Compact Modeling of Nanocluster Functionality as a Higher-Order Neuron

Disordered nanoclusters with multielectrode input–output functionality had recently been experimentally realized energy-efficient and emergent computational capacity, thus an interconnected network of several such proposed to realize artificial neural networks. To aid that end, here we show nanocluster can be fit the simplest dendritic neuron model (DNM), where only form nonlinearity is due multiplicative interactions. This work brings into spotlight higher-order networks (known for their efficient encoding geometric invariances) serve as explainable baseline nano-networks against which experimentalists compare more sophisticated models (deep or physics-based lin-min introduced here) provides ground designing novel approximate hardware a statistical mechanics analysis learning performance versus perceptrons (where neurons output nonlinear function weighted sum inputs). A just ten shown achieve classification accuracy than 96% on MNIST benchmark handwritten digit recognition (which required 100 times in three-layer perceptrons).