Tutorials

T1: Artificial Intelligence at the Deutsches Museum: Historical Research and Science Communication

Speakers: Michael Decker (Deutsches Museum), Rudolf Seising (Deutsches Museum), Dinah Pfau (Deutsches Museum), Helen Piel (Vienna University), Susanne Grube (Deutsches Museum Nürnberg)

Although AI is widely regarded as a pivotal technology of the future, it is rooted in a rich and multifaceted history. This tutorial explores approaches to researching AI’s past, examines how exhibitions on the subject can be conceived and designed, and reflects on the possibilities that lie ahead.

The Deutsches Museum explores and showcases technological and scientific achievements of the past, present, and future. Although AI is widely regarded as a key technology of tomorrow, it is deeply rooted in diverse historical processes.

This tutorial offers insights into historical research on AI, approaches to curating AI-related exhibitions, and perspectives on future developments. Across two sessions, we will present and discuss selected facets of AI history from multiple viewpoints:

• Session 1 begins with an introduction to the history of symbolic AI, focusing on the relationship between IJCAI and the early days of West German AI research. The second contribution examines the historical development of connectionist approaches and the enduring power of the organism–machine analogy.

• Session 2 opens with an overview of the emergence of cognitive science as an interdisciplinary field in West Germany and its connections to international AI development. The final segment addresses the challenges, ideas, and opportunities involved in communicating AI—its societal relevance, evolution, and applications—to broad audiences through exhibitions and a range of science communication formats.

T3: Safety and Security of LLMs, Agents and Robots

Speakers: Mark Maybury (Lockheed Martin)

Website: https://github.com/mtmaybury/SafeAITutorial

This tutorial presents an overview on the critical issues of safety, security, and resilience in Large Language Models (LLMs), AI agents and robots. Grounded in real world cases, this session will review responsible AI principles and computational methods to identify and mitigate vulnerabilities and associated harms to improve the safety and security of AI.

T4: Scalable Intelligence under Limited Supervision: Principles, Structures, and Adaptive Agents

Speakers: Quanming Yao (Tsinghua), Zixing Song (University of Bristol), Yaqing Wang (BIMSA)

Website: https://lars-group.github.io/pages/DEAL.html

AI systems are increasingly constrained not only by model scale, but by the availability, reliability, and cost of supervision. This tutorial studies scalable intelligence under limited supervision, covering scarce labels, weak feedback, costly interaction, and unstable self-generated experience. We present a unified view from data-efficient generalization to LLM-based agents that perceive, reason, act, and adapt through interaction. The tutorial is organized around three directions: principles of few-shot, meta-, and in-context learning; structures such as graphs and relational topology as implicit supervision; and adaptive agents that improve through experience augmentation, structural design, optimized in-context learning, parameter-efficient adaptation, and budget-efficient reinforcement learning. Across these topics, we emphasize experience reuse, error propagation, stability, cost-aware evaluation, and trustworthy deployment. The goal is to provide IJCAI attendees with a concise conceptual map and research agenda for building AI systems that can generalize and improve from limited supervision and interaction.

T5: Uncertainty in machine learning: from aleatoric to epistemic

Speakers: Willem Waegeman (Ghent University), Eyke Hüllermeier

Website: https://sites.google.com/view/aeuml-tutorialijcai/home

This tutorial aims to provide an overview of uncertainty representation and quantification in machine learning, a topic that has received increasing attention in the recent past. The main focus is on novel approaches for distinguishing and representing so-called aleatoric and epistemic uncertainty. By the end of the tutorial, attendees will have a comprehensive understanding of the fundamental concepts and recent advances in this field.

We will start the tutorial by defining what we mean with aleatoric and epistemic uncertainty. Subsequently, we will review the most popular methods that disentangle the two types of uncertainty, such as Bayesian methods, ensemble learning methods and evidential deep learning methods. We will also discuss typical applications of uncertainty disentanglement, such as classification with reject option and active learning. Furthermore, we will review recent critical papers about uncertainty disentanglement, emphasizing opportunities and potential pitfalls.

