Reshaping reservoirs with unsupervised Hebbian adaptation
Tanguy Cazalets, Joni Dambre · 2025
Read Reshaping reservoirs with unsupervised Hebbian adaptation by Tanguy Cazalets, Joni Dambre (2025) 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/pmc-12800270.txt.
Description
Tanguy Cazalets and Joni Dambre published this paper in 2025. It improves reservoir computing, a lightweight method for time-series tasks where a fixed random network echoes inputs and only a simple readout is trained. The random wiring is the weakness: it was never chosen for the task at hand.
Their rule, Hebbian Architecture Generation, grows the reservoir instead. Starting nearly empty, it adds connections between neurons that often fire together, the classic Hebbian slogan, without any labels or gradient steps. Across classification and forecasting tasks, the self-wired reservoirs beat standard echo state networks and reservoirs tuned by rival plasticity rules. The method keeps reservoir computing's cheap training while shedding its arbitrariness. The authors frame structural self-organization, the brain's own trick, as a practical route to robust time-series learning on modest hardware.
Catalog record
| CREATOR | Tanguy Cazalets, Joni Dambre |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 98 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |