8月2日｜热门工作，智能体Agent工具使用及其高质量数据生成方法，APIGen，xLAM...

报告主题：AI智能体的工具使用及其高质量数据生成方法

报告日期：8月2日（周五）10:30-11:30

报告要点：

本次讲座将深入探讨论文《APIGen: Automated Pipeline for Generating Verifiable and Diverse Function-Calling Datasets》中提出的进展和方法。APIGen旨在通过生成高质量、多样化的数据集来增强AI智能体模型的函数调用和工具使用能力。该框架通过多阶段验证过程确保数据的可靠性和正确性，包括格式检查、实际函数执行和语义验证三个阶段。通过APIGen自动化采集的数据可以使得我们的xLAM（Large Action Model）以较少的参数取得和GPT-4相近的性能 https://apigen-pipeline.github.io/

在本次讲座中，我们将讨论：

APIGen的动机及其在增强大模型函数调用能力的重要性。APIGen方法的详细设计，包括数据生成和验证的每个阶段。使用APIGen数据集训练的xLAM模型的性能评估。未来的研究方向和APIGen框架的潜在改进。

This talk will delve into the advancements and methodologies presented in the paper "APIGen: Automated Pipeline for Generating Verifiable and Diverse Function-Calling Datasets." APIGen aims to enhance the function-calling and tool-use ability of LLM agent models by generating high-quality and diverse datasets. The framework improves the reliability and correctness of the data through a multi-stage verification process, including format checking, actual function execution, and semantic verification. The automated data collected through APIGen enables our xLAM (Large Action Model) to achieve performance comparable to GPT-4 with fewer parameters.

In this talk, we will discuss:

The motivation behind APIGen and its importance in enhancing the function-calling capabilities of large models.The detailed design of the APIGen methodology, including each stage of data generation and verification.Performance evaluation of xLAM models trained with APIGen datasets.Future research directions and potential improvements to the APIGen framework.

报告嘉宾：

刘祖欣是Salesforce AI Research的研究科学家，分别从卡内基梅隆大学和北京航空航天大学获得博士学位和学士学位（2019届北航本科生最高荣誉沈元奖章获得者）。他的研究方向是深度强化学习，机器人，和基于大模型的AI智能体技术。博士期间与谷歌、亚马逊云等研究机构紧密合作。他曾在ICML、NeurIPS、ICLR、CVPR、CoRL、ICRA、IROS等顶级会议发表多篇有影响力的工作。

Zuxin Liu is a Research Scientist at Salesforce AI Research. He received his PhD from Carnegie Mellon University and his bachelor's degree from Beihang University (recipient of the 2019 President's Award, the highest undergraduate honor at Beihang University). His research focuses on deep reinforcement learning, robotics, and LLM-based AI agent. During his PhD, he worked closely with research institutions such as Google and Amazon Web Services. He has published many influential papers in top-tier conferences such as ICML, NeurIPS, ICLR, CVPR, CoRL, ICRA, and IROS.

扫码报名

---

近期热门报告