T6: Responsible Mechanism Design

Speakers: Pavel Naumov

Website: https://pavelnaumov.com/ijcai-26

Responsible Mechanism Design is an emerging interdisciplinary area at the intersection of AI, game theory, and philosophy that studies how collective decision-making mechanisms can be analysed and designed with individual accountability in view. This tutorial introduces formal models of responsibility used in AI, surveys recent work on responsibility gaps and diffusion, and discusses existing approaches, limitations, and open questions in responsibility-aware mechanism design.

T7: Data Centric AI: Addressing Missing Data Imputation in Image and Tabular Contexts

Speakers: Joana Cristo Santos (University of Lisbon – LASIGE), Ricardo Cardoso Pereira (University of Coimbra – CISUC), Pedro Henriques Abreu (University of Coimbra – CISUC)

Website: https://ricardodcpereira.com/missing-data-tutorial

Missing data is a common problem in both tabular and imaging datasets. One strategy is to remove records containing missing values, but this is often impractical, especially with small datasets or when missingness is not purely random. Imputation, as in replacing missing values with estimated ones, is the more common alternative, though different methods suit different scenarios. This tutorial presents the advantages and disadvantages of these strategies and helps participants identify when to use each one. The tutorial is organized into two parts. The first addresses missing data in tabular datasets: a lecture covering missingness mechanisms (MCAR, MAR, MNAR) and imputation methods, from statistical baselines to autoencoders and generative adversarial networks (GANs), followed by a hands-on session on a real-world tabular problem. The second part covers missing data in imaging datasets, including convolutional autoencoders and GANs for image reconstruction, with a practical session using healthcare imaging data. All exercises are provided in Python, with programming requirements kept accessible.

T9: Beyond Graph Distribution Shifts: LLMs, Adaptation, and Generalization

Speakers: Xin Wang(Tsinghua university) , Haoyang Li (Tsinghua university), Haibo Chen (Tsinghua university) and Wenwu Zhu (Tsinghua university)

Website: https://ood-generalization.com/ijcai2026Tutorial.htm

Graph machine learning has witnessed rapid progress across both academia and industry. However, most existing methods are developed under the in-distribution (I.D.) hypothesis, which assumes that training and testing graph data are drawn from the same distribution. In real-world applications—ranging from dynamic knowledge graphs to evolving biomedical networks—this assumption is frequently violated, resulting in severe performance degradation under distribution shifts. Addressing this challenge has become a key focus in recent years, leading to the development of novel paradigms that move beyond the I.D. setting. This tutorial presents a comprehensive overview of three emerging and synergistic directions for tackling distribution shifts in graph learning. First, we highlight Graph LLMs, which combine the representational power of large language models with graph structures to enable flexible, in-context, and few-shot learning on graphs. Second, we introduce adaptation techniques for both GNNs and Graph LLMs, including graph neural architecture search and continual learning strategies for evolving data. Third, we cover generalization methods that incorporate causality and invariance principles to build robust graph models under unseen distributions. We will advocate novel, high-quality research findings, as well as innovative solutions to the challenging problems in graph machine learning under distribution shifts and the applications on graphs. This topic is at the core of the scope of IJCAI, and is attractive to machine learning as well as data mining audience from both academia and industry.

T11: Out-of-distribution Generalized Generative AI

Speakers: Xin Wang (Tsinghua University), Zirui Pan (Tsinghua University), Yuwei Zhou (Tsinghua University), Wenwu Zhu (Tsinghua University)

Website: https://mn.cs.tsinghua.edu.cn/oodgenai-ijcai2026.html

This tutorial presents recent advances in multi-modal generative AI, with a focus on multi-modal large language models (MLLMs) for multi-modal understanding and diffusion-based models for visual generation, providing a systematic overview of their probabilistic formulations, architectural designs, and mechanisms for multi-modal interaction. It further examines the challenges of deploying these models in out-of-distribution (OOD) environments and discusses generalizable post-training strategies and unified frameworks to enable robust understanding and generation under distribution shifts and emerging application scenarios.

