Hopfield storage capacity via Hoeffding concentration on pattern overlaps

Layer 1 — Physicsin the neuromorphic-computing subtree

Hopfield associative-memory capacity via Hoeffding concentration. Setup: an N-neuron Hopfield network storing m random binary patterns {xi_mu_i} in a Hebbian weight matrix w_ij = (1/N) * sum_mu xi_mu_i * xi_mu_j. Retrieval at pattern mu is…

Related concepts

Hoeffding's inequality
Hopfield pattern-overlap concentration: P(|m-E[m]|>=1/2) <= 2*exp(-N/2)

Explore Hopfield storage capacity via Hoeffding concentration on pattern overlaps on the interactive knowledge graph →