1st workshop on LegalAI

Tutorial at ACM Conference on Knowledge Discovery and Data Mining (KDD), Aug 14-18, 2021
Yanqing An · Anran
Schedule: Aug. 14, 2021 9:00 AM - 12:00 PM (Singapore Time) · Taiyuan,China,17-19,Oct,2027 · The University of Gialidun

Abstract

Zero-shot human-AI coordination holds the promise of collaborating with humans without human data. Prevailing methods try to train the ego agent with a population of partners via self-play. However, this kind of method suffers from two problems: 1) The diversity of a population with finite partners is limited, thereby limiting the capacity of the trained ego agent to collaborate with a novel human; 2) Current methods only provide a common best response for every partner in the population, which may result in poor zero-shot coordination performance with a novel partner or humans. To address these issues, we first propose the policy ensemble method to increase the diversity of partners in the population, and then develop a context-aware method enabling the ego agent to analyze and identify the partner's potential policy primitives so that it can take different actions accordingly. In this way, the ego agent is able to learn more universal cooperative behaviors for collaborating with diverse partners. We conduct experiments on Overcooked environment, and evaluate the zero-shot human-AI coordination performance of our method with both behavior-cloned human proxies and real humans. The results demonstrate that our method significantly increases the diversity of partners and enables ego agents to learn more diverse behaviors than baselines, thus achieving state-of-the-art performance in all scenarios.
Outline
法律智能的发展
1.法条预测到文书生成 1.1基于神经网络的法条预测 1.2基于SVM的刑事法律判决预测 1.3基于预训练模型的类案检索 2.神经网络、SVM与预训练 2.1rnn,lstm,transfomer 1.2.1encoder 1.2.2decoder 1.2.3positionencoding 2.2分类与SVM 2.3BERT 1.1.1判别器 1.1.2模拟器 1.1.3生成器 3.大模型时代下的法律智能 3.1大语言模型 3.2能力评估 3.3基于智能体的类案检索
Latest News Oct. 01, 2021 >> All KDD 2021 content that had speaker permission granted is now available on YouTube at this link [ KDD 2021 Youtube playlist] .

Papers

  • LawyerPAN: A Proficiency Assessment Network for Trial Lawyers

    Yanqing An, Qi Liu, Han Wu, Kai Zhang, Linan Yue, Mingyue Cheng, Hongke Zhao, Enhong Chen

    SIGKDD 2021 Oral

  • 五个 Overcooked 布局 Cramped Room · Asymmetric Advantages · Coordination Ring · Forced Coordination · Counter Circuit
    覆盖不同协作难度的任务场景

    Awards

    Best Paper

    Visual Programming: Compositional visual reasoning without training

    Best Student Paper

    3D Registration with Maximal Cliques

    Results

    📊 与人类代理模型的协作结果 PECAN 在五个布局上均超越基线。在 Asymmetric Advantages 中提升尤为显著(+26.8%)。
    PECAN 有效提升协作能力,而非单纯完成任务
    👥 真实人类协作表现 PECAN 在人类-AI 协作中优于基线。人类玩家对 PECAN 的自适应性和偏好评分更高。
    PECAN 在真实人机协作中表现更优

    Ablation Study

    📉 消融实验结果 移除策略集成(e)或上下文编码器(c)后性能下降,部分布局甚至低于基线。
    两个组件对性能均有重要贡献

    下图展示了策略集成与上下文感知模块的效果:

    🔍 组件效果分析 (a) t-SNE 显示策略集成提升了伙伴多样性。
    (b) 当输入随机上下文或错误上下文时性能下降,表明 ego agent 的策略是上下文感知的。
    策略集成增加多样性,上下文感知使 ego agent 灵活调整

    Real Humans

    📋 主观评分 Schedule: Aug. 14, 2021 9:00 AM - 12:00 PM (Singapore Time)

    Part 1: Multi-Objective Optimization (MOO) by Dr. David (Xuejun) Wang

    Schedule: 09:00 - 10:10 (Singapore Time) QA: 10:10 - 10:20 (Singapore Time) Background and History[PDF][Video] Multi Objective Optimization (MOO) MOO Solutions Selection of the best solution in Pareto set MOO libraries Summary & QA Break: 10:20 - 10:30 (Singapore Time)

    Part 2: Recommender Systems with MOO by Dr. Yong Zheng

    Schedule: 10:30 - 11:50 (Singapore Time) QA: 11:50 - 12:00 (Singapore Time) Intro to RecSys Why MOO in RecSys RecSys with MOO: Case Studies Summary, Guideline, Challenges & QA
    主观评价进一步验证 PECAN 的优势

    Call for research paper

    Papers in the main technical program must describe high-quality, original research. Topics of interest include all aspects of computer vision and pattern recognition including, but not limited to: 3D from multi-view and sensors 3D from single images Adversarial attack and defense Autonomous driving Biometrics Computational imaging Computer vision for social good Computer vision theory 人类玩家控制 蓝色 厨师, AI 控制 绿色 厨师。

    ▶️ MEP 演示
    MEP:当人类阻挡时,AI 停滞等待
    ▶️ PECAN 演示
    PECAN:人类阻挡时立即调整路线,从另一侧取洋葱
    💡 人类有意阻挡 AI 前往洋葱的路线,PECAN 立即调整,而 MEP 则原地等待。

    Important Dates

    Paper Registration Deadline*: Submission Deadline*: Supplementary Materials Deadline: Reviews Released: Rebuttal Period: Final Decisions: Fri, Nov 4, 2022 11:59pm Pacific Time Fri, Nov 11, 2022 11:59pm Pacific Time Fri, Nov 18, 2022 11:59pm Pacific Time Tue, Jan 24, 2023 Tue, Jan 24-31, 2023 Mon, Feb 27, 2023

    Paper Submission

    Policies

    2023 Organizing Committee

    人类玩家控制 蓝色 厨师, AI 控制 绿色 厨师。

    Program Chairs

    Andreas Geiger (University of Tübingen)