pith. sign in

arxiv: 2605.03804 · v2 · pith:LVEOBLVNnew · submitted 2026-05-05 · 💻 cs.AI

ScrapMem: A Bio-inspired Framework for On-device Personalized Agent Memory via Optical Forgetting

Jiale Chang , Yuxiang Ren This is my paper

Pith reviewed 2026-07-01 00:06 UTC · model grok-4.3

classification 💻 cs.AI
keywords multimodal agent memoryoptical forgettingepisodic memory graphon-device storageLLM long-term memorystorage compressionpersonalized agentsbio-inspired memory
0
0 comments X

The pith

ScrapMem applies optical forgetting to lower the resolution of older memories inside a causal-temporal graph, making long-term multimodal agent memory feasible on edge devices.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents ScrapMem as a way to handle the high storage costs and complexity of keeping long-term personalized memory for multimodal LLM agents running on resource-limited devices. It stores experiences as Scrapbook Pages and uses optical forgetting to reduce detail in older entries over time. An Episodic Memory Graph organizes events into causal and temporal links to hold semantic consistency even after compression. Experiments on the ATM-Bench dataset test whether this combination delivers better task performance and recall than earlier methods while cutting memory use sharply. A reader would care because the approach targets the practical barrier that currently prevents agents from retaining personal history locally without constant external storage.

Core claim

ScrapMem integrates multimodal data into Scrapbook Pages and introduces Optical Forgetting, which progressively reduces the resolution of older memories to lower storage cost while suppressing low-value details, and constructs an Episodic Memory Graph to organize key events into a causal-temporal structure that maintains semantic consistency.

What carries the argument

Optical Forgetting, the mechanism that progressively reduces resolution of older memories, paired with the Episodic Memory Graph that structures events causally and temporally.

If this is right

  • Storage requirements for long-term agent memory drop substantially while joint task accuracy reaches new benchmark levels.
  • Recall on memory-dependent queries improves through the structured aggregation provided by the graph.
  • Multimodal data can be retained on-device for extended periods without requiring constant external resources.
  • The combination of compression and graph organization supports efficient handling of complex personal histories.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same resolution-reduction approach could be tested on sequential personal data streams such as sensor logs or interaction histories outside agent settings.
  • If the graph maintains consistency after compression, the method might scale to shared memories across multiple agents.
  • Hardware implementations that mimic the progressive detail loss could be explored for even lower power memory modules.

Load-bearing premise

Progressively lowering the resolution of older memories removes low-value details without damaging the semantic consistency that the episodic graph is meant to preserve.

What would settle it

A controlled test that measures semantic accuracy or task performance on recall queries before and after applying resolution reduction to the same set of older memories, where accuracy falls below baseline levels.

Figures

Figures reproduced from arXiv: 2605.03804 by Jiale Chang, Yuxiang Ren.

