Episode
$δ$-mem: Efficient Online Memory for Large Language Models
- Podcast
- Daily Paper Cast
- Published
- May 14, 2026
- Duration seconds
- 1471
- Processing state
not_requested- Canonical source
- https://share.transistor.fm/s/979c7e38
Actions
POST https://stenobird.com/v1/public/podcasts/daily-paper-cast-7079649/episodes/mem-efficient-online-memory-for-large-language-models/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/daily-paper-cast-7079649/mem-efficient-online-memory-for-large-language-models.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
🤗 Upvotes: 90 | cs.AI Authors: Jingdi Lei, Di Zhang, Junxian Li, Weida Wang, Kaixuan Fan, Xiang Liu, Qihan Liu, Xiaoteng Ma, Baian Chen, Soujanya Poria Title: $δ$-mem: Efficient Online Memory for Large Language Models Arxiv: http://arxiv.org/abs/2605.12357v1 Abstract: Large language models increasingly need to accumulate and reuse historical information in long-term assistants and agent systems. Simply expanding the context window is costly and often fails to ensure effective context utilization. We propose $δ$-mem, a lightweight memory mechanism that augments a frozen full-attention backbone with a compact online state of associative memory. $δ$-mem compresses past information into a fixed-size state matrix updated by delta-rule learning, and uses its readout to generate low-rank corrections to the backbone's attention computation during generation. With only an $8\times8$ online memory state, $δ$-mem improves the average score to $1.10\times$ that of the frozen backbone and $1.15\times$ that of the strongest non-$δ$-mem memory baseline. It achieves larger gains on memory-heavy benchmarks, reaching $1.31\times$ on MemoryAgentBench and $1.20\times$ on LoCoMo, while largely preserving general capabilities. These results show that effective memory can be realized through a compact online state directly coupled with attention computation, without full fine-tuning, backbone replacement, or explicit context extension.