Your Daily AI Research tl;dr — 2022–07–10 🧠

Mastering Code Generation, Predicting Models’ Generalization Abilities and a Special Investigation into Cruise

Welcome to your official daily AI research tl;dr (often with code and news) for AI enthusiasts where I share the most exciting papers I find daily, along with a one-liner summary to help you quickly determine if the article (and code) is worth investigating. I will also take this opportunity to share daily exciting news in the field.

Let’s get started with this iteration!

1️⃣ CodeRL: Mastering Code Generation through Pretrained Models and Deep Reinforcement Learning

A new framework for program synthesis (Code Generation) tasks through pretrained LMs and deep reinforcement learning (RL).

Link to the paper: https://arxiv.org/pdf/2207.01780.pdf

Code: https://github.com/salesforce/CodeRL

2️⃣ Neural Networks and the Chomsky Hierarchy

Really interesting paper by DeepMind: “In this work, we conduct an extensive empirical study (2200 models, 16 tasks) to investigate whether insights from the theory of computation can predict the limits of neural network generalization in practice.”

“We demonstrate that grouping tasks according to the Chomsky hierarchy allows us to forecast whether certain architectures will be able to generalize to out-of-distribution inputs.”

Link to the paper: https://arxiv.org/pdf/2207.02098.pdf

🆕 US safety regulators open special investigation into Cruise AV crash

The National Highway Traffic Safety Administration has opened a special investigation into a crash in San Francisco involving a Cruise autonomous vehicle that resulted in minor injuries. Given the circumstances below, it shows how much better autonomous driving vehicles need to be, compared to humans if we want to implement them in our lives. It doesn’t have to be better than us merely. It has to be near-perfection. This is why it is such a challenging field and will take a lot…