T12: Neural Networks and Differential Equations: From Infinite Layers to Continuous Modelling

Speakers: Cecília Coelho (Helmut Schmidt University, Germany), Luís Ferrás (University of Porto, Portugal), Andrzej Dulny (University of Würzburg, Germany)

Website: https://ceciliacoelho.github.io/tutorialNN4DEs

This tutorial offers an accessible but comprehensive journey into the emerging intersection of neural networks and differential equations, two pillars at the core of scientific modelling and modern AI. Beginning with an introduction to the theory and motivation behind differential equations, we build toward the conceptual and algorithmic foundations of architectures that integrate them with neural networks. Participants will develop a clear understanding of how approaches like Physics-Informed Neural Networks (PINNs), Neural Operators, and Neural Ordinary Differential Equations (Neural ODEs) are transforming modelling, simulation, and data-driven discovery across scientific domains.

T13: Automated Machine Learning

Speakers: Jovita Lukasik (University of Siegen), Marius Lindauer (L3S Research Center, Leibniz University Hannover), Marcel Wever (L3S Research Center, Leibniz University Hannover)

Website: https://sites.google.com/auto-ai.org/ijcai-ecai26-automl-tutorial

This half-day tutorial offers a comprehensive, up-to-date tour of automated machine learning (AutoML), spanning its classical foundations, its integration with modern foundation models and LLMs, human-centered design principles, neural architecture search, resource/sustainability-aware methods, and an outlook to the most recent state-of-the-art approaches. Attendees will leave equipped with both the theoretical grounding and practical intuitions needed to apply and advance AutoML in research and industry settings.

T14: AI for Visual Analytics

Speakers: Pranava Madhyastha (City, University of London; The Alan Turing Institute), Maeve Hutchinson (City, University of London)

Website: https://aiforvis.github.io/

Visual Analytics (VA) supports human reasoning and decision-making about complex data through data analysis and interactive visualization. Increasingly, AI is becoming a central medium through which people interpret, construct, and interact with visual representations of data. In this tutorial, we will provide a comprehensive and structured overview of the integration of AI into the core processes of data analysis and visualization. We will present this tutorial as a unified framework for AI-mediated visual data analysis across three core paradigms: reading visualization, where models perform reasoning over chart images through tasks such as question answering and insight extraction; and authoring visualization, where AI systems produce visual representations of data; and interactive visual analytics where we will explore the frontier of AI-augmented VA systems, where multi-turn, multimodal dialogue is enabling iterative and collaborative large-scale data exploration across multiple fields and domains.

T15: LLMs for Optimization: From Automated Modeling to Algorithmic Discovery

Speakers: Connor Lawless (Stanford University), Fei Liu (UZH&ETH Zurich), Hanzhang Qin (National University of Singapore), Ellen Vitercik (Stanford University)

Website: https://feiliu36.github.io/llm_opt_tutorial_ijcai2026/

Mathematical optimization is a foundational pillar of modern AI, underpinning decision-making in supply chains, energy systems, finance, and scheduling. Despite its importance, building and deploying optimization models remains a challenging, expert-driven process requiring significant domain knowledge and technical expertise.

This tutorial surveys the emerging intersection of LLMs and optimization along two complementary themes. First, we examine how LLMs can act as copilots across the optimization pipeline — assisting with problem formulation, model construction, solver configuration, and validation. Second, we explore the growing role of LLMs in algorithmic discovery: generating, refining, and uncovering new optimization algorithms and heuristics.

The tutorial bridges machine learning and optimization, covering foundational concepts, state-of-the-art methods and systems, and key open challenges — including correctness, robustness, and handling ambiguous problem specifications.

T17: Deep Reinforcement Learning for Combinatorial Optimization

Speakers: Zaharah Bukhsh (TU/e), Yaoxin Wu (TU/e), Yingqian Zhang (TU/e), Jesse van Remmerden (TU/e), Yue Yu (TU/e)

Website: https://ai-for-decision-making-tue.github.io/drl-co-tutorial/

Combinatorial optimization problems such as vehicle routing and job shop scheduling are ubiquitous in real-world decision-making but remain computationally challenging due to their NP-hard nature. The primary focus of this tutorial is a hands-on, step-by-step guide to developing end-to-end DRL solutions for COPs. Participants will be walked through the four essential design pillars 1) instance encoding schemes, 2) Markov decision process formulations, 3) neural network architectures, and 4) reinforcement learning training algorithms, with vehicle routing and job shop scheduling serving as concrete running examples throughout.

