Data labeling is an important, though often time-consuming and complex, element of modern machine learning (ML) operations.
Data labeling can also be the key to unlocking the broader enterprise potential of foundation models. While foundation models such as GPT-3 and DALL-E have great utility for creating text and images, they often lack the context needed for specific enterprise use cases. To improve the foundation model, tuning and additional training are required, and this often requires labeled data.
But what if a foundation model could be used to jumpstart the data labeling process to make smaller models useful for specific enterprise use cases? That’s the challenge that data labeling vendor Snorkel AI is now claiming to help solve.
“It’s one thing if you’re trying to do creative work where you’re producing some copy text or some creative images, but there’s a big gap between that and a complex production use case where you have Needs to perform at a high precision bar. For very specific data and tasks,” Alex Ratner, CEO and co-founder of Snorkel AI told VentureBeat.
To help address this challenge, Snorkel AI today announced a preview of its new data-centric foundation model development capabilities. The goal is to help customers of the company’s Snorkel Flow platform adopt Foundation models for enterprise use cases. Ratner explained that Snorkel’s core research and ideas are about finding more efficient ways to label data to train or refine models.
Going with the flow to create a new foundation for enterprise AI
There are also other vendors trying to develop technology to help fix foundation models more easily. Among them is Nvidia, which announced its NeMo LLM (large language model) service in September.
One of the core components of the Nvidia service is the ability for customers to train large models for specific use cases with an approach called prompt learning. With Nvidia’s instant learning approach, a companion model is trained to provide context to a pre-trained LLM, using instant tokens.
Snorkel is also using Prompt with the Foundation Model Prompt Builder feature as part of its Enterprise Foundation Model Management Suite. However, Ratner emphasized that indicators are only one part of a larger set of tools needed to optimize foundation models for enterprise use cases.
Another tool Snorkel offers is a foundation model warm-start capability, which uses an existing foundation model to help provide data labeling.
“So basically, when you upload a dataset to be labeled in Snorkel Flow, you can now get a push-button type of first-pass auto-labeling using the power of Foundation Models,” Ratner said. said
Ratner noted that WarmStart isn’t the solution to all data labeling, but it would yield “low-hanging fruit.” He suggests that users will likely use WarmStart with the Quick Builder as well as Snorkel’s foundation model fine-tuning feature to improve models. The fine-tuning feature enables organizations to distill the foundation model into a domain-specific training set.
Generative vs. Predictive AI Enterprise Use Cases
Using foundation models for real enterprise use cases is the goal of Snorkel AI.
For better or worse, Ratner said individuals are probably more familiar with generative AI today, which uses the Foundation model. He distinguished generative models from predictive AI models that help predict outcomes and are commonly used by businesses today.
As an anecdote, Ratner said he was trying to develop some Snorkel AI logos using Stable Diffusion because, “.. it was so much fun.” He said he looked at about 30 samples and never found exactly what he wanted — an octopus wearing a snorkel underwater — which is the original corporate logo.
“I guess it’s a lot weirder than a random image, but I got some pretty cool logos after about 30 samples as a creative, creative human loop process,” Ratner said. “If you think about it from a predictive automation perspective, though, 30 attempts to get a successful result is a 3.3% hit rate and you never send anything with that bad result. can.”
One of Snorkel’s customers is online video advertising optimization vendor Pixability. Ratner explained that Pixability has millions of data points from YouTube videos that need to be classified for ML. Using the foundation model capabilities within Snorkel Flow, they are able to quickly complete the classification with accuracy levels above 90%.
A major U.S. bank that is a Snorkel customer was able to improve accuracy for extracting text from complex legal documents using the foundation model approach, Ratner said.
“We see this technology being applicable to a whole universe of applications where you can tag anything in text, PDF, image and video with high accuracy for some kind of predictive automation task. trying to categorize, extract or label,” Ratner said. . “We think it’s going to accelerate all of the use cases we currently support, as well as add new ones that weren’t possible before with our current methods, so we’re pretty excited. “