{"id":8024,"date":"2025-05-10T09:45:52","date_gmt":"2025-05-10T01:45:52","guid":{"rendered":"http:\/\/tsinghualogic.net\/JRC\/?page_id=8024"},"modified":"2026-04-13T20:07:00","modified_gmt":"2026-04-13T12:07:00","slug":"labevents","status":"publish","type":"page","link":"http:\/\/tsinghualogic.net\/JRC\/labevents\/","title":{"rendered":"Events"},"content":{"rendered":"\t\t<div data-elementor-type=\"wp-page\" data-elementor-id=\"8024\" class=\"elementor elementor-8024\" data-elementor-settings=\"[]\">\n\t\t\t\t\t\t\t<div class=\"elementor-section-wrap\">\n\t\t\t\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-a82615c elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"a82615c\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-a8bf9fd\" data-id=\"a8bf9fd\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap\">\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-bd2fa7b elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"bd2fa7b\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-a174308\" data-id=\"a174308\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-bdc1c2f elementor-widget elementor-widget-heading\" data-id=\"bdc1c2f\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>\/*! elementor - v3.5.6 - 28-02-2022 *\/\n.elementor-heading-title{padding:0;margin:0;line-height:1}.elementor-widget-heading .elementor-heading-title[class*=elementor-size-]>a{color:inherit;font-size:inherit;line-height:inherit}.elementor-widget-heading .elementor-heading-title.elementor-size-small{font-size:15px}.elementor-widget-heading .elementor-heading-title.elementor-size-medium{font-size:19px}.elementor-widget-heading .elementor-heading-title.elementor-size-large{font-size:29px}.elementor-widget-heading .elementor-heading-title.elementor-size-xl{font-size:39px}.elementor-widget-heading .elementor-heading-title.elementor-size-xxl{font-size:59px}<\/style><h4 class=\"elementor-heading-title elementor-size-default\">[Advances in Logic and Artificial Intelligence] 26th March, 2026:<\/h4>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-ce015d0 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"ce015d0\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-8a8168d\" data-id=\"8a8168d\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-8d365dc elementor-widget elementor-widget-toggle\" data-id=\"8d365dc\" data-element_type=\"widget\" data-widget_type=\"toggle.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>\/*! elementor - v3.5.6 - 28-02-2022 *\/\n.elementor-toggle{text-align:left}.elementor-toggle .elementor-tab-title{font-weight:700;line-height:1;margin:0;padding:15px;border-bottom:1px solid #d4d4d4;cursor:pointer;outline:none}.elementor-toggle .elementor-tab-title .elementor-toggle-icon{display:inline-block;width:1em}.elementor-toggle .elementor-tab-title .elementor-toggle-icon svg{-webkit-margin-start:-5px;margin-inline-start:-5px;width:1em;height:1em}.elementor-toggle .elementor-tab-title .elementor-toggle-icon.elementor-toggle-icon-right{float:right;text-align:right}.elementor-toggle .elementor-tab-title .elementor-toggle-icon.elementor-toggle-icon-left{float:left;text-align:left}.elementor-toggle .elementor-tab-title .elementor-toggle-icon .elementor-toggle-icon-closed{display:block}.elementor-toggle .elementor-tab-title .elementor-toggle-icon .elementor-toggle-icon-opened{display:none}.elementor-toggle .elementor-tab-title.elementor-active{border-bottom:none}.elementor-toggle .elementor-tab-title.elementor-active .elementor-toggle-icon-closed{display:none}.elementor-toggle .elementor-tab-title.elementor-active .elementor-toggle-icon-opened{display:block}.elementor-toggle .elementor-tab-content{padding:15px;border-bottom:1px solid #d4d4d4;display:none}@media (max-width:767px){.elementor-toggle .elementor-tab-title{padding:12px}.elementor-toggle .elementor-tab-content{padding:12px 10px}}<\/style>\t\t<div class=\"elementor-toggle\" role=\"tablist\">\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-1481\" class=\"elementor-tab-title\" data-tab=\"1\" role=\"tab\" aria-controls=\"elementor-tab-content-1481\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><i class=\"fas fa-caret-right\"><\/i><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><i class=\"elementor-toggle-icon-opened fas fa-caret-up\"><\/i><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\" class=\"elementor-toggle-title\"> Resolution Chain-of-Thought for LLM Symbolic Reasoning, Yixiang Chen (East China Normal University)<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-1481\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"1\" role=\"tabpanel\" aria-labelledby=\"elementor-tab-title-1481\"><p>Resolution Chain-of-Thought for LLM Symbolic Reasoning<\/p><p><strong>Speaker<\/strong>: Yixiang Chen (East China Normal University)<\/p><p><strong>Time<\/strong>: 16:00-17:30, 26 March 2026<\/p><p><strong>Abstract<\/strong>:<\/p><section><span data-mpa-action-id=\"mfkkkti5hjp\" data-pm-slice=\"0 0 []\">Large language models still struggle with complex logical reasoning. Numerous studies have explored ways to strengthen their inference skills, broadly grouped into solver-based, prompt-based, and fine-tuning approaches. Among these, prompting techniques improve LLMs by explicitly modeling reasoning chains like Chain-of-Thought (CoT), Tree-of-Thought (ToT), by acquiring symbolic expressions such as SymbCoT and by adaptive selection of Symbolic Languages (SL).