Gal Vardi

Positive Distribution Shift as a Framework for Understanding Tractable Learning

ImpMIA: Leveraging Implicit Bias for Membership Inference Attack under Realistic Scenarios

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

Temperature is All You Need for Generalization in Langevin Dynamics and other Markov Processes

Benign Overfitting in Single-Head Attention

Flavors of Margin: Implicit Bias of Steepest Descent in Homogeneous Neural Networks

Transformers are almost optimal metalearners for linear classification

Approaching Deep Learning through the Spectral Dynamics of Weights

Provable Privacy Attacks on Trained Shallow Neural Networks

Trained Transformer Classifiers Generalize and Exhibit Benign Overfitting In-Context

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

Benign Overfitting in Single-Head Attention