to more than 120,000 hospitalized patients, more than 360,000 emergency room cases, and oversee nearly 5 million outpatient visits a year.

Original post:
UTSW team's new AI method may lead to 'automated scientists' - UT Southwestern

Groq Inc., a well-funded maker of artificial intelligence inference chips, has acquired fellow startup Definitive Intelligence Inc. for an undisclosed sum.

The companies announced the transaction today. The deal will help Groq enhance the capabilities of its newest offering, a cloud platform called GroqCloud that provides on-demand access to its AI chips.

Groq was founded in 2016 by Chief Executive Officer Jonathan Ross, a former Google LLC engineer who invented the search giants TPU machine learning processors. The company is backed by more than $360 million in funding. It raised the bulk of that capital through a Series C round co-led by Tiger Global Management and D1 Capital in early 2021.

Groqs flagship product is an AI chip known as the LPU Inference Engine. Its optimized to power large language models with a focus on inference, or the task of running an AI in production after it has been trained. In a November benchmark test, Groqs LPU set an inference speed record while running Meta Platform Inc.s popular Llama 2 70B LLM.

The LPU consists of cores dubbed TSPs that each include about 230 megabytes of memory. According to Groq, the TSPs are linked together by an on-chip network that provides detailed information on how much time it takes data to travel between the different cores. This information helps speed up LLM response times.

The faster a piece of data reaches a chip core via the onboard network, the sooner processing can begin. The information that the LPU provides about its onboard network allows LLMs to identify the fastest data travel routes and use them to speed up computations. Groq claims its chip can perform inference up to 10 times faster than competing products.

Definitive Intelligence, the company has acquired, is a Palo Alto, California-based analytics provider that previously raised more than $10 million in funding. It offers an AI-powered application that enables users to query datasets with natural language instructions. The software also lends itself to related tasks such as creating data visualizations.

Groq detailed on occasion of the acquisition that Definitive Intelligence had helped it build GroqCloud, a recently launched platform through which it provides on-demand access to LPUs. Developers can use the platform to familiarize themselves with the companys chips and build applications optimized for their architecture. A built-in library of learning resources promises to ease the onboarding process.

Following the acquisition, Definitive Intelligence co-founder and CEO Sunny Madra will join Groq to lead the business unit in charge of GroqCloud. The units initial priorities include expanding the platforms capacity and growing its user base. Groq said that the acquisition will also support the launch of a second division, Groq Systems, that will focus on helping organizations such as government agencies deploy the companys LPUs.

THANK YOU

Continue reading here:
AI chip startup Groq acquires Definitive Intelligence to scale its cloud platform - SiliconANGLE News

Capitalizing on this technology is criticalbut its notoriously difficult to launch. Many ML projects never progress beyond the modeling: the number-crunching phase. Industry surveys repeatedly show that most new ML initiatives dont make it to deployment, where the value would be realized.

Hype contributes to this problem. ML is mythologized, misconstrued as intelligent when it is not. Its also mismeasured as highly accurate, even when that notion is irrelevant and misleading. For now, these adulations largely drown out the words of consternation, but those words are bound to increase in volume.

Take self-driving cars. In the most publicly visible cautionary tale about ML hype, overzealous promises have led to slamming on the brakes and slowing progress. AsThe Guardianput it, The driverless car revolution has stalled. This is a shame, as the concept promises greatness. Someday, it will prove to be a revolutionary application of ML that greatly reduces traffic fatalities. This will require a lengthy transformation that is going to happen over 30 years and possibly longer, according Chris Urmson, formerly the CTO of Googles self-driving team and now the CEO of Aurora, which bought out Ubers self-driving unit. But in the mid-2010s, the investment and fanatical hype, including grandiose tweets by Tesla CEO Elon Musk, reached a premature fever pitch. The advent of truly impressive driver assistance capabilities were branded as Full Self-Driving and advertised as being on the brink of widespread, completely autonomous drivingthat is, self-driving that allows you to nap in the back seat.

Expectations grew, followed by . . . a conspicuous absence of self-driving cars. Disenchantment took hold and by the early 2020s investments had dried up considerably. Self-driving is doomed to be this decades jetpack.

What went wrong? Underplanning is an understatement. It wasnt so much a matter of overselling ML itself, that is, of exaggerating how well predictive models can, for example, identify pedestrians and stop signs. Instead, the greater problem was the dramatic downplaying of deployment complexity. Only a comprehensive, deliberate plan could possibly manage the inevitable string of impediments that arise while slowly releasing such vehicles into the world. After all, were talking about ML models autonomously navigating large, heavy objects through the midst of our crowded cities! One tech journalist poignantly dubbed them self-driving bullets. When it comes to operationalizing ML, autonomous driving is literally where the rubber hits the road. More than any other ML initiative, it demands a shrewd, incremental deployment plan that doesnt promise unrealistic timelines.

The ML industry has nailed the development of potentially valuable models, but not their deployment. A report prepared by theAI Journalbased on surveys by Sapio Research showed that the top pain point for data teams is Delivering business impact now through AI. Ninety-six percent of those surveyed checked that box. That challenge beat out a long list of broader data issues outside the scope of AI per se, including data security, regulatory compliance, and various technical and infrastructure challenges. But when presented with a model, business leaders refuse to deploy. They just say no. The disappointed data scientist is left wondering, You cant . . . or you wont? Its a mixture of both, according to a question asked by my survey with KDnuggets (see responsesto the question, What is the main impediment to model deployment?). Technical hurdles mean that they cant. A lack of approvalincluding when decision makers dont consider model performance strong enough or when there are privacy or legal issuesmeans that theywont.

Another survey also told this some cant and some wont story. After ML consultancy Rexer Analytics survey of data scientists asked why models intended for deployment dont get there, founder Karl Rexer told me that respondents wrote in two main reasons: The organization lacks the proper infrastructure needed for deployment and People in the organization dont understand the value of ML.

Unsurprisingly, the latter group of data scientiststhe wonts rather than the cantssound the most frustrated, Karl says.

