Were you unable to attend Transform 2022? Check out all of the summit sessions in our on-demand library now! Watch here.

Todays demand for real-time data analytics at the edge marks the dawn of a new era in machine learning (ML): edge intelligence. That need for time-sensitive data is, in turn, fueling a massive AI chip market, as companies look to provide ML models at the edge that have less latency and more power efficiency.

Conventional edge ML platforms consume a lot of power, limiting the operational efficiency of smart devices, which live on the edge. Thosedevices are also hardware-centric, limiting their computational capability and making them incapable of handling varying AI workloads. They leverage power-inefficient GPU- or CPU-based architectures and are also not optimized for embedded edge applications that have latency requirements.

Even though industry behemoths like Nvidia and Qualcomm offer a wide range of solutions, they mostly use a combination of GPU- or data center-based architectures and scale them to the embedded edge as opposed to creating a purpose-built solution from scratch. Also, most of these solutions are set up for larger customers, making them extremely expensive for smaller companies.

In essence, the $1 trillion global embedded-edge market is reliant on legacy technology that limits the pace of innovation.

MetaBeat 2022

MetaBeat will bring together thought leaders to give guidance on how metaverse technology will transform the way all industries communicate and do business on October 4 in San Francisco, CA.

ML company Sima AI seeks to address these shortcomings with its machine learning-system-on-chip (MLSoC) platform that enables ML deployment and scaling at the edge. The California-based company, founded in 2018, announced today that it has begun shipping the MLSoC platform for customers, with an initial focus of helping solve computer vision challenges in smart vision, robotics, Industry 4.0, drones, autonomous vehicles, healthcare and the government sector.

The platform uses a software-hardware codesign approach that emphasizes software capabilities to create edge-ML solutions that consume minimal power and can handle varying ML workloads.

Built on 16nm technology, the MLSoCs processing system consists of computer vision processors for image pre- and post-processing, coupled with dedicated ML acceleration and high-performance application processors. Surrounding the real-time intelligent video processing are memory interfaces, communication interfaces, and system management all connected via a network-on-chip (NoC). The MLSoC features low operating power and high ML processing capacity, making it ideal as a standalone edge-based system controller, or to add an ML-offload accelerator for processors, ASICs and other devices.

The software-first approach includes carefully-defined intermediate representations (including the TVM Relay IR), along with novel compiler-optimization techniques. This software architecture enables Sima AI to support a wide range of frameworks (e.g., TensorFlow, PyTorch, ONNX, etc.) and compile over 120+ networks.

Many ML startups are focused on building only pure ML accelerators and not an SoC that has a computer-vision processor, applications processors, CODECs, and external memory interfaces that enable the MLSoC to be used as a stand-alone solution not needing to connect to a host processor. Other solutions usually lack network flexibility, performance per watt, and push-button efficiency all of which are required to make ML effortless for the embedded edge.

Sima AIs MLSoC platform differs from other existing solutions as it solves all these areas at the same time with its software-first approach.

The MLSoC platform is flexible enough to address any computer vision application, using any framework, model, network, and sensor with any resolution. Our ML compiler leverages the open-source Tensor Virtual Machine (TVM) framework as the front-end, and thus supports the industrys widest range of ML models and ML frameworks for computer vision, Krishna Rangasayee, CEO and founder of Sima AI, told VentureBeat in an email interview.

From a performance point of view, Sima AIs MLSoC platform claims to deliver 10x better performance in key figures of merit such as FPS/W and latency than alternatives.

The companys hardware architecture optimizes data movement and maximizes hardware performance by precisely scheduling all computation and data movement ahead of time, including internal and external memory to minimize wait times.

Sima AI offers APIs to generate highly optimized MLSoC code blocks that are automatically scheduled on the heterogeneous compute subsystems. The company has created a suite of specialized and generalized optimization and scheduling algorithms for the back-end compiler that automatically convert the ML network into highly optimized assembly codes that run on the machine learning-accelerator (MLA) block.

For Rangasayee, the next phase of Sima AIs growth is focused on revenue and scaling their engineering and business teams globally. As things stand, Sima AI has raised $150 million in funding from top-tier VCs such as Fidelity and Dell Technologies Capital. With the goal of transforming the embedded-edge market, the company has also announced partnerships with key industry players like TSMC, Synopsys, Arm, Allegro, GUC and Arteris.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.

Link:
Machine learning at the edge: The AI chip company challenging Nvidia and Qualcomm - VentureBeat

Many of todays marketing processes are powered by AI and machine learning. Discover how these technologies are shaping the future of customer experience.

