Artificial Intelligence #121

Hey, in this issue: a new (and awesome!) text-to-image synthesis model from Google; accelerated PyTorch training on Mac; a backdoor problem of machine learning; a tutorial on automatic differentiation; lessons from deploying deep learning to production; applying BERT to speech data; and more.

This issue is sponsored by Superb AI and Snorkel AI.

• [Google] Imagen, a new text-to-image synthesis model that can generate photorealistic images from text
• [MIT] Hydra: a real-time spatial perception engine for 3D scene graph construction and optimization
• Introducing accelerated PyTorch training on Mac
• Machine learning has a backdoor problem
• [Sponsored] How to Prioritize Data Quality for Computer Vision

Data is the highest contributor to model performance for neural network architectures. However, dataset creation is one of the most costly and demanding components of the entire computation pipeline. Learn about data quality workflows and how to implement best practices.

• [Tutorial] Automatic differentiation: forward and reverse
• [The Gradient] Lessons from deploying deep learning to production
• [Sponsored] The future of data-centric AI, Snorkel AI's free virtual event, is back! Register today>
• [Project] A lightweight yet powerful audio-to-MIDI converter with pitch bend detection
• [The Gradient] An illustrated tour of applying BERT to speech data

The newsletter is now read by more than 580,000 subscribers. If you are building an AI or a data product or service, you are invited to join the list of these great companies by becoming a sponsor of one of the future newsletter issues. Feel free to reach out to [email protected] for more details on sponsorships.

Enjoy the newsletter? Help us make it bigger and better by sharing it with your colleagues and friends.

Have a nice weekend! See you next week. — Andriy