Figure 1
Figure 1. Figure 1: Comparison between human memory (CLS theory) and Scrapbook Memory. Top: The hippocampus rapidly encodes multimodal episodic experiences, while the neocortex gradually consolidates them into stable long-term knowledge. Bottom: ScrapMem similarly binds heterogeneous user data into scrapbook pages and progressively compresses old memories via optical for￾getting, preserving core semantics for efficient retrie… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the ScrapMem. (1) Consolidation and Perception: Unifies heterogeneous records (images, videos, text) into hybrid representations via OCR and vision-to-text extraction. (2) EM-Graph Construction: Organizes nodes into an Episodic Memory Graph with event-centric paths (EM-Paths) for structured retrieval and multi-hop reasoning. (3) Optical Forgetting: Compresses outdated memories through temporal … view at source ↗
Figure 3
Figure 3. Figure 3: Retrieval performance (Recall@K) under varying optical forgetting intensities. The clustering of different forgetting curves demonstrates that ScrapMem is highly robust to specific hyperparameter configura￾tions. quality (Q), resolution scaling factor (S), and tem￾poral stage boundaries (T) for Recent, Mid-term, and Old memories, respectively view at source ↗
Figure 4
Figure 4. Figure 4: Storage–performance trade-off on ATM￾Bench (Joint@10). The x-axis uses a logarithmic scale. ScrapMem (Timed-Gentle, orange star) reduces stor￾age by 93.0% relative to the raw-data baseline while retaining over 90% of SOTA performance (46.3% vs. 51.0%). The Pareto frontier indicates strong efficiency and graceful performance degradation, supporting on￾device deployment. strengthen long-range reasoning. Exte… view at source ↗
Figure 5
Figure 5. Figure 5: Prompt for entity node extraction. 2 Query Node Extraction System Prompt You extract semantic memory nodes from a user’s question so we can match them against a memory graph. Return JSON ONLY with key "semantic_nodes" (array of 3-10 concise phrases). Focus on entities (people, places, brands, products), objects, actions, events, and concrete time mentions. Use lowercase, short phrases, no stop words. User … view at source ↗
Figure 6
Figure 6. Figure 6: Prompt for query-side semantic node extraction. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Two-stage prompts for scrapbook page perception and path consolidation. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: EM-Graph failure: vague query with very old evidence. [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: EM-Graph failure: correct event, wrong day. [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: EM-Graph failure: numerical answer from wrong event. [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Optical Forgetting: date lost under aggressive old-bucket compression. [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Optical Forgetting: time-of-day lost, date preserved. [PITH_FULL_IMAGE:figures/full_fig_p017_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Optical Forgetting: partial evidence loss on list-recall. [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Optical Forgetting: coarse location preserved, fine-grained address lost. [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: LLM failure: cannot extract fine-grained visual detail despite full retrieval. [PITH_FULL_IMAGE:figures/full_fig_p018_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: LLM failure: confuses similar entity types across documents. [PITH_FULL_IMAGE:figures/full_fig_p018_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: LLM failure: multi-hop spatial reasoning breakdown. [PITH_FULL_IMAGE:figures/full_fig_p018_17.png] view at source ↗
read the original abstract

Long-term personalized memory for LLM agents is challenging on resource-limited edge devices due to high storage costs and multimodal complexity. To address this, we propose ScrapMem, a framework that integrates multimodal data into "Scrapbook Page." ScrapMem introduces Optical Forgetting, an optical compression mechanism that progressively reduces the resolution of older memories, lowering storage cost while suppressing low-value details. To maintain semantic consistency, we construct an Episodic Memory Graph (EM-Graph) that organizes key events into a causal-temporal structure. Extensive experiments on the multimodal ATM-Bench showcase that ScrapMem provides three main benefits: (1) strong performance, achieving a new state-of-the-art with a 51.0% Joint@10 score; (2) high storage efficiency, reducing memory usage by up to 93% via optical forgetting; and (3) improved recall, increasing Recall@10 to 70.3% through structured aggregation. ScrapMem offers an effective and storage-efficient solution for on-device long-term memory in multimodal LLM agents.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 0 minor

Summary. The paper proposes ScrapMem, a framework for long-term personalized memory in multimodal LLM agents on edge devices. It integrates data into 'Scrapbook Page' structures, applies 'Optical Forgetting' to progressively lower resolution of older memories for up to 93% storage reduction while suppressing low-value details, and uses an Episodic Memory Graph (EM-Graph) to organize events in a causal-temporal structure for semantic consistency. On the ATM-Bench dataset, it claims new SOTA results of 51.0% Joint@10, 70.3% Recall@10 via structured aggregation, and high storage efficiency.