Whether they cant or they wont, the lack of a well-established business practice is almost always to blame. Technical challenges abound for deployment, but they dont stand in the way so long as project leaders anticipate and plan for them. With a plan that provides the time and resources needed to handle model implementationsometimes, major constructiondeployment will proceed. Ultimately, its not so much that they cant but that they wont.

About the Author

Eric Siegel, Ph.D., is a leading consultant and former Columbia University professor who helps companies deploy machine learning. He is the founder of the long-runningMachine Learning Weekconference series and its new sister,Generative AI World, the instructor of the acclaimed online course Machine Learning Leadership and Practice End-to-End Mastery, executive editor ofThe Machine Learning Times, and afrequent keynote speaker. He wrote the bestsellingPredictive Analytics: The Power to PredictWho Will Click, Buy, Lie, or Die, which has been used in courses at hundreds of universities, as well asThe AI Playbook: Mastering the Rare Art of Machine Learning Deployment. Erics interdisciplinary work bridges the stubborn technology/business gap. At Columbia, he won the Distinguished Faculty award when teaching the graduatecomputer sciencecourses in ML and AI. Later, he served as abusiness schoolprofessor at UVA Darden. Eric also publishesop-eds on analytics and social justice.

Eric hasappeared onBloomberg TV and Radio, BNN (Canada), Israel National Radio, National Geographic Breakthrough, NPR Marketplace, Radio National (Australia), and TheStreet. Eric and his books have beenfeatured inBig Think, Businessweek, CBS MoneyWatch, Contagious Magazine, The European Business Review, Fast Company, The Financial Times, Forbes, Fortune, GQ, Harvard Business Review, The Huffington Post, The Los Angeles Times, Luckbox Magazine, MIT Sloan Management Review, The New York Review of Books, The New York Times, Newsweek, Quartz, Salon, The San Francisco Chronicle, Scientific American, The Seattle Post-Intelligencer, Trailblazers with Walter Isaacson, The Wall Street Journal, The Washington Post,andWSJ MarketWatch.

See original here:
Effective Machine Learning Needs Leadership Not AI Hype - The Machine Learning Times

Theres some in-fighting going on in the artificial intelligence (AI) world, and one prominent billionaire claims the future of the human race is at stake. Elon Musk is taking legal action against Microsoft-backed OpenAI and its CEO, Sam Altman, alleging the company has strayed from its original mission to develop artificial intelligence for the collective benefit of humanity.

Musks attorneys filed a lawsuit on Thursday (Feb. 29) in San Francisco, asserting that in 2015, Altman and Greg Brockman, co-founders of OpenAI, approached Musk to assist in establishing a nonprofit focused on advancing artificial general intelligence for the betterment of humanity.

Although Musk helped initiate OpenAI in 2015, he departed from its board in 2018. Previously, in 2014, he had voiced concerns about the risks associated with AI, suggesting it could pose more significant dangers than nuclear weapons.

The lawsuit highlights that OpenAI, Inc. still claims on its website to prioritize ensuring that artificial general intelligence benefits all of humanity. However, the suit contends that in reality, OpenAI, Inc. has evolved into a closed-source entity effectively operating as a subsidiary of Microsoft, the worlds largest technology company.

When it comes to cybersecurity, AI brings both risks and rewards. Google CEO Sundar Pichai and other industry leaders say artificial intelligence is key to enhancing online security. AI can accelerate and streamline the management of cyber threats. It leverages vast datasets to identify patterns, automating early incident analysis and enabling security teams to quickly gain a comprehensive view of threats, thus hastening their response.

Lenovo CTO Timothy E. Bates told PYMNTS that AI-driven tools, such as machine learning for anomaly detection and AI platforms for threat intelligence, are pivotal. Deep learning technologies dissect malware to decipher its composition and potentially deconstruct attacks. These AI systems operate behind the scenes, learning from attacks to bolster defense and neutralize future threats.

With the global shift toward a connected economy, cybercrime is escalating, causing significant financial losses, including an estimated $10.3 billion in the U.S. alone in 2022, according to the FBI.

Get set for lots more books that are authored or co-authored by AI. Inkitt, a startup leveraging artificial intelligence (AI) to craft books, has secured $37 million. Inkitts app enables users to self-publish their narratives. By employing AI and data analytics, it selects stories for further development and markets them on its Galatea app.

This technological shift offers both opportunities and challenges.

Zachary Weiner, CEO of Emerging Insider Communications, which focuses on publishing, shared his insights on the impact of AI on writing with PYMNTS. Writers gain significantly from the vast new toolkit AI provides, enhancing their creative process with AI-generated prompts and streamlining tasks like proofreading. AI helps them overcome traditional brainstorming limits, allowing for the fusion of ideas into more intricate narratives. It simplifies refining their work, letting them concentrate on their primary tasks.

But he warns of the pitfalls AI introduces to the publishing world. AI is making its way into all aspects of writing and content creation, posing a threat to editorial roles, he said. The trend towards replacing human writers with AI for cost reduction and efficiency gains is not just a possibility but a current reality.

The robots are coming, and they are getting smarter. New advancements in artificial intelligence (AI) are making it possible for companies to create robots with better features and improved abilities to interact with humans.

Figure AI has raised $675 million to develop AI-powered humanoid robots. Investors include Jeff Bezos Explore Investments and tech giants like Microsoft, Amazon, Nvidia, OpenAI, and Intel. Experts say this investment shows a growing interest in robotics because of AI.

According to Sarah Sebo, an assistant professor of computer science at the University of Chicago, AI can help robots understand their surroundings better, recognize objects and people more accurately, communicate more naturally with humans and improve their abilities over time through feedback.

Last March, Figure AI introduced the Figure 01 robot, designed for various tasks, from industrial work to household chores. Equipped with AI, this robot mimics human movements and interactions.

The company hopes these robots will take on risky or repetitive tasks, allowing humans to focus on more creative work.

Read more from the original source:
This Week in AI: A Battle for Humanity or Profits? - PYMNTS.com

Google CEO Sundar Pichai recently noted that artificial intelligence (AI) could boost online security, a sentiment echoed by many industry experts.