Building on these hands-on foundations, the tutorial offers two complementary perspectives. First, a research landscape overview introduces participants to the literature beyond classical TSP and JSP benchmarks, surveying how the core design pillars extend to real-world problem variants involving multiple objectives, diverse constraints, and accounting for uncertainties in decision-making. Second, the tutorial concludes with an open discussion of the field’s most pressing challenges.

T18: Learning General Representations to Act and Plan: Models and Policies

Speakers: Blai Bonet (Universidad Simón Bolívar), Hector Geffner (RWTH Aachen University)

Models for planning are usually hand crafted, while policies learned with reinforcement learning do not generalize. The tutorial covers research that addresses these limitations by learning action models and policies that are general, using both combinatorial and deep

learning methods.

T19: Strategic Behavior in Stable Matching Markets: from Preferences to Capacities

Speakers: Hadi Hosseini (Penn State University), Shivika Narang (UNSW Sydney), Shraddha Pathak (Penn State University), Rohit Vaish (IIT Delhi)

Website: https://sites.google.com/view/tutotial-sbsm/home

Two-sided matching markets are ubiquitous in the real world, modeling scenarios such as ride-sharing platforms and college admissions. Stability and strategyproofness are two desirable properties in these systems, but are known to be incompatible.

This tutorial surveys the literature on strategic behavior in these markets, covering classical results and recent developments, with a focus on manipulations via preference misreports (where agents alter their ordinal rankings over potential matches) and capacity modifications (where agents misreport the number of agents they can be matched to). Throughout, the tutorial will also highlight open problems and avenues for future work.

T20: Brain Information Computing and Decoding for Advanced BCIs: From Basic to Frontiers

Speakers: Yu Qi (Zhejiang University), Yang Yang (Zhejiang University)

Website: https://ijcai-brain-decoding.github.io/

Recent advances in artificial intelligence are reshaping the field of brain-computer interfaces. This tutorial delves into two key frontiers driving this transformation.

The first part focuses on Foundation Models for Brain Signals. Traditional deep learning models are often task-specific and dataset-limited, hindering their generalization. This section will introduce a paradigm shift towards developing large-scale, self-supervised pre-trained foundation models for both non-invasive EEG signals and invasive sEEG signals. We will explore how to design spatiotemporal pre-training strategies to learn universal representations from heterogeneous multi-dataset sources, and the effective learning of time-space-frequency features. The session will cover state-of-the-art model architectures, pre-training objectives, and their subsequent fine-tuning for diverse downstream tasks.

The second part of the tutorial focuses on Invasive Brain Neural Decoding for BCI. It will provide a foundational understanding of motor BCIs, from their neural basis to the computational models. We will then delve into state-of-the-art decoding methodologies, starting from foundational state-space models and progressing to modern architectures like Mamba. We will further explore how dynamic or domain adaptation techniques can achieve stable, cross-day decoding performance, addressing one of the most critical barriers to the real-world deployment of invasive BCIs.

T21: Deep Parameterized Logics as A Foundation for Neurosymbolic AI

Speakers: Luc De Raedt (KU Leuven), Giuseppe Marra (KU Leuven), Robin Manhaeve (KU Leuven), Vincent Derkinderen (KU Leuven)

Website: dtai.cs.kuleuven.be/tutorials/deeplog/editions/ijcai2026

Neurosymbolic AI integrates the learning capabilities of neural networks with the structured reasoning provided by symbolic methods. Although the field has produced a wide variety of approaches, spanning probabilistic and fuzzy logics, differentiable reasoning, and neural-guided symbolic systems, these methods often appear fragmented, difficult to compare and, therefore, to build upon. In this tutorial, we show that many neurosymbolic approaches share a common underlying representation and inference perspective. By making this shared structure explicit, we provide a unifying view that helps researchers better understand existing systems and design new ones.