Significance. If the performance, efficiency, and consistency claims hold with proper validation, the work could address a practically important challenge in on-device agent memory by providing a bio-inspired compression approach that trades resolution for storage without (claimed) loss of utility. The combination of optical-style forgetting with graph-structured episodic memory is a novel angle worth exploring for resource-constrained multimodal agents.

major comments (3)
  1. [Abstract] Abstract: The headline claims (51.0% Joint@10 SOTA, 93% memory reduction, 70.3% Recall@10) are presented with no experimental setup, baselines, error bars, dataset splits, or implementation description of Optical Forgetting or EM-Graph construction. This makes it impossible to assess whether the numbers can be attributed to the proposed mechanisms rather than unstated factors.
  2. [Abstract] Abstract: The central premise that Optical Forgetting 'suppresses low-value details' while the EM-Graph 'maintains semantic consistency' is stated but unsupported by any ablation, consistency metric, or comparison of full-resolution vs. forgotten memory on identical queries. Without such evidence the efficiency and recall claims rest on an untested assumption.
  3. [Abstract] Abstract: No equations, algorithms, or pseudocode are supplied for Optical Forgetting (resolution reduction schedule, optical model) or EM-Graph (node/edge definitions, causal-temporal construction, aggregation method), despite these being load-bearing for all three claimed benefits.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed feedback on the abstract. While abstracts are necessarily concise, we address each comment by clarifying where the supporting details appear in the full manuscript and indicate our willingness to revise the abstract for improved transparency.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The headline claims (51.0% Joint@10 SOTA, 93% memory reduction, 70.3% Recall@10) are presented with no experimental setup, baselines, error bars, dataset splits, or implementation description of Optical Forgetting or EM-Graph construction. This makes it impossible to assess whether the numbers can be attributed to the proposed mechanisms rather than unstated factors.

    Authors: The abstract is a high-level summary; the full experimental setup, baselines, error bars, dataset splits on ATM-Bench, and implementation details for Optical Forgetting and EM-Graph are provided in Sections 4 and 5. We will revise the abstract to include a brief reference to the evaluation protocol and main baselines. revision: partial

  2. Referee: [Abstract] Abstract: The central premise that Optical Forgetting 'suppresses low-value details' while the EM-Graph 'maintains semantic consistency' is stated but unsupported by any ablation, consistency metric, or comparison of full-resolution vs. forgotten memory on identical queries. Without such evidence the efficiency and recall claims rest on an untested assumption.

    Authors: Section 5.3 presents ablation studies comparing full-resolution vs. forgotten memory on identical queries, along with consistency metrics and the contribution of the EM-Graph. We will revise the abstract to reference these supporting experiments. revision: partial

  3. Referee: [Abstract] Abstract: No equations, algorithms, or pseudocode are supplied for Optical Forgetting (resolution reduction schedule, optical model) or EM-Graph (node/edge definitions, causal-temporal construction, aggregation method), despite these being load-bearing for all three claimed benefits.

    Authors: Section 3 provides the equations for the Optical Forgetting resolution schedule and optical model, plus formal definitions and construction details for the EM-Graph; pseudocode appears in the appendix. We will revise the abstract to briefly note the core mechanisms. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical claims rest on reported experiments without derivations or self-referential reductions.

full rationale

The provided abstract and structure contain no equations, parameter fittings, uniqueness theorems, or self-citations that could create circularity. Performance, efficiency, and recall figures are presented as experimental outcomes on ATM-Bench rather than derived by construction from inputs or prior author work. The framework description (Optical Forgetting, EM-Graph) is introduced as a proposal without load-bearing self-referential steps, making the derivation chain self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 3 invented entities

Abstract introduces three new named constructs without external grounding or prior citations; all rest on unstated assumptions about multimodal integration and compression preserving semantics.

axioms (2)
  • domain assumption Multimodal data can be integrated into Scrapbook Pages without loss of essential structure
    Stated as the basic representation step in the framework.
  • ad hoc to paper Optical forgetting can reduce resolution while the EM-Graph maintains semantic consistency
    Central unproven premise required for the claimed storage-recall tradeoff.
invented entities (3)
  • Optical Forgetting no independent evidence
    purpose: Progressively reduces resolution of older memories to lower storage cost
    Newly named compression mechanism with no independent evidence supplied.
  • Episodic Memory Graph (EM-Graph) no independent evidence
    purpose: Organizes key events into causal-temporal structure for semantic consistency
    Newly proposed graph structure without external validation.
  • Scrapbook Page no independent evidence
    purpose: Container for integrating multimodal data
    Basic representational unit introduced by the framework.