AI is transforming how security teams handle cyber threats, making their work faster and more efficient. By analyzing vast amounts of data and identifying complex patterns, AI automates the initial stages of incident investigation. The new methods allow security professionals to begin their work with a clear understanding of the situation, speeding up response times.

Tools like machine learning-based anomaly detection systems can flag unusual behavior, while AI-driven security platforms offer comprehensive threat intelligence and predictive analytics, Timothy E. Bates, chief technology officer at Lenovo, told PYMNTS in an interview. Then theres deep learning, which can analyze malware to understand its structure and potentially reverse-engineer attacks. These AI operatives work in the shadows, continuously learning from each attack to not just defend but also to disarm future threats.

Cybercrime is a growing problem as more of the world embraces the connected economy. Losses from cyberattackstotaled at least $10.3 billion in the U.S. in 2022, per an FBI report.

The tools used by attackers and defenders are constantly changing and increasingly complex, Marcus Fowler, CEO of cybersecurity firm Darktrace Federal, said in an interview with PYMNTS.

AI represents the greatest advancement in truly augmenting the current cyber workforce, expanding situational awareness, and accelerating mean time to action to allow them to be more efficient, reduce fatigue, and prioritize cyber investigation workloads, he said.

As cyberattacks continue to rise, improving defense tools is becoming increasingly important. Britains GCHQ intelligence agency recently warned that new AI tools could lead to more cyberattacks, making it easier for beginner hackers to cause harm. The agency also said that the latest technology could increase ransomware attacks, where criminals lock files and ask for money, according to a reportby GCHQs National Cyber Security Centre.

Googles Pichai pointed out that AI is helping to speed up how quickly security teams can spot and stop attacks. This innovation helps defenders who have to catch every attack to keep systems safe, while attackers only need to succeed once to cause trouble.

While AI may enhance the capabilities of cyberattackers, it equally empowers defenders against security breaches.

Artificial intelligence has the potential to benefit the field of cybersecurity far beyond just automating routine tasks, Piyush Pandey, CEO of cybersecurity firm Pathlock, noted in an interview with PYMNTS. As rules and security needs keep growing, he said, the amount of data for governance, risk management and compliance (GRC) is increasing so much that it may soon become too much to handle.

Continuous, automated monitoring of compliance posture using AI can and will drastically reduce manual efforts and errors, he said. More granular, sophisticated risk assessments will be available via ML [machine learning] algorithms, which can process vast amounts of data to identify subtle risk patterns, offering a more predictive approach to reducing risk and financial losses.

Using AI to spot specific patterns is one way to catch hackers who keep getting better at what they do. Todays hackers are good at avoiding usual security checks, so many groups are using AI to catch them, Mike Britton, CISO at Abnormal Security, told PYMNTS in an interview. He said that one way that AI can be used in cyber defense is through behavioral analytics. Instead of just searching for known bad signs like dangerous links or suspicious senders, AI-based solutions can spot unusual activity that doesnt fit the normal pattern.

By baselining normal behavior across the email environment including typical user-specific communication patterns, styles, and relationships AI could detect anomalous behavior that may indicate an attack, regardless of whether the content was authored by a human or by generative AI tools, he added.

AI systems can distinguish between fake and real attacks by recognizing ransomware behavior. The system can swiftly identify suspicious behavior, including unauthorized key generation, Zack Moore, a product security manager at InterVision, said in an interview with PYMNTS.

Generative AI, especially large language models (LLMs), allows organizations to simulate potential attacks and identify their weaknesses. Moore said that the most effective use of AI in uncovering and dissecting attacks lies in ongoing penetration testing.

Instead of simulating an attack once every year, organizations can rely on AI-empowered penetration testing to constantly verify their systems fortitude, he said. Furthermore, technicians can review the tools logs to reverse-engineer a solution after identifying a vulnerability.

The game of cat and mouse between attackers and defenders using AI is likely to continue indefinitely. Meanwhile, consumers are concerned about how to keep their data safe. A recent PYMNTS Intelligencestudy showed that people who love using online shopping features care the most about keeping their data safe, with 40% of shoppers in the U.S. saying its their top worry or very important.

Originally posted here:
Firms Turn to AI for Smarter Cybersecurity Solutions - PYMNTS.com

The holy grail in AI development is explainable AI, which is a means to reveal the decision-making processes that the AI model used to arrive at its output. In other words, we humans want to know why the AI did what it did before we staked our careers, lives, or businesses on its outputs.

Causal AI requires models to explain their prediction. In its simplest form, the explanation is a graph representing a cause-and-effect chain, says George Williams, GSI Technologys director of ML, data science and embedded AI. In its modern form, its a human understandable explanation in the form of text, he says.

Typically, AI models have no auditable trails in its decision-making, no self-reporting mechanisms, and no way to peer behind the cloaking curtains of increasingly complicated algorithms.

Traditional predictive AI can be likened to a black box where its nearly impossible to tell what drove an individual result, says Phil Johnson, VP data solutions at mPulse.

As a result, humans can trust hardly anything an AI model delivers. The output could be a hallucination -- a lie, fabrication, miscalculation, or a fairytale, depending on how generous you want to be in labeling such errors and what type of AI model is being used.

GenAI models still have the unfortunate side-effect of hallucinating or making up facts sometimes. This means they can also hallucinate their explanations. Hallucination mitigation is a rapidly evolving area of research, and it can be difficult for organizations to keep up with the latest research/techniques, says Williams.

Related:5 Ways to Use AI You May Have Never Even Considered

On the other hand, that same AI model could reveal a profound truth humans cannot see because their view is obscured by huge volumes of data.

Like the proverbial army of monkeys pounding on keyboards may one day produce a great novel, many crowds of humans may one day trip across an important insight buried in ginormous stores of data. Or we can lean on the speed of AI to find a useful answer now and focus on teaching it to reveal how it came to that conclusion. The latter is far more manageable than the former.

