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

2025-06-07

AI Summary

Whether biologically plausible learning rules—exemplified by e-prop—can achieve neural activity dynamics comparable to backpropagation through time (BPTT) while satisfying biological constraints remains unresolved. Method: We quantified geometric alignment between model neural trajectories and empirical neural recordings using Procrustes analysis, under strict task-performance matching across models. Systematic evaluation controlled for architecture, initialization, and learning rule. Results: First, model architecture and initial weights exert significantly greater influence on neural trajectory similarity than the choice of learning rule. Second, when matched for task accuracy, e-prop produces dynamical trajectories nearly indistinguishable from those of BPTT—demonstrating that biologically plausible rules can simultaneously approximate both functional performance and neurobiological fidelity of gradient-based methods. This provides critical empirical support for the biological plausibility of spiking neural networks.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Bio-inspired Learning
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Social networks and social learning
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

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.

Problem

• Assessing if biologically plausible learning rules match BPTT in neural similarity
• Evaluating e-prop's performance and alignment with neural recordings
• Exploring how model architecture affects neural similarity more than learning rules

Innovation

• E-prop uses gradient truncation for learning
• Procrustes analysis compares neural similarity
• Model architecture impacts neural similarity