Researcher, Google
1 paper at NeurIPS 2025
We prove that under appropriate conditions, a single-head softmax attention mechanism exhibits benign overfitting