By using artificial intelligence (AI) and machine learning (ML) along with analytics, brands are in a much better position to elevate customer service experiences at every touchpoint and create positive emotional connections.

This article will look at the ways that AI and ML are used by brands to improve customer service and support.

AI improves the customer service journey in several ways, including tracking conversations in real-time, providing feedback to service agents and using intelligence to monitor language, speech patterns and psychographic profiles to predict future customer needs.

This functionality can also drastically enhance the effectiveness of customer relationship management (CRM) and customer data platforms (CDP).

CRM platforms, including C2CRM, Salesforce Einstein and Zoho, have integrated AI into their software to provide real-time decisioning, predictive analysis and conversational assistants, all of which help brands more fully understand and engage their customers.

CDPs, such as Amperity, BlueConic, Adobes Real-Time CDP and ActionIQ, have also integrated AI into more traditional capabilities to unify customer data and provide real-time functionality and decisoning. This technology enables brands to gain a deeper understanding of what their customers want, how they feel and what they are most likely to do next.

Related Article: What's Next for Artificial Intelligence in Customer Experience?

Artificial intelligence and machine learning are now used for gathering and analyzing social, historical and behavioral data, which allows brands to gain a much more complete understanding of their customers.

Because AI continuously learns and improves from the data it analyzes, it can anticipate customer behavior. As such, AI- and ML-driven chatbots can provide customers with a more personalized, informed conversation that can easily answer their questions and if not, immediately route them to a live customer service agent.

Bill Schwaab, VP of sales, North America for boost.ai, told CMSWire that ML is used in combination with AI and a number of other deep learning models to support todays virtual customer service agents.

ML on its own may not be sufficient to gain a total understanding of customer requests, but its useful in classifying basic user intent, said Schwaab, who believes that the brightest applications of these technologies in customer service find the balance between AI and human intervention.

Virtual agents are becoming the first line in customer experience in addition to human agents, he explained. Because these virtual agents can resolve service queries quickly and are available outside of normal service hours, human agents can focus on more complex or valuable customer interactions. Round-the-clock availability provides brands with additional time to capture customer input and inform better decision-making.

Swapnil Jain, CEO and co-founder of Observe.AI, said that todays customer service agents no longer have to spend as much time on simpler, transactional interactions, as digital and self-serve options have reduced the volume of those tasks.

"Instead, agents must excel at higher-value, complex behaviors that meaningfully impact CX and revenue," said Jain, adding that brands are harnessing AI and ML to up-level agent skills, which include empathy and active listening. This, in turn, "drives the behavioral changes needed to improve CX performance at speed and scale."

Because customer conversations contain a goldmine of insights for improving agent performance, AI-powered conversation intelligence can help brands with everything from service and support to sales and retention, said Jain. Using advanced interaction analytics, brands can benefit from pinpointing positive and negative CX drivers, advanced tonality-based sentiment and intent analysis and evidence-based agent coaching.

Predictive analytics is the process of using statistics, data mining and modeling to make predictions.

AI can analyze large amounts of data in a very short time, and along with predictive analytics, it can produce real-time, actionable insights that can guide interactions between a customer and a brand. This practice is also referred to as predictive engagement and uses AI to inform a brand when and how to interact with each customer.

Don Kaye, CCO of Exasol, spoke with CMSWire about the ways brands are using predictive analytics as part of their data strategies that link to their overall business objectives.

Weve seen first-hand how businesses use predictive analytics to better inform their organizations decision-making processes to drive powerful customer experiences that result in brand loyalty and earn consumer trust, said Kaye.

As an example, he told CMSWire that banks use supervised learning or regression and classification to calculate the risks of loan defaults or IT departments to detect spam.

With retailers, weve seen them seeking the benefits of deep learning or reinforcement learning, which enables a new level of end-to-end automation, where models become more adaptable and use larger data volumes for increased accuracy, he said.

According to Kaye, businesses with advanced analytics also tend to have agile, open data architectures that promote open access to data, also known as data democratization.

Kaye is a big advocate for AI and ML and believes that the technologies will continue to grow and become routine across all verticals, with the democratization of analytics enabling data professionals to focus on more complex scenarios and making customer experience personalization the norm.

Related Article: What Customer-Centric Predictive Analytics Looks Like

AI-driven sentiment analysis enables brands to obtain actionable insights which facilitate a better understanding of the emotions that customers feel when they encounter pain points or friction along the customer journey as well as how they feel when they have positive, emotionally satisfying experiences.

