Hugging Face Daily PapersJingxuan Wu, Yuzhe Yang, Yiqiao HuangSep 29, 20261 min readpaperadvanced

MemFold: Learning Compact Soft Memory for Long-Context Personalization via On-Policy Optimization

Summary

MemFold learns a fixed‑size soft memory of K vectors for each user by optimizing the downstream task behavior via on‑policy rewards and a confidence‑gated teacher distillation. It improves accuracy on long‑context persona benchmarks and transfers to other evaluation sets without extra training.

Compresses query‑conditioned user history into K continuous vectors that serve as the model's memory interface.
Training combines group‑relative reward signals with on‑policy distillation from a frozen textual‑memory teacher, avoiding extra decoding at inference.
Achieves state‑of‑the‑art accuracy on PersonaMem‑32K/128K, with larger margins as history length grows.
Ablations show the reward term drives most gains; the teacher signal adds a smaller but consistent boost.

Engineers building long‑context LLM assistants need efficient, task‑aligned memory mechanisms to retain user preferences without exploding token length.

6/10

llm reinforcement learning evaluation memory personalization

Related reading

Continual Learning Mechanisms Compose for Long-Horizon Memorization arxiv.org
The paper studies long‑horizon memorization where a language model must learn 100 tasks via continual fine‑tuning without retaining data or task IDs. By composing data, function, and weight anchors with merged LoRA, they boost final retention from 1.2% to 34.9%, a 28× improvement.
Hugging Face Daily Papersarxiv.orgZheyuan Zhang, Alvin Zhang, Daniel Khashabi29d1 minpaperHN2
FactorEngram: Factorized N-gram Memory with Basis-Level Gating for Language Models arxiv.org
FactorEngram proposes a factorized n-gram memory with basis-level contextual gating for LLMs. This design allows polysemous patterns to selectively retrieve relevant memory components from a shared dictionary, leading to improved language modeling and downstream task performance.
Hugging Face Daily Papersarxiv.orgBowen Yang, Jingbo Zhou, Qinghong Miao8d1 minpaper
Do LLMs Have the Memory of a Goldfish?bytebytego.com
The article explains that LLMs don’t have persistent personal memory; all “memory” is supplied by the surrounding application via the context window, summaries, or external storage. It outlines the distinction between trained weights, working‑memory (token context), and persistent application memory, shows how to construct API calls to preserve conversation state, and discusses the cost and laten…
ByteByteGobytebytego.com20d12 min
MoME: Mixture-of-Memory Embeddings for Context-Aware Sparse Lookup arxiv.org
The paper proposes Mixture of Memory Embeddings (MoME), a context‑aware sparse lookup that replaces each token’s single memory row with a gated mixture of multiple slots. Experiments on Llama‑3, MobileLLM and Qwen3 show MoME outperforms existing memory‑embedding baselines at equal parameter and FLOP budgets and exhibits interpretable routing for polysemous tokens.
Hugging Face Daily Papersarxiv.orgMuchen Li, Leonid Sigal, Renjie Liao22d1 minpaper
Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents arxiv.org
Jev‑Mem splits agentic memory work between a fast System‑One controller (handling typing, organization, routing, budgeting, graph traversal, scoring, and stopping) and a slower System‑Two reasoning plane, cutting memory‑construction time 6.6×, query latency 36.7% and boosting LLM‑as‑a‑Judge score 11% on LoCoMo.
Hugging Face Daily Papersarxiv.orgDongming Jiang, Yi Li, Bingzhe Li15d1 minpaper
APM-Bench: Benchmarking Cross-session Persistent Memory for Egocentric Streaming Video Assistants arxiv.org
APM-Bench is a new benchmark for evaluating persistent memory in egocentric streaming video assistants across intermittent sessions. It reveals a significant utility-latency-storage trade-off, showing current models struggle with long-term recall, low overhead, and proactive assistance simultaneously.
Hugging Face Daily Papersarxiv.orgJianguo Huang, Jinming Liu, Qiyao Wang7d1 minpaper

MemFold: Learning Compact Soft Memory for Long-Context Personalization via On-Policy Optimization

Summary

Related reading

Continual Learning Mechanisms Compose for Long-Horizon Memorizationarxiv.org

FactorEngram: Factorized N-gram Memory with Basis-Level Gating for Language Modelsarxiv.org

Do LLMs Have the Memory of a Goldfish?bytebytego.com

MoME: Mixture-of-Memory Embeddings for Context-Aware Sparse Lookuparxiv.org

Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agentsarxiv.org

APM-Bench: Benchmarking Cross-session Persistent Memory for Egocentric Streaming Video Assistantsarxiv.org

Continual Learning Mechanisms Compose for Long-Horizon Memorization arxiv.org

FactorEngram: Factorized N-gram Memory with Basis-Level Gating for Language Models arxiv.org

MoME: Mixture-of-Memory Embeddings for Context-Aware Sparse Lookup arxiv.org

Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents arxiv.org

APM-Bench: Benchmarking Cross-session Persistent Memory for Egocentric Streaming Video Assistants arxiv.org