Tianyu Zhang

Shi Qiu, Shaoyang Guo, Zhuo-Yang Song et al.

Current benchmarks for evaluating the reasoning capabilities of Large Language Models (LLMs) face significant limitations: task oversimplification, data contamination, and flawed evaluation items. These deficiencies necessitate more rigorous assessment methods. To address these limitations, we introduce PHYBench, a benchmark of 500 original physics problems ranging from high school to Physics Olympiad difficulty. PHYBench addresses data contamination through original content and employs a systematic curation pipeline to eliminate flawed items. Evaluations show that PHYBench activates more tokens and provides stronger differentiation between reasoning models compared to other baselines like AIME 2024, OlympiadBench and GPQA. Even the best-performing model, Gemini 2.5 Pro, achieves only 36.9% accuracy compared to human experts' 61.9%. To further enhance evaluation precision, we introduce the Expression Edit Distance (EED) Score for mathematical expression assessment, which improves sample efficiency by 204% over binary scoring. Moreover, PHYBench effectively elicits multi-step and multi-condition reasoning, providing a platform for examining models' reasoning robustness, preferences, and deficiencies. The benchmark results and dataset are publicly available at https://www.phybench.cn/.

Tianyu Zhang, Suyuchen Wang, Chao Wang et al. · mila

Vision-language models (VLMs) benefit from multiple vision encoders, but naively stacking them yields diminishing returns while multiplying inference costs. We propose SCOPE, a Mixture-of-Encoders (MoEnc) framework that dynamically selects one specialized encoder per image-text pair via instance-level routing, unlike token-level routing in traditional MoE. SCOPE maintains a shared encoder and a pool of routed encoders. A lightweight router uses cross-attention between text prompts and shared visual features to select the optimal encoder from the routed encoders. To train this router, we introduce dual entropy regularization with auxiliary losses to balance dataset-level load distribution with instance-level routing confidence. Remarkably, SCOPE with one shared plus one routed encoder outperforms models using all four extra encoders simultaneously, while reducing compute by 24-49\%. This demonstrates that intelligent encoder selection beats brute-force aggregation, challenging the prevailing paradigm in multi-encoder VLMs.

Daniel Chin, Yian Zhang, Tianyu Zhang et al.

Learning to play an instrument is intrinsically multimodal, and we have seen a trend of applying visual and haptic feedback in music games and computer-aided music tutoring systems. However, most current systems are still designed to master individual pieces of music; it is unclear how well the learned skills can be generalized to new pieces. We aim to explore this question. In this study, we contribute Interactive Rainbow Score, an interactive visual system to boost the learning of sight-playing, the general musical skill to read music and map the visual representations to performance motions. The key design of Interactive Rainbow Score is to associate pitches (and the corresponding motions) with colored notation and further strengthen such association via real-time interactions. Quantitative results show that the interactive feature on average increases the learning efficiency by 31.1%. Further analysis indicates that it is critical to apply the interaction in the early period of learning.