Construction of a Neural Network with Temperature-Dependent Recall Patterns
Munetaka Sasaki · 2026
Read Construction of a Neural Network with Temperature-Dependent Recall Patterns by Munetaka Sasaki (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/openalex-w7163924565.txt.
Description
Munetaka Sasaki, of Kanagawa University in Japan, published a 2026 study building a neural network that remembers different patterns at different temperatures. Neural networks of the Hopfield type store patterns and recall them from partial cues. Sasaki stored one pattern in a fully connected network, where every unit links to every other, and a second pattern in a sparse network with fewer links.
Because the fully connected pattern resists heat-driven noise better, tuning the two patterns' relative weights lets temperature select which one is recalled. Computer simulations confirmed the switch and indicated it happens as a sharp, first-order phase transition. Annealing runs also showed a catch: if the energy barrier is too high, the sparse pattern fails to return at low temperature in the time available.
Catalog record
| CREATOR | Munetaka Sasaki |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | OpenAlex / Journal of the Physical Society of Japan |