cd /news/artificial-intelligence/some-memories-must-fade-persistence-… · home › topics › artificial-intelligence › article
[ARTICLE · art-173048] src=arxiv.org ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

Some Memories Must Fade: Persistence and Decay of User Memories in Model Parameters

A new arXiv paper (2611.25235v1) proposes a parametric memory system that stores user facts directly in model weights while letting transient information decay, built on a fixed-capacity LoRA adapter using k-hot rank combinations with independently controlled temporal decay and a lightweight learned controller. On a controlled dataset of long-horizon user trajectories with changing belief states, the policy reached 86% of oracle performance by balancing recall and abstention, whereas continual-learning and RAG baselines defaulted to one behavior. The authors describe it as the first system to manage memory directly in model weights, aimed at long-horizon parametric memory.

recommendation

When building long-horizon user memory, store facts in a fixed-capacity LoRA adapter with per-fact decay and train a controller to decide recall versus abstention; the source reports 86% of oracle performance.

more in playbook: Fine-tuning and data
by read1 min views1 publishedNov 20, 2026

arXiv:2611.25235v1 Announce Type: new Abstract: Parametric memory allows Large Language Models to internalize knowledge directly into their weights, offering a potentially powerful alternative to retrieval-based memory (RAG). However, existing approaches to parametric memory are difficult to deploy as they largely mirror the Continual Learning (CL) objective of retaining information indefinitely. This is neither sustainable nor sufficient for a memory system, which must govern knowledge across its lifecycle by retaining stable facts, reinforcing recurring observations, updating changed beliefs, and allowing transient information to fade. We propose a system for parametric memory management consisting of a memory substrate (handling information storage and forgetting) and a learned policy (what to remember, how deeply, and for how long). Our substrate is a fixed-capacity LoRA adapter that consists of k-hot rank combinations which store facts without catastrophic interference and maintain an independently controlled temporal decay, enabling outdated information to fade while long-term facts persist. On this substrate, we train a lightweight controller that governs each incoming user event to ensure the memory system reflects the user's true, updated belief state. To evaluate how memory systems manage user memory, we introduce a controlled dataset of long-horizon user trajectories with varying persistence of facts based on changing belief states and user-behavioral shifts, and probe queries that test recall while information is valid and abstention after it expires. Our policy learns to effectively manage user memory over its entire lifecycle, balancing recall and abstention to achieve 86% of oracle performance, unlike CL and RAG baselines that primarily default to one of the two behaviors. By providing the first system to manage memory directly in model weights, this work establishes a foundation for long-horizon parametric memory.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @arxiv 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/some-memories-must-f…] indexed:0 read:1min 2026-11-20 · —