AutoKaggle: A Multi-Agent Framework for Autonomous Data Science Competitions

Ziming Li, Qianbo Zang, David Ma, Jiawei Guo, Tuney Zheng, Minghao Liu, Xinyao Niu, Yue Wang, Jian Yang, Jiaheng Liu, Wanjun Zhong, Wangchunshu Zhou, Wenhao Huang, Ge Zhang

Project |

• Existing automated data science systems are oversimplified, inflexible, and lack transparency, making them inadequate for complex real-world data science tasks.
• We develop AutoKaggle, an end-to-end multi-agent framework that integrates phase-based workflows, iterative debugging, and comprehensive reporting systems to provide data scientists with efficient, reliable, and transparent automated solutions.

CodeEditorBench: Evaluating Code Editing Capability of Large Language Models

Jiawei Guo, Ziming Li, Xueling Liu, Kaijing Ma, Tianyu Zheng, Zhouliang Yu, Ding Pan, Yizhi LI, Ruibo Liu, Yue Wang, Shuyue Guo, Xingwei Qu, Xiang Yue, Ge Zhang, Wenhu Chen, Jie Fu

Project |

• Existing frameworks focus heavily on code generation, lacking systematic evaluation of code editing abilities. Code editing is crucial in software development, requiring dedicated evaluation benchmarks.
• We establish CodeEditorBench as the first comprehensive framework for evaluating LLMs’ code editing capabilities and advance code editing technology through open-source datasets and evaluation tools.

MedAgents: Large Language Models as Collaborators for Zero-shot Medical Reasoning

Xiangru Tang, Anni Zou, Zhuosheng Zhang, Ziming Li, Yilun Zhao, Xingyao Zhang, Arman Cohan, Mark Gerstein

Project |

• Current LLMs face challenges in medical domains due to limited domain knowledge and reasoning capabilities, necessitating a solution that doesn’t require additional training.
• We design a framework that simulates multi-disciplinary expert collaboration through role-playing and multi-round discussions to activate LLMs’ latent medical knowledge and enhance their reasoning abilities.