If one gets anything out of the experience of working with AI, it should be the re-discovery of the marvel that is the human brain. The more we fashion AI after our own brains, the more ways we find it a mere shadow of our own astounding capabilities.

And thats not a diss on AI, which is a truly astounding invention and itself a testament to human capabilities. Nonetheless, the creators truly want to know what the creation is actually up to.

Related:What to Know About Machine Customers

Most AI/ML is correlational in nature, not causal, explains David Guarrera, EY Americas generative AI leader. So, you cant say much about the direction of the effect. If age and salary correlate, you dont technically know if being older CAUSES you to have more money or money CAUSES you to age, he says.

Most of us would intuitively agree that its the lack of money that causes one to age, but we cant reliably depend on our intuition to evaluate the AIs output. Neither can we rely on AI to explain itself -- mostly because it wasnt designed to do so.

In many advanced machine learning models such as deep learning, massive amounts of data are ingested to create a model, says Judith Hurwitz, chief evangelist, Geminos Software and author of Causal Artificial Intelligence: The Next Step in Effective Business AI. One of the key issues with this approach to AI is that the models created by the data cannot be easily understood by the business. They are, therefore, not explainable.In addition, it is easy to create a biased result depending on the quality of the data used to create the model, she says.

This issue is commonly referred to as AIs black box. Breaking into the innards of an AI model to retrieve the details of its decision-making is no small task, technically speaking.

Related:Implementing Generative AI for Business Success

This involves the use of causal inference theories and graphical models, such as directed acyclic graphs (DAGs), which help in mapping out and understanding the causal relationships between variables, says Ryan Gross, head of data and applications at Caylent. By manipulating one variable, causal AI can observe and predict how this change affects other variables, thereby identifying cause-and-effect relationships.

Traditional AI models are fixed in time and understand nothing. Causal AI is a different animal entirely.

Causal AI is dynamic, whereas comparable tools are static. Causal AI represents how an event impacts the world later. Such a model can be queried to find out how things might work, says Brent Field at Infosys Consulting. On the other hand, traditional machine learning models build a static representation of what correlates with what. They tend not to work well when the world changes, something statisticians call nonergodicity, he says.

Its important to grok why this one point of nonergodicity is such a crucial difference to almost everything we do.

Nonergodicity is everywhere. Its this one reason why money managers generally underperform the S&P 500 index funds. Its why election polls are often off by many percentage points. Commercial real estate and global logistics models stopped working about March 15, 2020, because COVID caused this massive supply-side economic shock that is still reverberating through the world economy, Field explains.

Without knowing the cause of an event or potential outcome, the knowledge we extract from AI is largely backward facing even when it is forward predicting. Outputs based on historical data and events alone are by nature handicapped and sometimes useless. Causal AI seeks to remedy that.

Causal models allow humans to be much more involved and aware of the decision-making process. Causal models are explainable and debuggable by default -- meaning humans can trust and verify results -- leading to higher trust, says Joseph Reeve, software engineering manager at Amplitude. Causal models also allow human expertise through model design to be leveraged when training a model, as opposed to traditional models that need to be trained from scratch, without human guidance, he says.

Can causal AI be applied even to GenAI models? In a word, yes.

We could use causal AI to analyze a large amount of data and pair it with GenAI to visualize the analysis using graphics or explanations, says Mohamed Abdelsadek, EVP data, insights, and analytics at Mastercard. Or, on the flip side, GenAI could be engaged to identify the common analysis questions at the beginning, such as the pictures of damage caused by a natural event, and causal AI would be brought in to execute the data processing and analysis, he says.

There are other ways causal AI and GenAI can work together, too.

Generative AI can be an effective tool to support causal AI. However, keep in mind that GenAI is a tool not a solution, says Geminos Softwares Hurwitz. One of the emerging ways that GenAI can be hugely beneficial in causal AI is to use these tools to analyze subject matter information stored in both structured and instructed formats. One of the essential areas needed to create an effective causal AI solution is the need for what is called causal discovery -- determining what data is needed to understand cause and effect, she says.

Does this mean that causal AI is a panacea for all of AI or that it is an infallible technology?

Causal AI is a nascent field. Because the technology is not completely developed yet, the error rates tend to be higher than expected, especially in domains that dont have sufficient training for the AI system, says Flavio Villanustre, global chief information security officer of LexisNexis Risk Solutions. However, you should expect this to improve significantly with time.

So where does causal AI stand in the scheme of things?

In 2022 Gartner Hype Cycle, causal AI was deemed as more mature and ahead of generative AI, says Ed Watal, founder and principal at Intellibus. However, unlike generative AI, causal AI has not yet found a mainstream use case and adoption that tools like ChatGPT have provided over generative AI models like GPT, he says.

Read this article:
Causal AI: AI Confesses Why It Did What It Did - InformationWeek

If you thought the software supply chain security problem was difficult enough today, buckle up. The explosive growth in artificial intelligence (AI) use is about to make those supply chain issues exponentially harder to navigate in the years to come.

Developers, application security pros, and DevSecOps professionals are called to fix the highest risk flaws that lurk in what seems like the endless combinations of open source and proprietary components that are woven into their applications and cloud infrastructure. But it's a constant battle trying to even understand which components they have, which ones are vulnerable, and which flaws put them most at risk. Clearly, they're already struggling to sanely manage these dependencies in their software as it is.

What's going to get harder is the multiplier effect that AI stands to add to the situation.

AI and machine learning (ML)-enabled tools are software just the same as any other kind of application and their code is just as likely to suffer from supply chain insecurities. However, they add another asset variable to the mix that greatly increases the attack surface of the AI software supply chain: AI/ML models.

"What separates AI applications from every other form of software is that [they rely] in some way or fashion on a thing called a machine learning model," explains Daryan Dehghanpisheh, co-founder of Protect AI. "As a result, that machine learning model itself is now an asset in your infrastructure. When you have an asset in your infrastructure, you need the ability to scan your environment, identify where they are, what they contain, who has permissions, and what they do. And if you can't do that with models today, you can't manage them."

