20250926 Neurips 4papers

[NeurIPS'25] Four papers accepted to NeurIPS 2025 in the field of efficient and reliable AI. Congrats to our students and collaborators! Two papers from our lab:

  • HoliTom: A top-performing video LLM token compression method that maintains 99.1% performance while reducing FLOPs to just 6.9%—and it’s training-free! [arxiv] [code] [webpage]

  • FreqExit: A dynamic inference framework for visual autoregressive (VAR) models via early exit with novel frequency-aware guidance. [openreview] [code] [webpage]