Agent Reasoning with Graph World Models

Session Track: Graph Memory & Agents

Session Time: 12:30 pm - 01:00 pm April 15

Session description

World models have long been a key area of research in robotics and spatial applications, however as LLM-based software agents become more capable and more widely deployed, many of the key underlying ideas and principles of world model design and implementation provide a powerful conceptual framework for developers to improve agent performance across any domain.

In this session, we will review how graph representations can be used in world models, and you will learn how to think more explicitly about modelling states, actions, and environments. The focus will be on how to iterate on your existing graph schema and agent architecture, however the session will also be suitable for you if you don’t have those things implemented yet but are interested in how world models can inform your initial system design. You will learn to think differently about what a world model is and why this topic is important for everyone involved in the design and implementation of agents.