AI/ML models provide the foundation for an AI system's ability to recognize patterns, make predictions, make decisions, trigger actions, or create content. But the truth is that most organizations don't even know how to even start gaining visibility into all of the AI models embedded in their software. Models and the infrastructure around them are built differently than other software components, and traditional security and software tooling isn't built to scan for or understand how AI models work or how they're flawed. This is what makes them unique, says Dehghanpisheh, who explains that they're essentially hidden pieces of self-executing code.

"A model, by design, is a self-executing piece of code. It has a certain amount of agency," says Dehghanpisheh. "If I told you that you have assets all over your infrastructure that you can't see, you can't identify, you don't know what they contain, you don't know what the code is, and they self-execute and have outside calls, that sounds suspiciously like a permission virus, doesn't it?"

Getting ahead of this issue was the big impetus behind him and his co-founders launching Protect AI in 2022, which is one of a spate of new firms cropping up to address model security and data lineage issues that are looming in the AI era. Dehghanpisheh and co-founder Ian Swanson saw a glimpse of the future when they worked previously together building AI/ML solutions at AWS. Dehghanpisheh had been the global leader for AI/ML solution architects.

"During the time that we spent together at AWS, we saw customers building AI/ML systems at an incredibly rapid pace, long before generative AI captured the hearts and minds of everyone from the C-suite to Congress," he says, explaining that he worked with a range of engineers and business development experts, as well as extensively with customers. "That's when we realized how and where the security vulnerabilities unique to AI/ML systems are."

They observed three basic things about AI/ML that had incredible implications for the future of cybersecurity, he says. The first was that the pace of adoption was so fast that they saw firsthand how quickly shadow IT entities were cropping up around AI development and business use that escaped the kind of governance that would oversee any other kind of development in the enterprise.

The second was that the majority of tools that were being used whether commercial or open source were built by data scientists and up-and-coming ML engineers who had never been trained in security concepts.

"As a result, you had really useful, very popular, very distributed, widely adopted tools that weren't built with a security-first mindset," he says.

As a result, many AI/ML systems and shared tools lack the basics in authentication and authorization and often grant too much read and write access in file systems, he explains. Coupled with insecure network configurations and then those inherent problems in the models, organizations start getting bogged down cascading security issues in these highly complex, difficult-to-understand systems.

"That made us realize that the existing security tools, processes, frameworks no matter how shift left you went, were missing the context that machine learning engineers, data scientists, and AI builders would need," he says.

Finally, the third major observation he and Swanson made during those AWS days was that AI breaches weren't coming. They had already arrived.

"We saw customers have breaches on a variety of AI/ML systems that should have been caught but weren't," he says. "What that told us is that the set and the processes, as well as the incident response management elements, were not purpose-built for the way AI/ML was being architected. That problem has become much worse as generative AI picked up momentum."

Dehghanpisheh and Swanson also started seeing how models and training data were creating a unique new AI supply chain that would need to be considered just as seriously as the rest of the software supply chain. Just like with the rest of modern software development and cloud-native innovation, data scientists and AI experts have fueled advancements in AI/ML systems through rampant use of open source and shared componentry including AI models and the data used to train them. So many AI systems, whether academic or commercial, are built using someone else's model. And as with the rest of modern development, the explosion in AI development keeps driving a huge daily influx of new model assets proliferated across the supply chain, which means keeping track of them just keeps getting harder.

Take Hugging Face, for example. This is one of the most widely used repositories of open source AI models online today its founders say they want to be the GitHub of AI. Back in November 2022, Hugging Face users had shared 93,501 different models with the community. The following November, that had blown up to 414,695 models. Now, just three months later, that number has expanded to 527,244. This is an issue whose scope is snowballing by the day. And it is going to put the software supply chain security problem "on steroids," says Dehghanpisheh.

A recent analysis by his firm found thousands of models that are openly shared on Hugging Face can execute arbitrary code on model load or inference. While Hugging Face does some basic scanning of its repository for security issues, many models are missed along the way at least half of the highly risk models discovered in the research were not deemed unsafe by the platform, and Hugging Face makes it clear in documentation that determining the safety of a model is ultimately the responsibility of its users.

Dehghanpisheh believes the lynchpin of cybersecurity in the AI era will start first by creating a structured understanding of AI lineage. That includes model lineage and data lineage, which are essentially the origin and history of these assets, how they've been changed, and the metadata associated with them.

"That's the first place to start. You can't fix what you can't see and what you can't know and what you can't define, right?" he says.

Meantime, on the daily operational level Dehghanpisheh believes organizations need to build out capabilities to scan their models, looking for flaws that can impact not only the hardening of the system but the integrity of its output. This includes issues like AI bias and malfunction that could cause real-world physical harm from, say, an autonomous car crashing into a pedestrian.

"The first thing is you need to scan," he says. "The second thing is you need to understand those scans. And the third is then once you have something that's flagged, you essentially need to stop that model from activating. You need to restrict its agency."

MLSecOps is a vendor-neutral movement that mirrors the DevSecOps movement in the traditional software world.

"Similar to the move from DevOps to DevSecOps, you've got to do two things at once. The first thing you've got to do is make the practitioners aware that security is a challenge and that it is a shared responsibility," Dehghanpisheh says. "The second thing you've got to do is give context and put security into tools that keep data scientists, machine learning engineers, [and] AI builders on the bleeding edge and constantly innovating, but allowing the security concerns to disappear into the background."

In addition, he says organizations are going to have to start adding governance, risk, and compliance policies and enforcement capabilities and incident response procedures that help govern the actions and processes that take place when insecurities are discovered. As with a solid DevSecOps ecosystem, this means that MLSecOps will need strong involvement from business stakeholders all the way up the executive ladder.

The good news is that AI/ML security is benefiting from one thing that no other rapid technology innovation has had right out of the gate namely, regulatory mandates right out of the gate.

"Think about any other technology transition," Dehghanpisheh says. "Name one time that a federal regulator or even state regulators have said this early on, 'Whoa, whoa, whoa, you've got to tell me everything that's in it. You've got to prioritize knowledge of that system. You have to prioritize a bill of materials. There isn't any."