Julien Salinas, founder and CTO at NLP Cloud, told CMSWire that AI is often used to perform sentiment analysis to automatically detect whether an incoming customer support request is urgent or not. "If the detected sentiment is negative, the ticket is more likely to be addressed quickly by the support team."

Sentiment analysis can automatically detect emotions and opinions by classifying customer text as positive, negative or neutral through the use of AI, natural language processing (NLP) and ML.

Pieter Buteneers, director of engineering in ML and AI at Sinch, said that NLP enables applications to understand, write and speak languages in a manner that is similar to humans.

"It also facilitates a deeper understanding of customer sentiment, he explained. When NLP is incorporated into chatbots and voice bots it permits them to have seemingly human-like language proficiency and adjust their tones during conversations.

When used in conjunction with chatbots, NLP can facilitate human-like conversations based on sentiment. So if a customer is upset, for example, the bot can adjust its tone to diffuse the situation while moving along the conversation, said Buteneers. This would be an intuitive shift for a human, but bots that arent equipped with NLP sentiment analysis could miss the subtle cues of human sentiment in the conversation, and risk damaging the customer relationship."

Buteneers added that breakthroughs in NLP are making an enormous difference in how AI understands input from humans. For example, NLP can be used to perform textual sentiment analysis, which can decipher the polarity of sentiments in text."

Similar to sentiment analysis, AI is also useful for detecting intent. Salinas said that its sometimes difficult to have a quick grasp on a user request, especially when the users message is very long. In that case, AI can automatically extract the main idea from the message so the support agent can act more quickly.

While AI and ML have continued to evolve, and brands have found many ways to use these technologies to improve the customer service experience, the challenges of AI and ML can still be daunting.

Kaye explained that AI models need good data to deliver accurate results, so brands must also focus on quality and governance.

In-memory analytics databases will become the driver of creation, storage and loading features in ML training tools given their analysis capabilities, and ability to scale and deliver optimal time to insight, said Kaye. He added that these tools will benefit from closer integration with the companys data stores, which will enable them to run more effectively on larger data volumes to guarantee greater system scalability.

Iliya Rybchin, partner at Elixirr Consulting, told CMSWire that thanks to ML and the vast amount of data bots are collecting, they are getting better and will continue to improve. The challenge is that they will improve in proportion to the data they receive.

Therefore, if an under-represented minority with a unique dialect is not utilizing a particular service as much as other consumers, the ML will start to discount the aspects of that dialect as outliers vs. common language, said Rybchin.

He explained that the issue is not caused by the technology or programming, but rather, it is the result of the consumer-facing product that is not providing equal access to the bot. The solution is more about bringing more consumers to the product vs. changing how the product is built or designed."

AI and ML have been incorporated into the latest generations of CDP and CRM platforms, and conversational AI-driven bots are assisting service agents and enhancing and improving the customer service experience. Predictive analytics and sentiment analysis, meanwhile, are enabling brands to obtain actionable insights that guide the subsequent interactions between a customer and a brand.

See more here:
4 Ways AI, Analytics and Machine Learning Are Improving Customer Service and Support - CMSWire

What youll learn

Sometimes inspiration comes in the oddest ways. I like to watch CBS News Sunday Morning because of the variety of stories they air. Recently, they did one on Vertical Farming - A New Form of Agriculture (see video below).

CBS News Sunday Morning recently did a piece on vertical farming that spawned this article.

For those who didnt watch the video, vertical farming is essentially a method of indoor farming using hydroponics. Hydroponics isnt new; its a subset of hydroculture where crops are grown without soil. Instead, the plants grow in a mineral-enriched water. This can be done in conjunction with sunlight but typically an artificial light source is used.

The approach is useful in areas that dont provide enough light, or at times or in locations where the temperature or conditions outside would not be conducive for growing plants.

Vertical farming is hydroponics taken to the extreme, with stacks upon stacks of trays with plants under an array of lights. These days, the lights typically are LEDs because of their efficiency and the ability to generate the type of light most useful for plant growth. Automation can be used to streamline planting, support, and harvesting.

A building can house a vertical farm anywhere in the world, including in the middle of a city. Though lots of water is required, its recycled, making it more efficient than other forms of agriculture.

Like many technologies, the opportunities are great if you ignore the details. Thats where my usual contrary nature came into play, though, since I followed up my initial interest by looking for limitations or problems related to vertical farming. Of course, I found quite a few and then noticed that many of the general issues applied to another topic I cover a lotmachine learning/artificial intelligence (ML/AI).

