18
1

Biology-inspired joint distribution neurons based on Hierarchical Correlation Reconstruction allowing for multidirectional neural networks

Abstract

Biological neural networks seem qualitatively superior (e.g. in learning, flexibility, robustness) from current artificial like Multi-Layer Perceptron (MLP) or Kolmogorov-Arnold Network (KAN). Simultaneously, in contrast to them: have fundamentally multidirectional signal propagation~\cite{axon}, also of probability distributions e.g. for uncertainty estimation, and are believed not being able to use standard backpropagation training~\cite{backprop}. There are proposed novel artificial neurons based on HCR (Hierarchical Correlation Reconstruction) removing the above low level differences: with neurons containing local joint distribution model (of its connections), representing joint density on normalized variables as just linear combination among (fj)(f_\mathbf{j}) orthonormal polynomials: ρ(x)=jBajfj(x)\rho(\mathbf{x})=\sum_{\mathbf{j}\in B} a_\mathbf{j} f_\mathbf{j}(\mathbf{x}) for x[0,1]d\mathbf{x} \in [0,1]^d and BB some chosen basis, with basis growth approaching complete description of joint distribution. By various index summations of such (aj)(a_\mathbf{j}) tensor as neuron parameters, we get simple formulas for e.g. conditional expected values for propagation in any direction, like E[xy,z]E[x|y,z], E[yx]E[y|x], which degenerate to KAN-like parametrization if restricting to pairwise dependencies. Such HCR network can also propagate probability distributions (also joint) like ρ(y,zx)\rho(y,z|x). It also allows for additional training approaches, like direct (aj)(a_\mathbf{j}) estimation, through tensor decomposition, or more biologically plausible information bottleneck training: layers directly influencing only neighbors, optimizing content to maximize information about the next layer, and minimizing about the previous to minimize the noise.

View on arXiv
Comments on this paper