30% drop in O1-preview accuracy when Putnam problems are slightly variated

The new 220-terabyte dataset will help to evaluate neural networks for quantum chemistry

A new method to boost generative model training by up to 10 times will be useful for generating and processing images and text, as well as for solving reinforcement learning tasks…

SceneCraft: An LLM Agent for Synthesizing 3D Scenes as Blender Code

Chemical Language Models Have Problems with Chemistry: A Case Study on Molecule Captioning Task

Vision Transformers Need Registers

Turing Complete Transformers: Two Transformers Are More Powerful Than One

Word2Vec received 'strong reject' four times at ICLR2013

Transformer as a hippocampal memory consolidation model

Text Embeddings Reveal (Almost) as Much as Text

Using Language Models to Convert Between Natural Language and Game Commands

Large Language Models Are Human-Level Prompt Engineers

Symbolic Regression is NP-hard

Predictive Coding Approximates Backprop Along Arbitrary Computation Graphs

ImageNet-trained CNNs are biased towards texture (2019)

Deep Learning for Symbolic Mathematics

Recurrent Independent Mechanisms

Learning the Arrow of Time

Learning to Represent Edits

Approximating CNNs with Bag-of-Local-Features Models

Learning to Design RNA

Winner’s Curse? On pace, progress, and empirical rigor

A computational model of the Moth Olfactory Network learns to read MNIST

Neural Architecture Search with Reinforcement Learning

HolStep: A Machine Learning Dataset for Higher-order Logic Theorem Proving

A Differentiable Physics Engine for Deep Learning in Robotics

Outrageously large neural networks: the sparsely-gated mixture-of-experts layer

Deep Learning research from the University of Oxford and Google DeepMind can accurately deduce sentences from visual analysis of - LipNet achieves 93.4% accuracy