NeuronSoup: Evolving Asynchronous, Shared-Neuron Temporal Graphs without Backpropagation

2026-07-16

AI Summary

This work addresses the limitations of conventional deep learning—namely its reliance on backpropagation, fixed computational graphs, and synchronous inter-layer processing—by introducing a novel neural computing paradigm that does not require differentiable architectures. The proposed approach constructs an asynchronous, temporally structured graph network with shared neurons, where signals propagate through a shared neuron pool via delay modulation. Network topology, connection weights, and propagation delays are co-evolved using a genetic algorithm, eliminating the need for backpropagation. The model enables sample-adaptive computational depth and automatically discovers lateral interactions among pathways. Evaluated on a ten-class MNIST classification task using frozen ResNet18 features, the evolved architecture achieves 85.9% test accuracy after 10,000 generations with a model size of only 115 KB, demonstrating both the efficacy and representational capacity of the proposed framework.

Abstract

We present NeuronSoup, a neural computation architecture that replaces synchronous layer-by-layer processing with asynchronous, delay-mediated signal propagation through a pool of shared neurons. Each path in the network routes a continuous-valued signal from one input neuron to one output neuron through a variable number of intermediate hidden neurons. Hidden neurons are physically shared across paths: when two paths pass through the same neuron, the second arrival encounters the accumulated state left by the first, producing constructive or destructive interference that depends on signal polarity and arrival timing. The entire architecture -- topology, weights, delays, and connectivity -- is co-evolved by a genetic algorithm operating on a flat real-valued genome of 14,602 genes. On 10-class MNIST digit classification using frozen ResNet18 features as input, the system evolves a network of 204 active paths through 266 hidden neurons (156 shared across multiple paths, with one neuron participating in 11 distinct paths) and achieves 85.9\% test accuracy after 10,000 generations. The trained model occupies 115 KB. We argue that this architecture addresses fundamental limitations of current deep learning: it requires no differentiable computation graph, adapts its computation depth per-sample, and discovers lateral interactions between processing pathways that current architectures must engineer explicitly. We discuss why genetic algorithms are the correct optimization tool for this problem class, why CMA-ES fails at this scale, and how the architecture generalizes to arbitrary domains by substituting the encoder and output structure.

Problem

• asynchronous neural computation
• shared neurons
• non-backpropagation learning
• adaptive computation depth
• lateral interactions

Innovation

• asynchronous neural computation
• shared-neuron architecture
• genetic algorithm co-evolution
• delay-mediated signal propagation
• backpropagation-free learning