This means that many security leaders are more likely to get buy-in to build out AI security capabilities a lot earlier in the innovation life cycle. One of the most obvious signs of this support is the rapid shift to sponsor new job functions at organizations.

"The biggest difference that the regulatory mentality has brought to the table is that in January of 2023, the concept of a director of AI security was novel and didn't exist. But by June, you started seeing those roles," Dehghanpisheh says. "Now they're everywhere and they're funded."

Read the original here:
It's 10 p.m. Do You Know Where Your AI Models Are Tonight? - Dark Reading

Structured Query Language (SQL) is a complex language that requires an understanding of databases and metadata. Today, generative AI can enable people without SQL knowledge. This generative AI task is called text-to-SQL, which generates SQL queries from natural language processing (NLP) and converts text into semantically correct SQL. The solution in this post aims to bring enterprise analytics operations to the next level by shortening the path to your data using natural language.

With the emergence of large language models (LLMs), NLP-based SQL generation has undergone a significant transformation. Demonstrating exceptional performance, LLMs are now capable of generating accurate SQL queries from natural language descriptions. However, challenges still remain. First, human language is inherently ambiguous and context-dependent, whereas SQL is precise, mathematical, and structured. This gap may result in inaccurate conversion of the users needs into the SQL thats generated. Second, you might need to build text-to-SQL features for every database because data is often not stored in a single target. You may have to recreate the capability for every database to enable users with NLP-based SQL generation. Third, despite the larger adoption of centralized analytics solutions like data lakes and warehouses, complexity rises with different table names and other metadata that is required to create the SQL for the desired sources. Therefore, collecting comprehensive and high-quality metadata also remains a challenge. To learn more about text-to-SQL best practices and design patterns, see Generating value from enterprise data: Best practices for Text2SQL and generative AI.

Our solution aims to address those challenges using Amazon Bedrock and AWS Analytics Services. We use Anthropic Claude v2.1 on Amazon Bedrock as our LLM. To address the challenges, our solution first incorporates the metadata of the data sources within the AWS Glue Data Catalog to increase the accuracy of the generated SQL query. The workflow also includes a final evaluation and correction loop, in case any SQL issues are identified by Amazon Athena, which is used downstream as the SQL engine. Athena also allows us to use a multitude of supported endpoints and connectors to cover a large set of data sources.

After we walk through the steps to build the solution, we present the results of some test scenarios with varying SQL complexity levels. Finally, we discuss how it is straightforward to incorporate different data sources to your SQL queries.

There are three critical components in our architecture: Retrieval Augmented Generation (RAG) with database metadata, a multi-step self-correction loop, and Athena as our SQL engine.

We use the RAG method to retrieve the table descriptions and schema descriptions (columns) from the AWS Glue metastore to ensure that the request is related to the right table and datasets. In our solution, we built the individual steps to run a RAG framework with the AWS Glue Data Catalog for demonstration purposes. However, you can also use knowledge bases in Amazon Bedrock to build RAG solutions quickly.

The multi-step component allows the LLM to correct the generated SQL query for accuracy. Here, the generated SQL is sent for syntax errors. We use Athena error messages to enrich our prompt for the LLM for more accurate and effective corrections in the generated SQL.

You can consider the error messages occasionally coming from Athena like feedback. The cost implications of an error correction step are negligible compared to the value delivered. You can even include these corrective steps as supervised reinforced learning examples to fine-tune your LLMs. However, we did not cover this flow in our post for simplicity purposes.

Note that there is always inherent risk of having inaccuracies, which naturally comes with generative AI solutions. Even if Athena error messages are highly effective to mitigate this risk, you can add more controls and views, such as human feedback or example queries for fine-tuning, to further minimize such risks.

Athena not only allows us to correct the SQL queries, but it also simplifies the overall problem for us because it serves as the hub, where the spokes are multiple data sources. Access management, SQL syntax, and more are all handled via Athena.

The following diagram illustrates the solution architecture.

Figure 1. The solution architecture and process flow.

The process flow includes the following steps:

At this stage, the process is ready to receive the query in natural language. Steps 79 represent a correction loop, if applicable.

For this post, you should complete the following prerequisites:

You can use the following Jupyter notebook, which includes all the code snippets provided in this section, to build the solution. We recommend using Amazon SageMaker Studio to open this notebook with an ml.t3.medium instance with the Python 3 (Data Science) kernel. For instructions, refer to Train a Machine Learning Model. Complete the following steps to set up the solution:

In this section, we run our solution with different example scenarios to test different complexity levels of SQL queries.

To test our text-to-SQL, we use two datasets available from IMDB. Subsets of IMDb data are available for personal and non-commercial use. You can download the datasets and store them in Amazon Simple Storage Service (Amazon S3). You can use the following Spark SQL snippet to create tables in AWS Glue. For this example, we use title_ratings and title:

In this scenario, our dataset is stored in an S3 bucket. Athena has an S3 connector that allows you to use Amazon S3 as a data source that can be queried.

For our first query, we provide the input I am new to this. Can you help me see all the tables and columns in imdb schema?

The following is the generated query:

The following screenshot and code show our output.

For our second query, we ask Show me all the title and details in US region whose rating is more than 9.5.

The following is our generated query:

The response is as follows.

For our third query, we enter Great Response! Now show me all the original type titles having ratings more than 7.5 and not in the US region.

The following query is generated:

We get the following results.

This scenario simulates a SQL query that has syntax issues. Here, the generated SQL will be self-corrected based on the response from Athena. In the following response, Athena gave a COLUMN_NOT_FOUND error and mentioned that table_description cant be resolved:

To use the solution with other data sources, Athena handles the job for you. To do this, Athena uses data source connectors that can be used with federated queries. You can consider a connector as an extension of the Athena query engine. Pre-built Athena data source connectors exist for data sources like Amazon CloudWatch Logs, Amazon DynamoDB, Amazon DocumentDB (with MongoDB compatibility), and Amazon Relational Database Service (Amazon RDS), and JDBC-compliant relational data sources such MySQL, and PostgreSQL under the Apache 2.0 license. After you set up a connection to any data source, you can use the preceding code base to extend the solution. For more information, refer to Query any data source with Amazon Athenas new federated query.

