Spatial partitioning improves coordinate-based neuroevolution on mnist images
Breaking the Central Bias: Spatially Partitioned Experts for Coordinate-Based Neuroevolution
Neural and Evolutionary ComputingComputer Vision and Pattern RecognitionMachine Learning
Summary
Some AI methods evolve neural networks by mapping input pixels to coordinates. On the MNIST digit dataset, these methods get stuck focusing only on the center pixels, hurting accuracy. The authors show that splitting the image into parts and evolving separate specialist networks for each part forces coverage of the whole image, improving performance significantly. This change highlights that the problem was due to network design bias, not the evolutionary search process itself.
What this means in practice
- •For machine learning engineers: Improve evolutionary neural network training by using spatial partitions to avoid input bias, boosting performance on tasks similar to image classification.
- •For computer vision developers: Design specialized spatial expert networks that collectively enhance feature discovery across image regions for better input utilization in indirect encoding models.
Tested on one dataset.
Authors
Romain Claret, Arthur Gygax, Michael O'Neill, Paul Cotofrei, Michael Palma Mendes, Pascal Felber
Abstract
Evolvable-Substrate HyperNEAT (ES-HyperNEAT), a bio-inspired indirect encoding that determines neuron placement and connection weights from spatial coordinates, exhibits a failure mode on MNIST as a diagnostic benchmark. Because input pixels map to a coordinate space centered at the origin, evolved networks converge on a small central cluster of input pixels, a spatial-concentration bias; prior work observed only 21% mean accuracy in this regime. Is this bias an optimization artifact or an architectural ceiling? Inspired by Mixture-of-Experts (MoE) principles, we partition the input into non-overlapping spatial segments, each assigned to a separately evolved specialist network. With 13 such experts, this design reaches 43% mean accuracy, a 106% relative improvement over the baseline. The architectural gain does not depend on data-driven aggregation: equal-weighted averaging, which uses no validation data, already yields a 70% improvement; the gain comes from partitioning, not the weighting. Receptive-field analysis shows the mechanism: partitioning forces evolution to discover features across the entire image, expanding active pixel coverage from 4% to 79%. Absolute accuracy stays below gradient-trained baselines, but the relative gain points to central bias, not the evolutionary search. Two tools are designed to generalize beyond MNIST: a receptive-field diagnostic for silent input-coverage collapse, and a spatial-partitioning remedy that restores coverage.