Equilibrium Propagation for Non-Conservative Systems
Antonino Emanuele Scurria, Dimitri Vanden Abeele, Bortolo Matteo Mognetti, Serge Massar · 2026
Read Equilibrium Propagation for Non-Conservative Systems by Antonino Emanuele Scurria, Dimitri Vanden Abeele, Bortolo Matteo Mognetti, Serge Massar (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03670.txt.
Description
Equilibrium propagation trains physical networks by comparing two settled states, but it only worked for systems with an energy function. Antonino Emanuele Scurria and 3 co-authors lift that restriction in a 2026 paper. Non-conservative systems, whose interactions are not reciprocal, cover many real devices and even feedforward networks, and earlier generalizations could not compute the exact gradient.
The framework keeps learning from stationary states but adds a correction term, scaled by the interaction's non-reciprocal part, to the learning-phase dynamics. That change recovers the cost function's exact gradient. A variational version over an enlarged state space yields the same algorithm. In numerical tests it learns faster and performs better than previous schemes.
Catalog record
| CREATOR | Antonino Emanuele Scurria, Dimitri Vanden Abeele, Bortolo Matteo Mognetti, Serge Massar |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03670