Optimality of short-term synaptic plasticity in modelling certain dynamic environments
Timoleon Moraitis,Abu Sebastian,Evangelos Eleftheriou
DOI: https://doi.org/10.48550/arXiv.2009.06808
2020-09-15
Neural and Evolutionary Computing
Abstract:Biological neurons and their in-silico emulations for neuromorphic artificial intelligence (AI) use extraordinarily energy-efficient mechanisms, such as spike-based communication and local synaptic plasticity. It remains unclear whether these neuronal mechanisms only offer efficiency or also underlie the superiority of biological intelligence. Here, we prove rigorously that, indeed, the Bayes-optimal prediction and inference of randomly but continuously transforming environments, a common natural setting, relies on short-term spike-timing-dependent plasticity, a hallmark of biological synapses. Further, this dynamic Bayesian inference through plasticity enables circuits of the cerebral cortex in simulations to recognize previously unseen, highly distorted dynamic stimuli. Strikingly, this also introduces a biologically-modelled AI, the first to overcome multiple limitations of deep learning and outperform artificial neural networks in a visual task. The cortical-like network is spiking and event-based, trained only with unsupervised and local plasticity, on a small, narrow, and static training dataset, but achieves recognition of unseen, transformed, and dynamic data better than deep neural networks with continuous activations, trained with supervised backpropagation on the transforming data. These results link short-term plasticity to high-level cortical function, suggest optimality of natural intelligence for natural environments, and repurpose neuromorphic AI from mere efficiency to computational supremacy altogether.