pith-pipeline@v0.9.1-grok · 5708 in / 1442 out tokens · 45665 ms · 2026-07-01T00:06:40.219329+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 26 canonical work pages · 10 internal anchors

  1. [1]

    Sina Abdollahi, Mohammad M Maheri, Javad Forough, Amir Al Sadi, Josh Millar, David Kotz, Marios Kogias, and Hamed Haddadi. 2026. Agentee: Confidential llm agent execution on edge devices. In Proceedings of the Sixth European Workshop on Machine Learning and Systems, pages 473--480

    2026

  2. [2]

    Shuai Bai, Yuqi Cai, Ruoyi Chen, Kai Chen, Xu Chen, Zhihao Cheng, and 1 others. 2025 a . Qwen3-vl technical report. arXiv preprint arXiv:2511.21631

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, and 1 others. 2025 b . Qwen2.5-vl technical report. arXiv preprint arXiv:2502.13923

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  4. [4]

    TING BAI, JIAYANG FAN, XIAOSHUAI WEN, JIAZHENG KANG, HENGZHI LAN, RUOCHEN ZHAO, PINGZHENG WU, ZEPENG ZHANG, YUTIAN ZHONG, GEZI LI, and 1 others. 2025. Survey on ai memory: Theories, taxonomies, evaluations, and emerging trends

    2025

  5. [5]

    Ting Bai, Le Huang, Yue Yu, Cheng Yang, Cheng Hou, Zhe Zhao, and Chuan Shi. 2025. https://doi.org/10.1145/3736403 Efficient multi-task prompt tuning for recommendation . ACM Transactions on Information Systems, 43(4)

    work page doi:10.1145/3736403 2025

  6. [6]

    Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George Bm Van Den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, and 1 others. 2022. Improving language models by retrieving from trillions of tokens. In International conference on machine learning, pages 2206--2240. PMLR

    2022

  7. [7]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, and 1 others. 2020. Language models are few-shot learners. In Advances in Neural Information Processing Systems (NeurIPS), volume 33, pages 1877--1901

    2020

  8. [8]

    Dongqi Cai, Shangguang Wang, Chen Peng, Zeling Zhang, and Mengwei Xu. 2024. Recall: Empowering multimodal embedding for edge devices. arXiv preprint arXiv:2409.15342

    work page arXiv 2024

  9. [9]

    Jialin Chen, Ziyu Zhao, Gaukhar Nurbek, Aosong Feng, Ali Maatouk, Leandros Tassiulas, Yifeng Gao, and Rex Ying. 2026. Trace: Grounding time series in context for multimodal embedding and retrieval. Advances in Neural Information Processing Systems, 38:2440--2472

    2026

  10. [10]

    Zhe Chen, Weiyun Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Erfei Cui, Jinguo Zhu, Shenglong Ye, Hao Tian, Zhaoyang Liu, and 1 others. 2024. Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling. arXiv preprint arXiv:2412.05271

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  11. [11]

    Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Yadav. 2025. Mem0: Building production-ready ai agents with scalable long-term memory. arXiv preprint arXiv:2504.19413

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  12. [12]

    Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, Dasha Metropolitansky, Robert Osazuwa Ness, and Jonathan Larson. 2024. From local to global: A graph rag approach to query-focused summarization. arXiv preprint arXiv:2404.16130

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  13. [13]

    Lang Feng, Fuchao Yang, Feng Chen, Xin Cheng, Haiyang Xu, Zhenglin Wan, Ming Yan, and Bo An. 2026. Agentocr: Reimagining agent history via optical self-compression. arXiv preprint arXiv:2601.04786

    work page arXiv 2026

  14. [14]

    Yicheng Fu, Raviteja Anantha, and Jianpeng Cheng. 2025. https://doi.org/10.18653/v1/2025.realm-1.3 Camphor: Collaborative agents for multi-input planning and high-order reasoning on device . In Proceedings of the 1st Workshop for Research on Agent Language Models (REALM 2025), pages 33--51, Vienna, Austria. Association for Computational Linguistics

    work page doi:10.18653/v1/2025.realm-1.3 2025

  15. [15]

