Behavior-Inspired Neural Networks for Relational Inference

2024-06-20

AI Summary

This work addresses the challenge of physics-based relational inference in multi-agent systems, noting that conventional modeling of relations as mutually exclusive categorical distributions contradicts the inherently mixed and dynamic nature of real-world interactions. To this end, we propose a behavior-driven neural framework: (i) a novel preference mapping layer that projects individual agent behaviors onto latent relational category preferences; (ii) integration of physics-informed proximity constraints with nonlinear opinion dynamics to enable interpretable relational identification, long-horizon trajectory forecasting, and controllable behavior generation. The architecture unifies graph neural networks with multi-scale spatiotemporal modeling. Evaluated on multiple benchmarks, our method achieves state-of-the-art performance in long-horizon trajectory prediction while significantly enhancing relational interpretability and controllability.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Multiagent Systems: Agent-Based Simulation and Emergent Behavior
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Studies of user behavior, including longitudinal effects of personalized systems

Abstract

From pedestrians to Kuramoto oscillators, interactions between agents govern how a multitude of dynamical systems evolve in space and time. Discovering how these agents relate to each other can improve our understanding of the often complex dynamics that underlie these systems. Recent works learn to categorize relationships between agents based on observations of their physical behavior. These approaches are limited in that the relationship categories are modelled as outcomes of categorical distribution, when in real world systems categories often intermingle and interact. In this work, we introduce a level of abstraction between the observable behavior of agents and the latent categories that determine their behavior. To do this, we learn a mapping from agent behavior to agent preferences for each latent category in a graph neural network. We integrate the physical proximity of agents and their preferences in a nonlinear opinion dynamics model which provides a mechanism to identify mutually exclusive latent categories, predict an agent's evolution in time, and control an agent's physical behavior. We demonstrate the utility of our model for learning interpretable categories, and its efficacy on long-horizon prediction across several benchmarks where we outperform existing methods.

Problem

• Discovering agent relationships in dynamical systems
• Modeling mixed and interacting relationship categories
• Predicting and controlling agent behavior over time

Innovation

• Mapping agent observations to latent preferences
• Nonlinear opinion dynamics model integration
• Interpretable categories and trajectory prediction