Snorkel AI

06/01/2022

Building NLP techniques for financial services is challenging, costly, and labor-intensive. Join Machine Learning Engineer, Aarti Bagul, and learn how the world's largest banks are using Snorkel Flow's data-centric AI platform to build highly-accurate NLP applications in a matter of days. Save your spot ↓

Learn how the world’s largest banks are using Snorkel Flow’s data-centric AI platform to build highly accurate NLP applications in a matter of days.

05/27/2022

In this post, Josh McGrath presents a primer on active learning. Machine learning systems are created by better training sets, not just new algorithms, yet curating a dataset is more labor-intensive than creating a new model. ↓

Active learning is a human-in-the-loop framework for iteratively building training sets and models. The goal is to decrease the overall cost of labeling by using a model to choose informative examples to label.

Josh discusses the design decisions machine learning engineers
must make when building an active learning pipeline, and how this technique compares to and can interact with other methods.

Experimental results show that active learning is a reasonable baseline for marginal cost savings over random sampling. However, in studies, it doesn’t achieve the same quality or efficiency gains as other methods such as data augmentation, self-supervision, and weak supervision.

A primer on active learning. What is active learning, how it works, its methods, and how it compares to other machine learning techniques

05/26/2022

In this episode of ScienceTalks, Braden Hancock talks to Thomas Wolf, Chief Science Officer at Hugging Face. Thomas discusses important design decisions behind the widely adopted Transformers library. ↓

To solve critical technical problems around large datasets, we need to use techniques like model compression and quantization, and we also need to share and trace such datasets.

Thomas explains that the first barrier to using machine learning in industry is the fear of novelty, and the second barrier is that good performance on academic datasets does not correlate to good applicability to industry use cases.

Hugging Face Chief Science Officer, Thomas Wolf shares his story about how he got into machine learning and discusses important design decisions behind the widely adopted Transformers library, as well as the challenges of bringing research projects into production.

05/26/2022

Managing bias in ML is not just a technical problem. It’s a social-technical problem. The right solution needs to be very carefully identified, because it is very easy to end up solving the wrong problem. Learn more about how ↓

Panel discussion on ethical AI, with Swati Gupta, Thomas Sasalsa, Sakshi Jain, Skip McCormick, & Alexis Zumwalt discussing ethical AI systems

05/24/2022

Government agencies are experiencing rapid adoption of AI technologies, making it challenging to keep up. Snorkel AI hosted a virtual event on April 21 to explore the challenges of and provide solutions using Snorkel Flow ↓

Gregory Ihrie, CTO of the , discussed how recent events have accelerated the need for auditable machine learning models with traceable lineage. The FBI applies legal and ethical controls to ensure that AI applications operate within the constitution, law, and policy.

Swati Gupta, Thomas Sasala, Sakshi Jain, Skip McCormick, and Alexis Zumwalt discussed how a roadmap can streamline the development and deployment of trustworthy AI for the government.

In his presentation, Alex Ratner explained how high-quality AI can be developed and deployed for the government quickly and efficiently with a data-centric roadmap using Snorkel Flow.

Braden Hancock discussed the benefits of programmatic labeling for trustworthy AI and how Snorkel Flow enables easy training data auditing, systematically correcting bias, ensuring supply train integrity, and more.

Learn how to unblock and adopt trustworthy AI for government with academia and industry experts in this insightful post on best practices of...

05/23/2022

[Last chance] If you have been waiting to see how Snorkel Flow automates data labeling while keeping humans in the loop, join us tomorrow at 11.00 AM PT for a live Snorkel Flow demo. →

Join a live demonstration of Snorkel Flow, the data-centric AI development platform used by Fortune 500 enterprises and government agencies to accelerate their AI development by 10-100x. Snorkel Flow can be used to classify and extract information from unstructured text like documents and social med...

05/23/2022

Aortic valve disease is the most common congenital malformation of the heart, and it is associated with poor health outcomes. In this paper, Snorkeler Jason Fries and team discuss how machine learning models could help identify aortic valve malformations from cardiac MRIs → https://snkl.ai/wsav

Obtaining labeled training data for machine learning models is a practical roadblock to building deep learning models for use in medicine. The UK Biobank dataset of >500,000 individuals with comprehensive medical record data prior to enrollment, long-term follow-up data, and medical imaging presents an attractive candidate for use with deep learning.

However, most of the imaging data is unlabeled. The team uses weak supervision to train a deep learning model to classify aortic valve malformations using unlabeled cardiac MRI sequences. Their model identified individuals with a 1.8-fold increase in the risk of a major adverse cardiac event.

05/17/2022

Interested in becoming the newest Snorkeler? We have open roles across several teams. Help us accelerate for the enterprise, come and join one of the most talented, passionate, and supportive teams in tech! ↓ https://snorkel.ai/careers

Join us to make your career-defining move. Search for open positions.