Can Biologically Plausible Temporal Credit Assignment Rules Match BPTT for Neural Similarity? E-prop as an Example

TL;DR

This paper shows that a biologically plausible learning rule called e-prop can match the neural data similarity of traditional backpropagation methods while maintaining high task performance, emphasizing the importance of architecture and initial conditions.

Contribution

It demonstrates that e-prop, a biologically plausible learning rule, can achieve neural data similarity comparable to BPTT, highlighting progress in biologically plausible learning algorithms.

Findings

E-prop matches BPTT in neural data similarity at similar task accuracy.

Model architecture and initial conditions influence neural similarity more than the learning rule.

BPTT and biologically plausible models exhibit similar dynamics at comparable performance.

Abstract

Understanding how the brain learns may be informed by studying biologically plausible learning rules. These rules, often approximating gradient descent learning to respect biological constraints such as locality, must meet two critical criteria to be considered an appropriate brain model: (1) good neuroscience task performance and (2) alignment with neural recordings. While extensive research has assessed the first criterion, the second remains underexamined. Employing methods such as Procrustes analysis on well-known neuroscience datasets, this study demonstrates the existence of a biologically plausible learning rule -- namely e-prop, which is based on gradient truncation and has demonstrated versatility across a wide range of tasks -- that can achieve neural data similarity comparable to Backpropagation Through Time (BPTT) when matched for task accuracy. Our findings also reveal that model architecture and initial conditions can play a more significant role in determining neural similarity than the specific learning rule. Furthermore, we observe that BPTT-trained models and their biologically plausible counterparts exhibit similar dynamical properties at comparable accuracies. These results underscore the substantial progress made in developing biologically plausible learning rules, highlighting their potential to achieve both competitive task performance and neural data similarity.