AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

A Biologically Plausible Dense Associative Memory with Exponential Capacity

Mohadeseh Shafiei Kafraj, Dmitry Krotov, Peter E. Latham

ICLR 2026 · Architectures and efficiency › RNNs and normalization · approach: Hebbian and predictive coding · 5 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace winner-take-all nonlinearity with Heaviside thresholdKrotov-Hopfield two-layer DAM · MNIST, N_h=50stored memories recalled: 50 -> 60,000 memories (12%-90% -> 98% recall)Improved
store CIFAR-10 images with threshold DAM (N_h=500)threshold DAM · CIFAR-10unique stable minima: 49982 of 50,000Improved
classify from recalled hidden vs visible vs original on CIFAR-10threshold DAM · CIFAR-10accuracy: 88% -> 56% (visible), 40% (hidden)Hurt
classify from recalled hidden patterns vs original on MNISTthreshold DAM · MNISTaccuracy: 99% -> 95% (hidden), 98% (visible)No clear effect
set threshold to theoretical optimum 0.5 (ablation)threshold DAM · capacity vs N_h (random weights)capacityImproved
use smooth sigmoid instead of Heaviside (ablation)threshold DAM · capacityrecallNo clear effect
use asymmetric weights and heterogeneous thresholds (ablation)threshold DAM · MNIST and CIFAR-10 recallrecallNo clear effect
ratio tau_v/tau_h of only 4 (ablation)threshold DAM · capacity, N_h=10capacityImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy