Chenxiang Zhang

How does the optimizer implicitly bias the model merging loss landscape?

Training with Dynamic Sparse Heads as the Key to Effective Ensembling

Spurious Privacy Leakage in Neural Networks