<\/span><\/section><section><\/section><section><span data-mpa-action-id=\"mfkkkti5hjp\" data-pm-slice=\"0 0 []\">Building on this line of work, we introduce bidirectional reasoning into the improved method and implement an automated reasoning process based on the generation of large language models through the design of prompt words. Technically, Bi-Resolution first converts the natural language problem into first-order logic formulation, and selects the corresponding version of resolution algorithm. During resolution, bidirectional reasoning guides constraint instantiation to prune redundant clauses and reduce complexity. Bi-Resolution enables the model to judge statements that are \u201cneither fully true nor fully false\u201d more accurately. Experimental results show that our method successfully improves the logical inference accuracy of large language models.<\/span><\/section><p>=====<\/p><p style=\"font-weight: 400;\">Speaker Bio: Professor Yixiang Chen, a professor at the School of Software Engineering, East China Normal University, currently serves as the first Chair of the Artificial Intelligence Logic Committee of the Chinese Association for Artificial Intelligence, the first Chair of the Trusted Intelligent Systems Committee of the Shanghai Association for Artificial Intelligence. He is engaged in foundational and engineering research on the trustworthiness of artificial intelligence. He has established the spatio-temporally consistent intelligent system specification language STeC and its hybrid clock logic system, designed technical methods for the optimized hardware and software design of intelligent systems, and developed multidimensional attribute-based software trustworthiness measurement, evaluation methods, and enhancement specifications.<\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-d6297b9 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"d6297b9\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-4d478a8\" data-id=\"4d478a8\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-644badd elementor-widget elementor-widget-heading\" data-id=\"644badd\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h4 class=\"elementor-heading-title elementor-size-default\">[Advances in Logic and Artificial Intelligence, lectures] 26th February, 6th March, 13th March, 2026:<\/h4>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-b8e80b6 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"b8e80b6\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-d18bc32\" data-id=\"d18bc32\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-a883ea9 elementor-widget elementor-widget-toggle\" data-id=\"a883ea9\" data-element_type=\"widget\" data-widget_type=\"toggle.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-toggle\" role=\"tablist\">\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-1761\" class=\"elementor-tab-title\" data-tab=\"1\" role=\"tab\" aria-controls=\"elementor-tab-content-1761\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><i class=\"fas fa-caret-right\"><\/i><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><i class=\"elementor-toggle-icon-opened fas fa-caret-up\"><\/i><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\" class=\"elementor-toggle-title\">Probabilistic Causal Models, Algorithms for Causal Learning, and Foundations of Causal Learning | Hanti Lin (UC Davis)<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-1761\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"1\" role=\"tabpanel\" aria-labelledby=\"elementor-tab-title-1761\"><p><strong><em>Probabilistic Causal Models<\/em><\/strong><\/p><p><strong>Speaker<\/strong>: Hanti Lin<\/p><p><strong>Time<\/strong>: 9:50-12:15, 27 February 2026<\/p><p><strong><em>Algorithms for Causal Learning<\/em><\/strong><\/p><p><strong>Speaker<\/strong>: Hanti Lin<\/p><p><strong>Time<\/strong>: 9:50-12:15, 6 March 2026<\/p><p><strong><em>Probabilistic Causal Models<\/em><\/strong><\/p><p><strong>Speaker<\/strong>: Hanti Lin<\/p><p><strong>Time<\/strong>: 9:50-12:15, 13 March 2026<\/p><p>=====<\/p><p style=\"font-weight: 400;\">Speaker Bio: Hanti Lin\u00a0is a philosopher of science and formal epistemologist, with papers published in philosophy as well as theoretical computer science. Before he joined UC Davis, he was a postdoc at the Australian National University.<\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-d2c74b5 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"d2c74b5\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-c0e42e1\" data-id=\"c0e42e1\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-58694b2 elementor-widget elementor-widget-heading\" data-id=\"58694b2\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h4 class=\"elementor-heading-title elementor-size-default\">[Advances in Logic and Artificial Intelligence] 18th September, 2025:<\/h4>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-e7e44a6 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"e7e44a6\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-e8a0578\" data-id=\"e8a0578\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-d3d5723 elementor-widget elementor-widget-toggle\" data-id=\"d3d5723\" data-element_type=\"widget\" data-widget_type=\"toggle.