If you made it this far, you know how Im looking at the difference between machine learning and vertical farming. They obviously have no relationship in terms of their technology and implementation, but they do have much in common when one looks at the potential problems and solutions related to those technologies.

As electronic system designers and developers, we constantly deal with potential solutions and their tradeoffs. Machine learning is one of those generic categories that has proven useful in many instances. However, one must be wary of the issues underlying those flashy approaches.

Vertical farming, like machine learning, is something one can dabble in. To be successful, though, it helps to have an expert or at least someone who can quickly gain that experience. This tends to be the case with new and renewed technologies in general. I suspect significantly more ML experts are available these days for a number of reasons like the cost of hardware, but the demand remains high.

Vertical farming uses a good bit of computer automation. The choice of plants, fertilizers, and other aspects of hydropic farming are critical to the success of the farm. Then theres the maintenance aspect. ML-based solutions are one way of reducing the expertise or time required by the staff to support the system.

ML programmers and developers also are able to obtain easier-to-use tools, thereby reducing the amount of expertise and training required to take advantage of ML solutions. These tools often incorporate their own ML models, which are different than those being generated.

Hydroponics works well for many plants, but unfortunately for multiple others, thats not the case. For example, crops like microgreens work well. However, a cherry or apple tree often struggles with this treatment.

ML suffers from the same problem in that its not applicable to all computational chores. But, unlike vertical farms, ML applications and solutions are more diverse. The challenge for developers comes down to understanding where ML is and isnt applicable. Trying to force-fit a machine-learning model to handle a particular problem can result in a solution that provides poor results at high cost.

Vertical farms require power for lighting and to move liquid. ML applications tend to do lots of computation and thus require a good deal of power compared to other computational requirements. One big difference between the two is that ML solutions are scalable and hardware tradeoffs can be significant.

For example, ML hardware can improve performance thats orders of magnitude better than software solutions while reducing power requirements. Likewise, even software-only solutions may be efficient enough to do useful work even while using little power, simply because developers have made the ML models work within the limitations of their design. Vertical farms do not have this flexibility.

Large vertical farms do require a major investment, and theyre not cheap to run due to their scale. The same is true for cloud-based ML solutions utilizing the latest in disaggregated cloud-computing centers. Such data centers are leveraging technologies like SmartNIC and smart storage to use ML models closer to communication and storage than was possible in the past.

The big difference with vertical farming versus ML is scalability. Its now practical for multiple ML models to be running in a smartwatch with a dozen sensors. But that doesnt compare to dealing with agriculture that must scale with the rest of the physical world requirements, such as the plants themselves.

Still, these days, ML does require a significant investment with respect to development and developing the experience to adequately apply ML. Software and hardware vendors have been working to lower both the startup and long-term development costs, which has been further augmented by the plethora of free software tools and low-cost hardware thats now generally available.

Cut the power on a vertical farm and things come to a grinding halt rather quickly, although its not like having an airplane lose power at 10,000 feet. Still, plants do need sustenance and light, though theyre accustomed to changes over time. Nonetheless, responding to failures within the system is important to the systems long-term usefulness.

ML applications tend to require electricity to run, but that tends to be true of the entire system. A more subtle problem with ML applications is the source of input, which is typically sensors such as cameras, temperature sensors, etc. Determining whether the input data is accurate can be challenging; in many cases, designers simply assume that this information is accurate. Applications such as self-driving cars often use redundant and alternative inputs to provide a more robust set of inputs.

Vertical-farming technology continues to change and become more refined, but its still maturing. The same is true for machine learning, though the comparison is like something between a penny bank and Fort Knox. There are simply more ML solutions, many of which are very mature with millions of practical applications.

That said, ML technologies and applications are so varied, and the rate of change so large, that keeping up with whats availablelet alone how things work in detailcan be overwhelming.

Vertical farming is benefiting from advances in technology from robotics to sensors to ML. The ability to track plant growth, germination, and detecting pests are just a few tasks that apply across all of agriculture, including vertical farming.

As with many Whats the Difference articles, the comparisons are not necessarily one-to-one, but hopefully you picked up something about ML or vertical farms that was of interest. Many issues dont map well, like problems of pollination for vertical farms. Though the output of vertical farms will likely feed some ML developers, ML is likely to play a more important part in vertical farming given the level of automation possible with sensors, robots, and ML monitoring now available.

Read more from the original source:
What's the Difference Between Vertical Farming and Machine Learning? - Electronic Design