Topological Similarity between Artificial and Biological Neural Networks

Inspired by biological neural networks (BNNs), deep artificial neural networks (ANNs) have largely reshaped artificial intelligence nowadays. Neural encoding and decoding studies have shown how and to what extent the information representation in ANNs functionally resembles the brains. Meanwhile, researchers start to investigate how ANNs' predictive performance relates to their topological structures by building relational graphs of computational ANNs. These studies bring the opportunity to assess how is the structural organization of ANNs analogous to BNNs. However, further efforts are necessary to answer this question as the graphical metrics and BNNs are limited in previous studies. In this study, we evaluate the topological similarities between several representative ANNs and a battery of BNNs in different species using a rich set of graphical metrics. We sought to answer two questions: 1) What are the appropriate graphical metrics to characterize the topological similarity between ANNs and BNNs? 2) Is the evolution of ANNs analogous to that of BNNs? Our results show that: 1) the ANN-BNN topological similarity patterns are distinguishable in several graphical metrics; 2) The evolution of ANNs to some extent is analogous to that of BNNs. These findings may provide novel clues for designing brain-inspired neural architectures.