To clean up the resources, you can start by cleaning up your S3 bucket where the data resides. Unless your application invokes Amazon Bedrock, it will not incur any cost. For the sake of infrastructure management best practices, we recommend deleting the resources created in this demonstration.

In this post, we presented a solution that allows you to use NLP to generate complex SQL queries with a variety of resources enabled by Athena. We also increased the accuracy of the generated SQL queries via a multi-step evaluation loop based on error messages from downstream processes. Additionally, we used the metadata in the AWS Glue Data Catalog to consider the table names asked in the query through the RAG framework. We then tested the solution in various realistic scenarios with different query complexity levels. Finally, we discussed how to apply this solution to different data sources supported by Athena.

Amazon Bedrock is at the center of this solution. Amazon Bedrock can help you build many generative AI applications. To get started with Amazon Bedrock, we recommend following the quick start in the following GitHub repo and familiarizing yourself with building generative AI applications. You can also try knowledge bases in Amazon Bedrock to build such RAG solutions quickly.

Sanjeeb Panda is a Data and ML engineer at Amazon. With the background in AI/ML, Data Science and Big Data, Sanjeeb design and develop innovative data and ML solutions that solve complex technical challenges and achieve strategic goals for global 3P sellers managing their businesses on Amazon. Outside of his work as a Data and ML engineer at Amazon, Sanjeeb Panda is an avid foodie and music enthusiast.

Burak Gozluklu is a Principal AI/ML Specialist Solutions Architect located in Boston, MA. He helps strategic customers adopt AWS technologies and specifically Generative AI solutions to achieve their business objectives. Burak has a PhD in Aerospace Engineering from METU, an MS in Systems Engineering, and a post-doc in system dynamics from MIT in Cambridge, MA. Burak is still a research affiliate in MIT. Burak is passionate about yoga and meditation.

Excerpt from:
Build a robust text-to-SQL solution generating complex queries, self-correcting, and querying diverse data sources ... - AWS Blog

We previously reported that Meta Platforms, the parent company of Facebook, plans to deploy its own custom-designed artificial intelligence chips, codenamed Artemis, into its data centers this year, but would continue using Nvidia H100 GPUs alongside them - for the foreseeable future at least.

However, The Register now claims job advertisements for ASIC engineers with expertise in architecture, design, and testing have been spotted in Bangalore, India, and Sunnyvale, California, indicating Meta's intentions to develop its own AI hardware.

The job descriptions suggest that Meta is seeking professionals to "help architect state-of-the-art machine learning accelerators" and to design complex SoCs and IPs for datacenter applications. Some of these roles were initially posted on LinkedIn in late December 2023 and re-posted two weeks ago, with the Sunnyvale roles offering salaries nearing $200,000.

While the exact nature of Meta's project remains unspecified, it's likely linked to the company's previously announced "Meta Training Inference Accelerators," set to be launched later this year.

Meta's ambitions also extend to artificial general intelligence, a venture that might necessitate specialized silicon.

With the increasing demand for AI and Nvidia's struggle to meet this demand, Meta's move to develop its own technology is a strategic step to ensure it doesn't have to compete with rivals for hardware in a super-hot market.

The Register reports that the Indian government will likely welcome Meta's decision to advertise in Bangalore, as the nation seeks to become a significant player in the global semiconductor industry.

In addition, rumors suggest that Microsoft is also reducing its dependence on Nvidia by developing a server networking card to optimize machine-learning workload performance. This trend suggests Nvidia's most formidable rivals are looking for ways to become less reliant on its massively in-demand hardware.

Read this article:
Meta looking to use exotic, custom CPU in its datacenters for machine learning and AI yet another indication that ... - TechRadar

Introduction

Background

Leveraging Machine Learning

Methodology

Data Collection and Preprocessing

Model selection

Algorithmic Framework

Model Training

Feature Importance and Analysis

Model evaluation

Continuous learning and Adaptation

Conclusion

The Automotive Market has been dynamically increasing all over the World, therefore making it difficult to estimate the value of a used car. Due to the increase and rise of new technologies, people have found it difficult to use traditional methods, therefore relying on todays technology due to some of the factors influencing the cars worth. This article will guide us on how to predict the price of a used car using todays technology, that is machine learning. We are going to see how these algorithms have led to cheap means of predicting a cars worth. Our model will provide a more precise and adaptable solution for determining the accurate price of used cars by using the historical data .

Understanding the market of used cars is very pivotal in coming up with the best technology to solve the problems that have been encountered in the field. The challenges of traditional methods of pricing have been increasing day by day as the market evolves. These traditional methods fail to keep in pace with the new features of the automotive market therefore making it important to come up with a model that solves the challenge. Factors influencing the price of a car including make, model, year of manufacturer, among others are vital when coming up with an effective model.

The power of machine learning is of unlimited strength and use. By using sophisticated datasets, employing effective algorithms to models may lead to an accurate prediction of used car price. Machine learning models may keep in pace with the dynamic and fluctuating market therefore making it more effective. Our project will use this technology to assess the worth of pre owned vehicles.

We start by collecting the data and making them available for the next steps. Diverse set of features such as make, model, year ,customer review among others allows machine learning to make predictions on the price of used cars. Gathering these comprehensive datasets is the first step.

The collected data is then cleaned and preprocessed so that the model is not hindered by missing values, inconsistencies and outliers. Cleaning and preprocessing data ensures that the data is ready to be trained by machine model for a high performance.

In This phase, we are going to make a choice on which model is essential and effective for our problem. Regression models such as linear regression or decision tree regression are commonly used in predicting numerical values making them ideal for predicting the price of used cars. In our case we are going to use a regression model.

