Evaluating adversarial robustness in simulated cerebellum

TL;DR

This study investigates whether key features of the cerebellum enhance its robustness against adversarial attacks, finding no significant improvements and suggesting biological systems may also be vulnerable.

Contribution

First examination of adversarial robustness in simulated cerebellum models, exploring three cerebellar features for potential robustness benefits.

Findings

No significant robustness improvements from proposed mechanisms

Cerebellum likely vulnerable to adversarial examples

Encourages neuroscientists to test biological systems with adversarial attacks

Abstract

It is well known that artificial neural networks are vulnerable to adversarial examples, in which great efforts have been made to improve the robustness. However, such examples are usually imperceptible to humans, and thus their effect on biological neural circuits is largely unknown. This paper will investigate the adversarial robustness in a simulated cerebellum, a well-studied supervised learning system in computational neuroscience. Specifically, we propose to study three unique characteristics revealed in the cerebellum: (i) network width; (ii) long-term depression on the parallel fiber-Purkinje cell synapses; (iii) sparse connectivity in the granule layer, and hypothesize that they will be beneficial for improving robustness. To the best of our knowledge, this is the first attempt to examine the adversarial robustness in simulated cerebellum models. The results are negative in the experimental phase -- no significant improvements in robustness are discovered from the proposed three mechanisms. Consequently, the cerebellum is expected to be vulnerable to adversarial examples as the deep neural networks under batch training. Neuroscientists are encouraged to fool the biological system in experiments with adversarial attacks.