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-toggle\" role=\"tablist\">\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-2221\" class=\"elementor-tab-title\" data-tab=\"1\" role=\"tab\" aria-controls=\"elementor-tab-content-2221\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><i class=\"fas fa-caret-right\"><\/i><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><i class=\"elementor-toggle-icon-opened fas fa-caret-up\"><\/i><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\" class=\"elementor-toggle-title\"> Towards Logical and Causal Reasoning of Large Language Models, Haoxuan Li (Peking University)<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-2221\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"1\" role=\"tabpanel\" aria-labelledby=\"elementor-tab-title-2221\"><p><strong><em>Towards Logical and Causal Reasoning of Large Language Models<\/em><\/strong><\/p><p><strong>Speaker<\/strong>: Haoxuan Li (Peking University)<\/p><p><strong>Time<\/strong>: 16:00-17:30, 18 September 2025<\/p><p><strong>Abstract<\/strong>:<\/p><p style=\"font-weight: 400;\">Large language\u00a0models\u00a0(LLMs) have achieved remarkable successes in various natural language tasks, but still have significant limitations to their logical and causal reasoning abilities. In this talk, we first comprehensively introduce the most cutting-edge LLM logical reasoning approaches with a proposed new taxonomy. Specifically, to accurately answer complex logic questions, previous methods can be categorized based on reliance on external solvers, prompts, and fine-tuning. To avoid logical contradictions, we discuss concepts and solutions of various logical consistencies, including implication, negation, transitivity, factuality consistencies, and their composites. Secondly, we discuss the benefits of introducing causality into LLM reasoning, in which the\u00a0key insight is that correlation does not necessarily\u00a0imply causation. For example, there is high ice cream sales and crime rates in summer, but this does not indicate that ice cream sales have a causal influence\u00a0on crime rates.\u00a0We conclude that logical rules can be regarded as the causal invariance of LLM reasoning\u00a0based on\u00a0natural language examples.<\/p><p>=====<\/p><p style=\"font-weight: 400;\">Speaker Bio: Haoxuan Li is an assistant researcher at Peking University, also as research fellow at Tsinghua-UvA Joint Research Center for Logic and the University of Oxford. He graduated\u00a0from\u00a0the\u00a0experimental class for gifted children in Beijing No.8 Middle School, which enables him to pursue his PhD at the age of 19. His research interests include causal inference and logical reasoning of\u00a0large language\u00a0models, and has more than 50 publications as the first author or the corresponding\u00a0author appeared in top-tier CCF-A conferences, reported by MIT Technology Review and CAAI. Moreover, he is supported by the Young Scientists Fund of the National Natural Science Foundation of China (\u00a5300,000) and Young Elite Scientists Sponsorship Program by CAST &#8211; Doctoral Student Special Plan (via CCF). He has been selected as the 2024 Peking University Person of the Year and representative of National Scholarship reported by People&#8217;s Daily.<\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-704429b elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"704429b\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-ef54301\" data-id=\"ef54301\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-c375fac elementor-widget elementor-widget-heading\" data-id=\"c375fac\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h4 class=\"elementor-heading-title elementor-size-default\">[TALK] 18th May, 2025:<\/h4>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-5744446 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"5744446\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-bd03889\" data-id=\"bd03889\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-8e44203 elementor-widget elementor-widget-toggle\" data-id=\"8e44203\" data-element_type=\"widget\" data-widget_type=\"toggle.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-toggle\" role=\"tablist\">\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-1491\" class=\"elementor-tab-title\" data-tab=\"1\" role=\"tab\" aria-controls=\"elementor-tab-content-1491\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><i class=\"fas fa-caret-right\"><\/i><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><i class=\"elementor-toggle-icon-opened fas fa-caret-up\"><\/i><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\" class=\"elementor-toggle-title\"> Developing And Assessing Language Models For Logical Reasoning Over Natural Language, Qiming Bao (University of Auckland)<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-1491\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"1\" role=\"tabpanel\" aria-labelledby=\"elementor-tab-title-1491\"><p><strong><em>Developing And Assessing Language Models For Logical Reasoning Over Natural Language<\/em><\/strong><\/p><p><strong>Speaker<\/strong>: Qiming Bao (University of Auckland)<\/p><p><strong>Time<\/strong>: 10:00 AM, 18 May 2025<\/p><p><strong>Abstract<\/strong>: Recent advancements in AI have highlighted the importance of integrating deep learning with symbolic logic reasoning. Language models such as RoBERTa, DeBERTa, LLaMA, Alpaca, Vicuna, GPT-3.5, and GPT-4 have advanced the performance of AI systems in various natural language processing tasks to human-like levels. However, the generalization of language models in logical reasoning remains underexplored. One of the main reasons is the limitation posed by the lack of extensive, balanced, and real-world datasets for logical reasoning. This presentation has three research objectives, addressing the main research gap\/limitation:<br \/>To improve the models&#8217; out-of-distribution performance on multi-step logical reasoning tasks through logic-driven data augmentation.