Related Experiment Video
Updated: Jun 27, 2025
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
9.1K
Is Learning in Biological Neural Networks Based on Stochastic Gradient Descent? An Analysis Using Stochastic
Sören Christensen1, Jan Kallsen2
1Department of Mathematics, Kiel Universiy, 24118 Kiel, Germany christensen@math.uni-kiel.de.
Neural Computation
|April 26, 2024
Summary
Learning in biological neural networks (BNNs) may use stochastic gradient descent. Our study shows that many local updates during learning opportunities approximate a gradient step, suggesting BNNs might optimize like artificial neural networks.
Area of Science:
- Computational Neuroscience
- Machine Learning Theory
Background:
- A key debate in neuroscience and AI concerns whether biological neural networks (BNNs) learn using local information only, precluding gradient-based optimization.
- Artificial neural networks (ANNs) commonly employ stochastic gradient descent (SGD) for efficient learning.
Purpose of the Study:
- To investigate the theoretical possibility of gradient-based learning in BNNs.
- To analyze a stochastic model of supervised learning in BNNs.
Main Methods:
- Development of a stochastic model for supervised learning in BNNs.
- Mathematical analysis of the conditions under which local updates approximate a continuous gradient step.
Main Results:
- Demonstration that a continuous gradient step is approximated when numerous local updates process each learning opportunity.
- Identification of a potential mechanism for gradient-based optimization within BNNs.
Conclusions:
- The findings suggest that stochastic gradient descent may indeed play a role in the optimization of biological neural networks.
- This bridges the gap between theoretical arguments against and potential biological plausibility of gradient-based learning in the brain.
