Snorkel AI cofounders (L to R): Alex Ratner (CEO), Chris R (Board Member), Paroma Varma (Head of ... [+] Solutions), Braden Hancock (Head of Technology), and Henry Ehrenberg (Head of Engineering)

In 2014, computer science PhD candidate Alex Ratner and a team of fellow Stanford PhDs, advised by associate professor and MacArthur Fellow Chris R, were working on a research project at the universitys prominent AI Lab. The main issue they focused on was companies not being able to deploy AI as widely and effectively as they wanted to, due to the costly and time-consuming manual labeling of the data that machine learning models learn from.

Like many academic projects, it was meant to be just an afternoon of messing around and a whiteboard with some math, Ratner says. Soon it turned out that this question that we had started with, of what if we changed the paradigm from labeling by hand to labeling programmatically, was quite interesting to a lot of people.

After spending five years developing the product and deploying it at organizations like Google, Apple, Intel, and the departments of Justice and Defense, in 2019 the research team spun out of the AI Lab and created a company called Snorkel AI.

Today, the enterprise came out of stealth mode announcing that it had raised a total of $15 million (combined seed and Series A rounds), from investors like Greylock Partners, GV, and In-Q-Tel.

We were motivated by this mission of not just publishing more papers on some of these fun algorithmic or theoretical ideas, but actually making AI more broadly practical with a new end to end platform that focuses centrally on the problem of data labeling, Ratner, who serves as the companys CEO, says.

Snorkel AI's platform

The companys flagship product is the end-to-end Machine Learning platform called Snorkel Flow, which allows for AI applications to be deployed programmatically at much faster rates.

Snorkel Flow would serve as a replacement of armies of human labelers which at the moment do it by hand. An example of those manual processes include training AI applications to assist a radiologist in triaging chest X-rays. The radiologist would have to sit through a ton of images labeling which ones are emergency and which ones arent to teach the AI algorithm. Another example is a bank wanting AI to classify, sort and pull information out of someones loan portfolio. The companies would need to have their legal team check and label thousands of documents by hand every single time they want to change something.

Our key focus has been on sectors where labeling data by hand is not just a slower or more expensive option, but is often just a non-starter, Ratner says.

According to Ratner, this is usually due to one or more of three factors: the data is private so companies cant outsource it to get it labeled outside of the organization, the data requires in-demand experts (doctors or legal analysts), and the data changes frequently so companies find themselves labeling and relabeling all the time.

Snorkel AIs platform enables a programmatic approach so that instead of labeling one document at a time, the user can write a function (for example if they see the word employment in the header, they can label it as an employment contract).

The advantage is that writing a dozen or two dozen of these labeling functions to label your AI solution is orders of magnitude faster than labeling documents by hand, Ratner says.

Snorkel AI label

The Palo Alto-based Snorkel AI which counts around two dozen employees, raised a $3 million seed round in June of last year, and a $12 million Series A in October. The companys current customers include two top US banks, government agencies and other Fortune 500 companies.

Saam Motamedi, an Under 30 honoree and a general partner at Greylock Partners which co-led the seed round and led the Series A, says that Greylock immediately wanted to partner with the Snorkel AI cofounders given the caliber of the team, the traction of the open source Snorkel project and the power of the paradigm shift they are pioneering around this data-centric approach.

Customers have been able to go from what took months to deploy AI applications to now being able to deploy it in hours because they can programmatically manage the data, Motamedi says.

Continued here:

Meet The Stanford AI Lab Alums That Raised $15 Million To Optimize Machine Learning - Forbes