Nature Communications · 6 November 2026 · From issue 63
Redundancy maximization as a principle of associative memory learning in Hopfield networks
In a computer model, training memory units to agree raised storage capacity more than tenfold.
Mark Blümel, Andreas C. Schneider et al.
A memory system inspired by the brain can store far more patterns when its parts are trained to agree with each other. These systems, called Hopfield networks, work like a brain's memory. They recall a full pattern from a partial clue. Scientists checked how each tiny unit combines two signals: a clue from outside and signals from other units. When units were trained to make those two signals overlap, the network's storage limit jumped from 0.14 to 1.59 times its number of units.
The new memory storage limit, up from 0.14 in standard networks.
Computer simulation only, no real neurons or devices tested.
Integrity screen: passed (3 checks) Checked 9 November 2026. Retraction record: none. DOI resolves at doi.org. Metadata record found (Nature Communications). Read the source
This screen checks process signals. It does not tell you the findings are true.