MICCAI 2026 poster: MIS-Ground tests 3D anatomical spatial grounding across CT, MRI, views, coordinates, prompts, and terminology. MIS-SemSam changes the token-selection rule and raises Qwen3-VL-32B accuracy from 53.4% to 66.5%.
A geometry-driven mixed-precision KV-cache quantization poster showing that keys carry more information than values, so key-favored bit allocation preserves accuracy while reducing memory.
Ranking reasoning LLMs under repeated sampling, comparing 72 ranking methods across four Olympiad-style math benchmarks and packaging them in Scorio.
Proposed a Bayesian framework that estimates models' success probabilities with quantified uncertainty, yielding more reliable rankings and enabling categorical evaluation of LLMs.
NeurIPS 2025 poster on DFloat11: a lossless compression framework that shrinks LLMs and diffusion transformers to ~70% of their size with bit-for-bit identical outputs, plus a GPU kernel that decompresses on the fly.