The tutorial first introduces the motivation and key applications of integrating learning and reasoning, followed by an overview of the major dimensions along which neurosymbolic systems differ. We then present a broad neurosymbolic formulation that captures a wide range of existing approaches. Building on this, we revisit foundational concepts—including algebraic structures, model counting, and arithmetic circuits—that underpin many modern neurosymbolic inference methods. Using these foundations, we introduce the DeepLog framework as a language for constructing neurosymbolic systems and illustrate its use through a running example. The tutorial concludes with a walk-through session in which participants learn how to build a neurosymbolic system from scratch, demonstrating the practical implementation of these ideas.

T23: A Tutorial on Recent Advances in Generative Conversational Recommender Systems

Speakers: Ahmadou Wagne (TU Wien, Austria), Thomas E. Kolb (TU Wien, Austria), Julia Neidhardt (TU Wien), Yashar Deldjoo (Polytechnic University of Bari, Italy)

Website: https://recsys-lab.at/gen-conv-recsys-tutorial-ijcai-ecai-2026/

This tutorial introduces Generative Conversational Recommender Systems (Gen-CRS), a rapidly emerging paradigm that integrates generative models into conversational recommendation. Moving beyond traditional modular pipelines, Gen-CRS enable coherent multi-turn interactions, nuanced intent understanding, and context-aware personalization. Based on a survey of recent literature (2023–2026), we organize the design space along three key dimensions: system type (modular, unified, agentic), dialogue initiative (user-, system-, mixed-initiative), and recommendation method (retrieval-based, generative, hybrid). The tutorial provides a comprehensive overview of architectures, model adaptation techniques, knowledge integration, simulation methods, and evaluation practices. Combining theoretical foundations with practical guidance, it equips participants to design and assess next-generation systems in a rapidly evolving research field while highlighting open challenges such as controllability, grounding, and evaluation in LLM-driven conversational environments. The tutorial is an updated and revised version incorporating feedback collected during a previous edition presented at ACM RecSys 2025 in Prague. Furthermore, it is supported by an upcoming survey paper on Gen-CRS.

T24: The State-Of-the-Art in Explanation Methods for Two-Dimensional Embeddings of Data

Speakers: Edith Heiter (Ghent University), Fuyin Lai (Ghent University), Jefery Lijffijt (Ghent University)

Website: https://sites.google.com/view/x-embeddingtutorial/home

Dimensionality reduction methods are widely used to create two-dimensional views of high-dimensional data, but it is not so straightforward how to make sense of apparent patterns such as clusters or paths. This tutorial offers a structured guide to which explainability techniques exist that link these patterns back to the original features, ranging from feature-importance and counterfactual approaches to interactive annotation of low-dimensional embeddings.

T25: Compression vs. Accuracy: Compact Models for Efficiency and Interpretation

Speakers: Malte Luttermann (University of Hamburg), Jan Speller (University of Münster), Marcel Gehrke (University of Hamburg), Tanya Braun (University of Münster)

Website: https://www.uni-muenster.de/Informatik.AGBraun/en/research/tutorials/ijcai-26.html

Our surrounding world is inherently uncertain and relational. The field of Statistical Relational AI (StaRAI) has emerged to account for both. StaRAI explicitly encodes objects and relations in probabilistic models, which enables algorithms to exploit repeated structures, i.e., isomorphic subgraphs with matching associated probability functions, for efficiency gains during inference. While such repeated structures frequently occur in many practical applications, they are generally not explicitly represented in a learned model and thus cannot be exploited. It is therefore crucial to efficiently identify and compress these structures. Next to a significant reduction in storage requirements, dedicated inference algorithms can use these compressed structures for efficiency gains, yielding tractability in the number of random variables. This tutorial provides a look at recent advances in the task of computing a highly compressed model from a given propositional model. We consider how the compression can efficiently be realised. Furthermore, we discuss the approximation of a compressed representation, give error bounds for the induced approximation error as well as investigate how to obtain a compressed model for a given error bound. Additionally, for the trade-off between accuracy and compression, we also discuss a hierarchical approach and provide different points of view on the clustering in preparation for better interpretability.

