Biological connectomes as a representation for the architecture of artificial neural networks

Abstract

Grand efforts in neuroscience are working toward mapping the connectomes of many new species, including the near completion of the Drosophila melanogaster. It is important to ask whether these models could benefit artificial intelligence. In this work we ask two fundamental questions: (1) where and when biological connectomes can provide use in machine learning, (2) which design principles are necessary for extracting a good representation of the connectome. Toward this end, we translate the motor circuit of the C. Elegans nematode into artificial neural networks at varying levels of biophysical realism and evaluate the outcome of training these networks on motor and non-motor behavioral tasks. We demonstrate that biophysical realism need not be upheld to attain the advantages of using biological circuits. We also establish that, even if the exact wiring diagram is not retained, the architectural statistics provide a valuable prior. Finally, we show that while the C. Elegans locomotion circuit provides a powerful inductive bias on locomotion problems, its structure may hinder performance on tasks unrelated to locomotion such as visual classification problems.

Training Spiking Neural Networks Using Lessons From Deep Learning

“…Stochastic approaches can trade computation with naturally arising circuit noise [131], [132], [133]. A brief summary of several common approaches to mitigating the spatial credit assignment problem is provided in the following [134].…”

Assessing Deep Learning: A Work Program for the Humanities in the Age of Artificial Intelligence

“…One initial aim of creating correspondences between the workings of the brain and AI systems was to better understand the human brain, self, and behavior (see, e.g., [192][193][194]). Indeed, AI can be very helpful in researching human beings, but its architectural similarity with the human brain should not be overstated, as the majority of what we know, e.g., about the learning process in the brain, has not been integrated in DL-or only in an immensely simplified manner [195][196][197]. DL anthropomorphism, however, and the dynamics of seeing ourselves in the image of our technology, has a pedigree reaching back to antiquity.…”

Embodied Cognitive Morphogenesis as a Route to Intelligent Systems

“…As a computational approach, connectionist models such as neural networks can be related to a variety of biological connectomes. While biophysical realism is not essential to the exhibition of intelligent behavior, replicating the features of specific connectomes is not particularly useful for simulating generalized intelligence [23]. Yet critical to the emergence and function of a connectome is anatomical specificity, and more specifically, orientation to the environment [24].…”