Large language models (LLMs) have been widely studied in areas such as mathematical reasoning, complex coding, and scientific problem solving. However, their ability to generate executable code that visually depicts physical scenarios and their qualitative dynamics remains underexplored. We propose SimuScene, the first systematic study that trains and evaluates LLMs on code generation for physics-inspired animations across 52 concepts spanning five physics domains. We build an automated data collection pipeline with human verification to ensure data quality. The resulting dataset contains 7,659 scenarios, including a 334-example human-verified test set. We evaluate 10 contemporary LLMs and find that even the strongest model achieves only a 21.5\% Avg@8 accuracy, demonstrating the difficulty of generating animations that are both executable and visually aligned with physical scenario descriptions. Finally, we introduce a reinforcement learning pipeline that uses visual rewards from code-generated videos to train text-only LLMs, with a vision-language model evaluating videos through verification questions. Experiments show that training with our data and video-based rewards improves LLM performance on physics-inspired animation generation.
Is Bio-Inspired Learning Better than Backprop? Benchmarking Bio Learning vs. Backprop
Institute for Infocomm Research (I2R), A*STAR(信息通信研究院(I2R),新加坡科技研究局(A*STAR));School of Mechanical and Aerospace Engineering, Nanyang Technological University (NTU)(南洋理工大学机械与宇航工程学院);Faculty of Science, National University of Singapore (NUS)(新加坡国立大学理学院);Centre for Frontier AI Research (CFAR), A*STAR(新加坡科技研究局前沿人工智能研究中心(CFAR))
U2-BENCH: Benchmarking Large Vision-Language Models on Ultrasound Understanding
Fudan University(复旦大学);University of Oxford(牛津大学);Dolphin AI;Beihang University(北京航空航天大学);Hong Kong Baptist University(香港浸会大学);University of Cambridge(剑桥大学);The Chinese University of Hong Kong(香港中文大学);Esonic Imaging;Technical University of Denmark(丹麦技术大学)
Benchmarking Relief-Based Feature Selection Methods for Bioinformatics Data Mining
University of Pennsylvania(宾夕法尼亚大学);Ursinus College(乌尔辛纳斯学院)
现代生物医学数据采掘要求特征选择方法能够(1)应用于大规模特征空间(例如“组学”数据),(2)在噪声问题中发挥作用,(3)检测复杂的关联模式(例如基因-基因相互作用),(4)灵活地适应各种问题域和数据类型(例如,遗传变体、基因表达和临床数据)和(5)在计算上是易处理的。为此,这项工作研究了一套过滤器式的特征选择算法的启发“救济”算法,即救济为基础的算法(RBAs)。我们在一个名为ReBATE(Relief-Based Algorithm Training Environment)的开源框架中实现并扩展了这些RBA。我们应用一个全面的遗传模拟研究比较现有的RBA,建议RBA称为MultiSURF,和其他建立特征选择方法,在各种问题。本研究的结果(1)支持这样的断言,即RBA是特别灵活、有效和强大的特征选择方法,其区分具有单变量、多变量、上位性或异质关联的相关特征,(2)证实了分类与回归、离散与连续特征、缺失数据、多个类别或类别不平衡的扩展的功效,(3)识别特定RBA之前未知的局限性,(4)表明虽然MultiSURF* 在明确识别纯双向交互方面表现最佳,但MultiSURF在各种问题类型中产生最可靠的特征选择性能。