Yi-Zhe Chen

Transformers trained with ABNAT learn to mimic clean attention maps, achieving up to 85.77% less attention deviation and 11.28% higher accuracy on noisy analog hardware.

Turning published scientific figures back into executable Python remains a brittle frontier for top VLMs, with even frontier models routinely failing on multi-component layouts, precise axis semantics, and domain typography.

Early denoising stages in diffusion models are up to 2.58× more vulnerable to hardware noise, but a novel approach can mitigate this without retraining.

MatRank's innovative approach to pseudo-labeling reduces prediction error by effectively utilizing unlabeled data, achieving a remarkable NMAE of 0.896.

Accepting selective mismatches in autoregressive decoding can boost throughput by over 15% without any additional training or model adjustments.

ReTopK achieves a 3.07x speedup in attention computation with only a 0.50% increase in perplexity, revolutionizing long-context processing efficiency.