Enhancing Population-based Search with Active Inference

TL;DR

This work investigates enhancing population-based metaheuristics with Active Inference to create anticipatory, cognitive-augmented search agents. It instantiates the idea by embedding belief updates and a free-energy objective into Ant Colony Optimization for the Travelling Salesman Problem, capturing $F = path\_length - (b \\log b + (1-b) \\log (1-b))$ and a belief update $b = 1 - rac{current\_path\_length}{best\_path\_length}$ to steer edge choices and reinforce high-quality tours. Empirical results on randomly generated symmetric graphs show that AI-augmented ACO yields statistically significant improvements in tour length compared to basic ACO, with larger gains for bigger problems and only a marginal relative increase in computation time, though performance varies with graph structure. The findings highlight the potential of integrating Active Inference into PBMH to produce more proactive optimization strategies and motivate extending the approach to other PBMH (e.g., GA, PSO, DE) and to develop theoretical justifications for when and why such augmentations succeed.

Abstract

The Active Inference framework models perception and action as a unified process, where agents use probabilistic models to predict and actively minimize sensory discrepancies. In complement and contrast, traditional population-based metaheuristics rely on reactive environmental interactions without anticipatory adaptation. This paper proposes the integration of Active Inference into these metaheuristics to enhance performance through anticipatory environmental adaptation. We demonstrate this approach specifically with Ant Colony Optimization (ACO) on the Travelling Salesman Problem (TSP). Experimental results indicate that Active Inference can yield some improved solutions with only a marginal increase in computational cost, with interesting patterns of performance that relate to number and topology of nodes in the graph. Further work will characterize where and when different types of Active Inference augmentation of population metaheuristics may be efficacious.

Figures (1)

• Figure 1: Scaling of solution quality improvement and computational overhead with problem size.