T26: Hands-On Cognitive Robotics: Knowledge-Driven Action Execution using a Virtual Research Environment

Speakers: Michael Beetz (University of Bremen), Michaela Kümpel (University of Bremen), Vanessa Hassouna (University of Bremen)

Website: https://vrb.ease-crc.org/aicor-tutorial/

This two-day tutorial provides a comprehensive and hands-on introduction to knowledge-driven approaches for robot action representation, reasoning, and execution. The tutorial integrates theoretical foundations with practical exercises, guiding participants through the full pipeline of cognitive robotics, including scene understanding, knowledge acquisition and semantic reasoning, as well as robot action execution in simulated environments.

Participants will learn how to design a virtual environment including a robot agent, perceive objects and access semantic object properties to enable robots to perform a manipulation task like ”getting the milk from the fridge”. The methods and tools introduced are generalizable to a wide range of robotic applications.

The hands-on components are delivered through interactive Jupyter Notebooks within a Docker-based environment, ensuring accessibility and reproducibility across different technical setups. An immersive learning experience is provided through physics-based simulation, enabling participants to visualize and experiment with robot behaviors. By combining established teaching material with an end-to-end system perspective, this tutorial offers a scalable and reproducible framework for education in cognitive robotics and knowledge-based AI systems.

T27: A Tutorial on Unified Multimodal Models

Speakers: Jindong Wang (William & Mary), Yinyi Luo (Carnegie Mellon University), Haoyue Bai (William & Mary)

Website: https://aifrontierlab.github.io/IJCAI_Tutorial/

Multimodal understanding and generation have long been considered related but separated tasks.

The emergence of unified multimodal models (UMMs) makes it possible to unify the two tasks within a single foundation model.

Such capability integration has triggered new research interest on the architecture design, training dynamics, data integration techniques, evaluation pipelines, and potential downstream applications.

This tutorial is dedicated to present the first comprehensive overview and lecture of UMMs to both academic and industrial researchers by surveying, categorizing, and analyzing existing literature.

We aim to provide a holistic understanding of the representation learning, training dynamics, and architecture design of UMMs, as well as evaluation benchmarks and downstream applications.

Finally, we discuss several opening challenges, sharing insights for future research.

T28: Dynamical causal modelling: a Bayesian computational framework for multi-scale hypothesis-testing

Speakers: Ulrich Stoof (University College London), Richard Rosch (King’s College London), Klaus Krause (King’s College London)

Website: https://github.com/IJCAI-DCM/Tutorial

Dynamic causal modelling (DCM) is an established causal modelling technique in neuroscience, which is widely used to test hypothesis related to neural imaging (functional magnetic resonance imaging (fMRI)) and electrophysiological data (magneto- / electroencephalography (M/EEG)). The technique is effective, as it relies on approximate Bayesian inference using variational Laplace, transparent and flexible, as an open source academic freeware which can be adjusted as needed, and it allows hierarchical hypothesis-testing via model specifications, priors, and group-level constraints.

This tutorial will teach the conceptual background and practical skills of mesoscopic model building in (human) neuroscience using DCM. We will focus on multi-modal hypothesis-testing with a concrete case study of how to link structural neural data (neurotransmitter) to functional electroencephalographic data. The taught skills are essential to utilise vast amounts of (openly-accessible) neural / biological data effectively for testing multi-scale hypothesis about structure-function relationships in (human) brains; it facilitates translational / pharmaceutical / medical, and computational research.

