AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

MLP Memory: A Retriever-Pretrained Memory for Large Language Models

Rubin Wei, Jiaqi Cao, Jiarui Wang, Jushi Kai, Qipeng Guo, Bowen Zhou, Zhouhan Lin

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Retrieval-augmented reasoning

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
add MLP Memory (interpolated) to the base LLMMistral-7B-v0.3 · Five QA benchmarksaverage score: 32.12 -> 36.06Improved
add MLP Memory to the base LLMLlama2-7B · Five QA benchmarksaverage score: 32.81 -> 35.38Improved
continued pretraining or LoRA on Wikipedia instead of MLP MemoryLlama2-7B · Five QA benchmarksaverage score: 32.81 -> 29.66 (CPT), 31.34 (LoRA)Hurt
use RAG (top-5 documents)Mistral-7B-v0.3 · WebQAexact match: 29.28 -> 24.90Hurt
add MLP MemoryMistral-7B-v0.3 · Nine general NLP tasksaverage accuracy: 67.86 -> 73.07Improved
add MLP MemoryMistral-7B-v0.3 · HaluEval dialogue, QA, summarizationaccuracy: +9.68, +10.08, +2.14Improved
apply CPT to the base modelMistral-7B-v0.3 · HaluEval dialogueaccuracy: 57.18 -> 51.68Hurt
use MLP Memory architecture instead of continued training of decoder-only GPT-2GPT-2 family · WikiText-103 scalingpower-law exponent: -0.143 -> -0.168Improved
use MLP Memory architecture instead of continued training of decoder-only GPT-2GPT-2 family · Web dataset scalingpower-law exponent: -0.216 -> -0.268Improved
vary KL/CE loss weight alpha away from 0.4 (to 0.0 or 1.0) (ablation)MLP Memory training · WikiText-103 (GPT-2 small/medium/large)perplexityHurt
attach MLP Memory at about 70% network depth instead of the final layer (ablation)MLP Memory · WikiText-103 (GPT-2 small/medium/large)perplexityImproved
raise interpolation weight lambda to 0.9 (ablation)MLP Memory · HaluEval dialogueaccuracy: 64.07 (lambda 0.45) -> 49.71 (lambda 0.90)Hurt

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