AI/ML Developers Weekly #2

Hey Everyone, Happy Friday! in this issue, we will discuss latest AI research on OOD detection and robots automation, tutorial on GPT2 with Pytorch, and ML system design.

[?? Breakthrough]?VOS: Learning What You Don't Know by Virtual Outlier Synthesis

This paper presents a novel framework for OOD detection. "The nice part is that it's architectural change of the detection network, with a new contrastive loss which does not introduce additional hyper-parameters. No additional data required." Martin Gorner twittered. 15 mins read.

[?? Breakthrough] Can Robots Follow Instructions for New Tasks?

Google AI presents "BC-Z: Zero-Shot Task Generalization with Robotic Imitation Learning" that studies how robots can generalize to new tasks that they were not trained to do. 10 mins read.

[????Events]?MLCon2.0: AI and ML Developers Conference (Feb. 22~23)

Learn from 35+ AI experts from DeepMind, Spotify, Twitter, HuggingFace, Instacart, LinkedIn, Pinterest, Mobileye, AstraZeneca, and more in sessions about building real-world AI applications.

[???? Events] Subsurface LIVE Winter 2022: The Cloud Data Lake Conference (Mar 2~3)

A two-days virtual ML, Cloud and Data Lake community event, presented by Dremio, AWS, and Microsoft Azure.

[?? Learning Blog]?Fast Product Iteration with Offline Replay Experiment

Maxine Qian (Senior Data Scientist at Pinterest) spoke the topic at AICamp AI/ML Seminar Series. She wrote this blog to share on the offline experimentation framework. The framework allows to select optimal candidates in a matter of hours instead of days required for online experiments. It can forecast the output of key metrics offline through a replay approach on past data, and enables quickly iterate product design. 15 mins read.

[????Learning Video] Tutorial: Text Classification using GPT2 and Pytorch

In this tutorial, George Mihaila walks you through on how to use GPT2 from HuggingFace for text classification. We start with downloading customized dataset, installing required components, selecting pre-trained models, and then train the model. We finally evaluate the results and how to optimize further. 90 mins watch.

[?? Learning Blog]?ML System Design: Data Distribution Shifts and Monitoring

This 13,000-word lecture note, created by Chip Huyen, discussed common problems in developing and designing ML systems in production, such as data distribution shifts, monitoring, and causes of ML failures. and how to solve them.

Thanks for reading. Have a great weekend.