T29: Human-AI Co-Creativity

Speakers: Adish Singla (Max Planck Institute for Software Systems, Germany)

Website: https://genaicreativity.org/ijcai2026/

Recent advances in large generative models have turned AI agents into everyday companions. Millions of people now rely on them across domains from design and communication to science and education. These advances offer unprecedented opportunities to support people in open-ended creative domains. For instance, these agents can provide a medium for brainstorming ideas and exploring design choices to improve creative outcomes. However, existing models and interfaces are primarily designed for automation. Consequently, their increasing use for creative tasks could lead to design fixation and idea homogeneity, as well as raise issues of content copyright and authorship. The goal of this tutorial is to provide an overview of these advances, opportunities, and research challenges in generative AI, creativity, and human-AI co-creation. The tutorial will focus on two thrusts: (i) leveraging recent advances in generative AI to support people in open-ended creative tasks; (ii) identifying unique challenges when integrating generative AI into creative workflows that require safeguards along with technical innovations. The tutorial will provide a common ground to facilitate tighter connections among researchers, industry professionals, and practitioners interested in generative AI, creativity, and human-AI co-creation.

T30: Neuroevolution of Intelligent Agents

Speakers: Yujin Tang (Sakana.ai), Sebastian Risi (IT University of Copenhagen), David Ha (Sakana.ai), Risto Miikkulainen (The University of Texas at Austin)

Website: https://www.cs.utexas.edu/~risto/talks/ijcai26-tutorial/

Neuroevolution, or optimization of neural networks through evolutionary computation, is a method for constructing intelligent agents through population-based search. It is particularly useful in partially observable domains with sparse and multiobjective reinforcement; compared to other policy search techniques, its power comes from extensive exploration that allows it to find effective, often surprising solutions. Prime application domains include robotic control, game-playing agents, and decision-making. More recently it has also been extended to optimizing deep-learning architectures, understanding how biological intelligence evolved, and optimizing neural networks for hardware implementation. It can also be used synergistically with reinforcement learning and LLMs, adding an element of exploration to those techniques. The tutorial introduces participants to neuroevolution fundamentals, progresses to several advanced topics that make neuroevolution effective and general, reviews example application areas, and proposes further research questions. It is accompanied by a hands-on exercise that makes the concepts concrete and allows the participants to take advantage of neuroevolution immediately.

T31: LLMs for Constraint Modeling

Speakers: Tias Guns (KU Leuven), Serdar Kadioglu (Brown University & Fidelity Investments), Dimos Tsouros (UOWM)

Website: https://sites.google.com/view/ijcai26-llm-con

Constraint Programming (CP) and Operations Research (OR) are powerful tools for optimization, but their adoption is often limited by the expertise needed to translate real-world problems into formal models. Recent advances in Large Language Models (LLMs) offer a promising way to reduce this modeling bottleneck by assisting in the generation of variables, constraints, and objectives from natural language descriptions.

This tutorial provides a concise overview of LLM-assisted constraint modeling for discrete and continuous optimization. It surveys recent methods, systems, and benchmarks for generating executable models, including prompt-based pipelines, intermediate representations, retrieval-augmented learning, self-reflection, fine-tuning, and agentic approaches. The tutorial also focuses on evaluation, considering correctness, solution quality, robustness, and practical usability. Participants will gain a clear understanding of the state of the art, the key challenges, and future opportunities for LLM-based modeling assistants.

T32: Current Advances in Reasoning with Large Language Models

Speakers: Akhil Arora (Aarhus University), Vishrav Chaudhary (Meta Superintelligence Labs), Julia Kreutzer (Cohere Labs), Nearchos Potamitis (Aarhus University), Lars Klein (EPFL), Nouha Dziri (Allen Institute for AI), Niket Tandon (Microsoft Research)

Website: https://llmreasoning.github.io/

Reasoning has become a central capability in modern AI systems built on large language models (LLMs), with rapid advances spanning evaluation, inference-time strategies, post-training, agentic systems, and multilingual generalization. This tutorial is organized around three questions of broad interest to the AI community: How well do current models reason? How can we make them reason better? What are the next frontiers for reliable, efficient, and inclusive reasoning?

We present a unified and practical overview of recent progress in LLM reasoning, covering evaluation beyond accuracy, structured and agentic inference-time methods, reinforcement-learning-based post-training, and frontiers in multilingual, high-stakes, and compute-efficient reasoning. The tutorial combines conceptual synthesis with demos and short guided hands-on exercises, and is intended for AI researchers and practitioners seeking a balanced, up-to-date introduction to reasoning in contemporary foundation models.