Qifan Song

Knowledge Distillation Detection for Open-weights Models

LLM Safety Alignment is Divergence Estimation in Disguise

Efficient Bayesian DNN Compression through Sparse Quantized Sub-distributions