Our algorithmic framework involves our chosen model as a regression model. Regression model allows us to establish a relationship between the selected variables with the target variable which is our car price. We will train our model on a subset of dataset using optimization techniques to minimize prediction errors.

Training involves feeding the model with a cleaned dataset to make it easy for the model to learn the patterns and relationship between the input variables provided and target output variable. Iterative adjustments are made to the model based on its performance and until optimal accuracy is achieved.

In order for the machine learning model to be effective and of high performance it is important to understand the features which significantly impact the price of used cars. To provide insights to feature important machine learning models ensemble analysis techniques such as random forests or gradient boosting. These analyses aid at making informed decisions rather than providing interpretability of the model regarding the pricing strategies.

In order to ensure accuracy and reliability to our machine learning model, its essential for the model to be evaluated. Different metrics are used to evaluate machine learning models such as Mean Absolute Error(MAE), Mean Squared Error(MSE), and R-squared. The model performance must be assessed on different subsets of data by cross-validation techniques to minimize the risk of overfitting.

The automotive market, especially the car market, is a dynamic field that is influenced by economic fluctuations, consumer preferences and other external factors. However, machine learning that is designed to predict the price of used cars must be adaptable and capable of continuous learning. These models require regular updates to the model incorporating new data and retraining at intervals to ensure its relevance and accuracy over time.

By using the historical datasets and sophisticated algorithms, we can now unravel a web of factors influencing the value of pre owned cars. Our articles embrace the rise of new technology of machine learning by predicting the price of used cars. By using the datasets and machine learning algorithms we have been able to offer a dynamic and accurate solution to the challenges of assessing used cars.

As we bid farewell to outdated traditional methodologies, the adoption of machine learning in the used car market brings forth a new era of precision and efficiency. The ability to adapt to changing dynamic positions, machine learning is an important tool to both sellers and buyers, offering transparency, accuracy and a glimpse to the future of the automotive market.

Our article describes this technology as a paradigm shift of how we perceive and navigate the realm of pre owned vehicles rather than an advancement of technology. Use of machine learning is not a choice but a step towards a more informed and dynamic automotive future.

Call-to-Action

We invite fellow enthusiasts and industry professionals to explore the possibilities of machine learning in their projects. Embrace the data-driven revolution and contribute to the evolution of predictive modeling in diverse domains. For those intrigued by the technical aspects of our project, further details, code snippets, and datasets are available here;

https://github.com/mkwasi5930/used-car-price-prediction

Stay tuned for future updates as we continue refining and expanding our used car price prediction model, pushing the boundaries of what is possible in the dynamic world of machine learning and automotive valuation.

For more articles, tutorials and updates you can follow me here:

https://github.com/mkwasi5930

https://www.linkedin.com/in/abednego-mutuku-a91935236?utm_source=share&utm_campaign=share_via&utm_content=profile&utm_medium=android_app

https://twitter.com/mkwasi_?t=_P1YiYUIZDRtiAMg5hC1nQ&s=09

Continue reading here:
Predicting The Price Of Used Cars Using Machine Learning - Medium

March 01, 2024 -The American Heart Association (AHA) released a scientific statement in Circulation this week detailing the current state of artificial intelligence (AI) use in the diagnosis and treatment of cardiovascular disease.

The statement is the first of its kind from the AHA, underscoring continued interest from healthcare organizations in how AI could potentially transform the industry. The report outlined limitations of these technologies, potential applications, challenges, and how AI may be deployed safely and effectively.

Here, we present the state-of-the-art including the latest science regarding specific AI usesfrom imaging and wearables to electrocardiography and genetics, said the chair of the statements writing committee Antonis Armoundas, PhD, a principal investigator at the Cardiovascular Research Center at Massachusetts General Hospital and an associate professor of medicine at Harvard Medical School, in a press release. Among the objectives of this manuscript is to identify best practices as well as gaps and challenges that may improve the applicability of AI tools in each area.

Multiple factors limiting the use of AI in cardiovascular care were described: lack of protocols for appropriate information sourcing and sharing; legal and ethical hurdles; the need to grow the scientific knowledge base around these technologies; and the absence of robust regulatory pathways, among others.

Robust prospective clinical validation in large diverse populations that minimizes various forms of bias is essential to address uncertainties and bestow trust, which, in turn, will help to increase clinical acceptance and adoption, Armoundas noted.

The statement also reviewed potential cardiovascular applications for AI tools, some of which are already in use.

AI and machine learning have significant potential to improve medical imaging, but challenges abound. The AHAs statement emphasized that using these tools for image interpretation is difficult due to a lack of representative, high-quality datasets, and further indicated that these technologies need to be validated in each potential use case prior to deployment.

AI could also be useful in interpreting information from implants, wearables, electrocardiograms, and genetic data.

Numerous applications already exist where AI/machine learning-based digital tools can improve screening, extract insights into what factors improve an individual patients health and develop precision treatments for complex health conditions, said Armoundas.

The statement also asserted that education and research are crucial to making good on the promise of healthcare AI.

There is an urgent need to develop programs that will accelerate the education of the science behind AI/machine learning tools, thus accelerating the adoption and creation of manageable, cost-effective, automated processes. We need more AI/machine learning-based precision medicine tools to help address core unmet needs in medicine that can subsequently be tested in robust clinical trials, Armoundas continued. This process must organically incorporate the need to avoid bias and maximize generalizability of findings in order to avoid perpetuating existing health care inequities.

The AHA is the latest national healthcare stakeholder to weigh in on how AI should be implemented across the industry.

This week, the American Medical Association (AMA) and Manatt Health published The Emerging Landscape of Augmented Intelligence in Health Care report, which outlines key terms, potential use cases, and risks associated with these tools.

The report explored both clinical and administrative applications for AI in an effort to assist clinicians as they navigate the implementation of the technology.

Alongside opportunities and risks, the AMA also laid out critical questions that healthcare organizations should be asking themselves as they consider adopting AI and other advanced analytics tools.

View original post here:
AHA Issues Statement on the Use of AI in Cardiovascular Care - HealthITAnalytics.com