Gilad Yehudai

Provable Unlearning with Gradient Ascent on Two-Layer ReLU Neural Networks

Depth-Width Tradeoffs for Transformers on Graph Tasks

Emergence of Linear Truth Encodings in Language Models

When Can Transformers Count to n?

Compositional Reasoning with Transformers, RNNs, and Chain of Thought

Quality over Quantity in Attention Layers: When Adding More Heads Hurts

Reconstructing Training Data From Real-World Models Trained with Transfer Learning