<br \/>To enhance the models&#8217; performance on real-world logical reasoning datasets by constructing an Abstract Meaning Representation based logic-driven data augmentation method.<br \/>Although large language models demonstrate impressive performance on current logical reasoning leaderboards, it remains underexplored whether they truly possess strong capabilities in logical reasoning.<br \/>The first part of the presentation focuses on improving language models&#8217; ability in multi-step logical reasoning, particularly when faced with unbalanced reasoning steps. Inspired by DeepLogic, we present IMA-GloVe-GA, an RNN-based model with a gate attention mechanism, developed to accommodate varying reasoning depths. This is facilitated by our PARARULE-Plus dataset, created for deeper reasoning tasks. Our results show notable enhancements in model performance under both standard and out-of-distribution conditions.<br \/>The second part of the presentation focuses on generating diverse training data to address the scarcity of real-world logical reasoning datasets and enhance large language models (LLMs) for logical reasoning tasks. We introduce AMR-LDA, a data augmentation method that converts text into Abstract Meaning Representation (AMR) graphs, improving reasoning datasets. This approach benefits various models, including GPT-3.5 and GPT-4, and improves performance, notably achieving the top rank on the ReClor leaderboard.<br \/>The third part of the presentation examines how Large Language Models (LLMs) like GPT-3.5 and GPT-4 respond to trivial changes in logical reasoning datasets. We created ReClor-plus, LogiQA-plus, and LogiQAv2-plus, which include shuffled options and modified correct choices to test LLMs&#8217; logical reasoning. Although LLMs excel on standard datasets, they exhibit degraded performance with these modified versions. Our findings reveal that incorporating task variations, perturbations in training sets, and logic-driven data augmentation significantly enhances LLMs&#8217; generalisation and robustness in logical reasoning scenarios.<br \/>This presentation explores several different approaches to demonstrate a more robust QA system that aids computers in thinking and reasoning over natural language texts through logical reasoning. Our methods have been evaluated and now lead the public logical reasoning leaderboard, ReClor. We are the first group in the world to have scored above 90% on the ReClor hidden test set.<\/p><p><strong>About the speaker<\/strong>: Qiming Bao is a Ph.D. graduated from the Strong AI Lab, NAOInstitute, University of Auckland, New Zealand, supervised by Professor Michael Witbrock and Associate Professor Jiamou Liu. His research interests include natural language processing and reasoning. He has over five years of research and development experience, and has published several papers in top conferences in the fields of AI\/NLP\/Reasoning, including ACL, AAAI, IJCAI, ICLR, EACL, LLM@IJCAI, AGI@ICLR and IJCLR-NeSy. His method named AMR-LDA (GPT-4 + AMR-LDA Prompt Augmentation) has achieved the #1 ranking on a one of the most challenged logical reasoning reading comprehension leaderboards (ReClor) and we are the first group scored above 90% on the hidden test set around the world. Two of his logical reasoning datasets called PARARULE-Plus and AbductionRules have been collected by LogiTorch, ReasoningNLP, Prompt4ReasoningPapers, OpenAI\/Evals, A Survey on Evaluation of Large Language Models and Reasoning Language Models: A Blueprint. Qiming has given public guest talks and academic visit at Microsoft Research Asia, Samsung AI Center Cambridge UK, IEEE Vehicular Technology Society, ZJU-NLP Group, Zhejiang University, The University of Melbourne, Institute of Automation, Chinese Academy of Sciences, Shenzhen MSU-BIT University, University of Massachusetts &#8211; Amherst and Penn State University on his main research topic, &#8220;Natural Language Processing and Reasoning&#8221;.<\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"<p>[Advances in Logic and Artificial Intelligence] 26th Ma [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":[],"_links":{"self":[{"href":"http:\/\/tsinghualogic.net\/JRC\/wp-json\/wp\/v2\/pages\/8024"}],"collection":[{"href":"http:\/\/tsinghualogic.net\/JRC\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"http:\/\/tsinghualogic.net\/JRC\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"http:\/\/tsinghualogic.net\/JRC\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/tsinghualogic.net\/JRC\/wp-json\/wp\/v2\/comments?post=8024"}],"version-history":[{"count":28,"href":"http:\/\/tsinghualogic.net\/JRC\/wp-json\/wp\/v2\/pages\/8024\/revisions"}],"predecessor-version":[{"id":10018,"href":"http:\/\/tsinghualogic.net\/JRC\/wp-json\/wp\/v2\/pages\/8024\/revisions\/10018"}],"wp:attachment":[{"href":"http:\/\/tsinghualogic.net\/JRC\/wp-json\/wp\/v2\/media?parent=8024"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}