Learning by non-interfering feedback chemical signaling in physical networks

TL;DR

This paper introduces a novel chemical signaling learning algorithm for physical networks inspired by slime mold, which encodes error information chemically without needing to store multiple states, and demonstrates its effectiveness and biological plausibility.

Contribution

The paper proposes a new chemical signaling-based learning algorithm that avoids storing multiple states, inspired by biological systems like slime mold, and proves its gradient descent capability.

Findings

Achieved 93% accuracy on Iris dataset

Proved the algorithm performs gradient descent

Compared favorably with EP and CL algorithms

Abstract

Both non-neural and neural biological systems can learn. So rather than focusing on purely brain-like learning, efforts are underway to study learning in physical systems. Such efforts include equilibrium propagation (EP) and coupled learning (CL), which require storage of two different states-the free state and the perturbed state-during the learning process to retain information about gradients. Inspired by slime mold, we propose a new learning algorithm rooted in chemical signaling that does not require storage of two different states. Rather, the output error information is encoded in a chemical signal that diffuses into the network in a similar way as the activation/feedforward signal. The steady state feedback chemical concentration, along with the activation signal, stores the required gradient information locally. We apply our algorithm using a physical, linear flow network and test it using the Iris data set with 93% accuracy. We also prove that our algorithm performs gradient descent. Finally, in addition to comparing our algorithm directly with EP and CL, we address the biological plausibility of the algorithm.