7 papers
WorldBench: A Challenging and Visually Diverse Multimodal Reasoning Benchmark
Yida Yin, Harish Krishnakumar, Chung Peng Lee +9
In real-world applications, models are expected to perform reliably across diverse settings. Yet, many existing multimodal benchmarks expand task types without capturing the visual…
Massive Spikes in LLMs are Bias Vectors: Mechanistic Uncovering and Spike-Free Quantization
Yung-Chin Chen, Chung Peng Lee, Ze-Wei Liou +1
Massive activation spikes in Large Language Models (LLMs) severely degrade quantization by stretching dynamic ranges. While prior hypotheses characterize these as high-level scalar…
The Lie We Tell: Correcting the Euclidean Fallacy in Vision Language Action Policies via Score Matching on Tangent Space
Bing-Cheng Chuang, I-Hsuan Chu, Bor-Jiun Lin +3
Diffusion-based Vision-Language-Action policies achieve remarkable success in robotic manipulation, yet commit a fundamental geometric error we term the $\textbf{Euclidean Fallacy}…
ESPADA: Execution Speedup via Semantics Aware Demonstration Data Downsampling for Imitation Learning
Byung-ju Kim, Jinu Pahk, Chungwoo Lee +6
Behavior-cloning based visuomotor policies enable precise manipulation but often inherit the slow, cautious tempo of human demonstrations, limiting practical deployment. However, p…
How Do Data Owners Say No? A Case Study of Data Consent Mechanisms in Web-Scraped Vision-Language AI Training Datasets
Chung Peng Lee, Rachel Hong, Harry H. Jiang +3
The internet has become the main source of data to train modern text-to-image or vision-language models, yet it is increasingly unclear whether web-scale data collection practices…
The Geometry of Alignment Collapse: When Fine-Tuning Breaks Safety
Max Springer, Chung Peng Lee, Blossom Metevier +5
Fine-tuning aligned language models on benign tasks unpredictably degrades safety guardrails, even when training data contains no harmful content and developers have no adversarial…