AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

MemoryLLM: Plug-n-Play Interpretable Feed-Forward Memory for Transformers

AJAY KUMAR JAISWAL, Lauren Hannah, Han-Byul Kim, Duc N.M Hoang, Arnav Kundu, Mehrdad Farajtabar, Minsik Cho

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Attention head and block redesign · 0 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
train FFNs on context-free token embeddings (MemoryLLM) instead of dense FFNs, compared at equal total paramsdense base 750M · C4perplexity: 19.730 -> 22.079Hurt
MemoryLLM vs dense model at matched active parametersdense base 250M · C4perplexity: 23.190 -> 22.079Improved
move beta*h^2 FFN params from memory to FFN compute (Flex-MemoryLLM-3h2)MemoryLLM 1B · LAMBADAaccuracy: 36.13 -> 44.19Improved
move beta*h^2 FFN params from memory to FFN compute (Flex-MemoryLLM-3h2)MemoryLLM 1B · HellaSwagaccuracy: 37.24 -> 41.64Improved
reduce FFN contribution alpha from 1.0 to 0.5 (ablation)MemoryLLM-1B · LAMBADAaccuracy: 0.3613 -> 0.1086Hurt
reduce FFN contribution alpha from 1.0 to 0.5 (ablation)MemoryLLM-1B · Wikitext-2perplexity: 24.348 -> 53.7072Hurt
reduce FFN contribution alpha from 1.0 to 0.5 (ablation)MemoryLLM-1B · BoolQaccuracy: 0.6205 -> 0.6131No clear effect
reduce FFN contribution alpha from 1.0 to 0.5 (ablation)MemoryLLM-1B · PIQAaccuracy: 0.7018 -> 0.6801No clear effect
reduce FFN contribution (alpha) in MemoryLLM versus conventional base (ablation)MemoryLLM vs dense base · Wikitext-2, SciQ, PIQA, HellaSwagperformance changeImproved
add MemoryLLM/Flex-MemoryLLM design instead of pruning a dense model (ablation)dense base 1B and 750M · C4perplexityImproved
use independent layer norm per FFNMemoryLLM · training convergenceconvergenceImproved

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