CORA: Per-Slice Coherent Orthogonal Rotation for SVD-based Low-Rank Adaptation

Collection of recent arXiv submissions (stat.ML) covering advances in parameter-efficient fine-tuning (CORA: per-slice coherent orthogonal rotations for SVD-based low-rank adaptation), theory and generalization of generative models (proof that benign overfitting/double descent do not occur in diffusion models), robust multi-task/federated learning under contamination (minimax rates and a filtering-based robust gradient method), reliable selection for CATE predictions via Denoised Conformal Alignment (FDR control with variance denoising), a formal hierarchy of policy-learning problems, manifold