Tutorials

Overview

Graphs are the fabric of spatial intelligence, multi-agent system and biological structures, showing fundamental importance in the era of large models. The rise of foundation models has motivated Graph Foundation Models (GFMs), yet current methods often struggle with structural heterogeneity, transferability, and interpretability. This tutorial introduces the advances in GFMs. We start from the roles of graphs in the era of large models, and then provide a unified geometric perspective. Furthermore, we introduce Geometric Morphology Learning and Sheaf Networks for learning graphs and structures. Participants will leave with a coherent view of how graph and geometry empower artificial general intelligence.

Program

Speakers

Overview

High-quality data is essential for data mining and machine learning, yet manual and rule-based management cannot effectively address complex anomalies. AI-powered high quality data management overcomes these limitations by applying machine learning, deep learning, and large language models to enable autonomous adaptation and contextual understanding. This tutorial reviews AI-driven advances in three technologies for creating and maintaining high-quality datasets: data discovery, data preparation, and data integration. It presents case studies involving relational, graph, and spatio-temporal data. By connecting methodologies with applications, the tutorial equips researchers and practitioners with tools for intelligent data management and encourages interdisciplinary research in data-centric AI.

Program

Speakers

Overview

LLMs have demonstrated remarkable performance on different reasoning tasks because of their superior capabilities in natural language understanding, but still struggle with complex tasks such as multi-hop reasoning and long-context reasoning. In this tutorial, we will introduce the state-of-the-art advances in neuro-symbolic reasoning with LLM and KG. The main topics cover neuro-symbolic logical reasoning, KG foundation model, and applications in interdisciplinary science. The technical details will be showcased, and the open challenges are also summarized.

Program

Speakers

Overview

Graphs underpin data science, from social platforms and knowledge bases to financial systems and AI-driven applications. As graph data becomes larger, more dynamic, more sensitive, and integrated with large language models, graph computation must evolve beyond isolated algorithms or learning models. This tutorial begins with an introduction to modern graph computation and then examines four complementary themes: scalable graph query processing, privacy-preserving graph analytics, graph-based financial security, and LLM-assisted graph reasoning. By connecting algorithmic foundations with requirements for privacy, robustness, scalability, and verifiability, it provides a coherent understanding of modern graph computation for the data mining and machine learning community.

Program

Speakers

Overview

Large language models (LLMs) are transforming relational data processing in two synergistic directions: as operator-level primitives—LLM-Enhanced Relational Operators (LROs)—and as the reasoning core of intelligent data agents that orchestrate these operators end-to-end. This tutorial provides a comprehensive 2.5-hour introduction spanning two complementary perspectives: (1) the operator perspective—a formal LRO taxonomy, multi-LRO system architectures, optimization techniques, and benchmark methodology; and (2) the agent perspective—how LROs serve as building blocks within data agents, featuring structured context management, layered skill orchestration, and self-evolving evaluation frameworks. We bridge these perspectives through a unified view of LLM–database co-design, covering both foundational theory and practical system implementations.

Program

Speakers

Overview

This tutorial explores advances in context-aware time series forecasting with large language models, moving beyond historical pattern recognition toward context-grounded cognitive reasoning. It explains why real-world forecasts depend not only on temporal patterns, but also on external environments, events, domain knowledge, and system states. The tutorial introduces three generations of context intelligence: multimodal semantic understanding, slow-thinking reasoning over predictive evidence, and agentic interaction with tools and environments. It shows how these capabilities support interpretation, reasoning, and information acquisition, with representative applications in electricity load forecasting, renewable power generation, traffic, industry, and scientific forecasting, while highlighting open challenges and future directions.

Program

Speakers