![[2602.03546] How to Train Your Resistive Network: Generalized Equilibrium Propagation and Analytical Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.03546] How to Train Your Resistive Network: Generalized Equilibrium Propagation and Analytical Learning
Summary
This paper presents a novel algorithm for training resistive networks using Generalized Equilibrium Propagation, aiming to enhance energy efficiency in machine learning applications.
Why It Matters
The research addresses the pressing need for energy-efficient computing solutions in machine learning, particularly through analog systems. By developing a method to train resistive networks, the authors contribute to advancing sustainable AI technologies, which is crucial as digital hardware becomes increasingly energy-intensive.
Key Takeaways
- Introduces Generalized Equilibrium Propagation for resistive networks.
- Demonstrates a method to calculate gradients using Kirchhoff's laws.
- Shows that training can occur at the output layer without needing full network readouts.
- Proposes that only a subset of resistance values can be updated with minimal performance loss.
- Highlights the potential for analog computing to reduce energy consumption in machine learning.
Computer Science > Machine Learning arXiv:2602.03546 (cs) [Submitted on 3 Feb 2026 (v1), last revised 14 Feb 2026 (this version, v3)] Title:How to Train Your Resistive Network: Generalized Equilibrium Propagation and Analytical Learning Authors:Jonathan Lin, Aman Desai, Frank Barrows, Francesco Caravelli View a PDF of the paper titled How to Train Your Resistive Network: Generalized Equilibrium Propagation and Analytical Learning, by Jonathan Lin and 3 other authors View PDF HTML (experimental) Abstract:Machine learning is a powerful method of extracting meaning from data; unfortunately, current digital hardware is extremely energy-intensive. There is interest in an alternative analog computing implementation that could match the performance of traditional machine learning while being significantly more energy-efficient. However, it remains unclear how to train such analog computing systems while adhering to locality constraints imposed by the physical (as opposed to digital) nature of these systems. Local learning algorithms such as Equilibrium Propagation and Coupled Learning have been proposed to address this issue. In this paper, we develop an algorithm to exactly calculate gradients using a graph theoretic and analytical framework for Kirchhoff's laws. We also introduce Generalized Equilibrium Propagation, a framework encompassing a broad class of Hebbian learning algorithms, including Coupled Learning and Equilibrium Propagation, and show how our algorithm compares. ...