    Bernal J Guti \'e rrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su. 2024. Hipporag: Neurobiologically inspired long-term memory for large language models. Advances in neural information processing systems, 37:59532--59569

    2024

  16. [16]

    Bernal Jim \'e nez Guti \'e rrez, Yiheng Shu, Weijian Qi, Sizhe Zhou, and Yu Su. 2025. From rag to memory: Non-parametric continual learning for large language models. arXiv preprint arXiv:2502.14802

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  17. [17]

    Canny, and Ian Fischer

    Kuang-Huei Lee, Xinyun Chen, Hiroki Furuta, John F. Canny, and Ian Fischer. 2024. https://proceedings.mlr.press/v235/lee24c.html A human-inspired reading agent with gist memory of very long contexts . In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, pages 22928--22951. PMLR

    2024

  18. [18]

    u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt \

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K \"u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt \"a schel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 33:9459--9474

    2020

  19. [19]

    Jinze Li, Yang Zhang, Xin Yang, Jiayi Qu, Jinfeng Xu, Shuo Yang, Junhua Ding, and Edith Cheuk-Han Ngai. 2026. Ocr-memory: Optical context retrieval for long-horizon agent memory. arXiv preprint arXiv:2604.26622

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  20. [20]

    Junming Liu, Yifei Sun, Weihua Cheng, Haodong Lei, Yirong Chen, Licheng Wen, Xuemeng Yang, Daocheng Fu, Pinlong Cai, Nianchen Deng, Yi Yu, Shuyue Hu, Botian Shi, and Ding Wang. 2025. https://api.semanticscholar.org/CorpusID:283466545 Memverse: Multimodal memory for lifelong learning agents . ArXiv, abs/2512.03627

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  21. [21]

    Miao Lu, Weiwei Sun, Weihua Du, Zhan Ling, Xuesong Yao, Kang Liu, and Jiecao Chen. 2025. https://doi.org/10.48550/arXiv.2510.06727 Scaling llm multi-turn rl with end-to-end summarization-based context management . arXiv preprint arXiv:2510.06727

    work page doi:10.48550/arxiv.2510.06727 2025

  22. [22]

    Yun Luo, Zhen Yang, Fandong Meng, Yafu Li, Jie Zhou, and Yue Zhang. 2025. An empirical study of catastrophic forgetting in large language models during continual fine-tuning. IEEE Transactions on Audio, Speech and Language Processing

    2025

  23. [23]

    Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. 2024. https://doi.org/10.18653/v1/2024.acl-long.747 Evaluating very long-term conversational memory of llm agents . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13851--13870

    work page doi:10.18653/v1/2024.acl-long.747 2024

  24. [24]

    James L McClelland, Bruce L McNaughton, and Randall C O'Reilly. 1995. Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychological review, 102(3):419

    1995

  25. [25]

    Jingbiao Mei, Jinghong Chen, Guangyu Yang, Xinyu Hou, Margaret Li, and Bill Byrne. 2026. According to me: Long-term personalized referential memory qa. arXiv preprint arXiv:2603.01990

    work page arXiv 2026

  26. [26]

    Charles Packer, Vivian Fang, Shishir\_G Patil, Kevin Lin, Sarah Wooders, and Joseph\_E Gonzalez. 2023. Memgpt: towards llms as operating systems

    2023

  27. [27]

    Dipkumar Patel

    Joon Sung Park, Joseph O'Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2023. https://doi.org/10.1145/3586183.3606763 Generative agents: Interactive simulacra of human behavior . In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST), pages 1--22

    work page doi:10.1145/3586183.3606763 2023

  28. [28]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 3982--3992

    2019

  29. [29]

    Yaorui Shi, Shugui Liu, Yu Yang, Wenyu Mao, Yuxin Chen, Qi Gu, Hui Su, Xunliang Cai, Xiang Wang, and An Zhang. 2026. Memocr: Layout-aware visual memory for efficient long-horizon reasoning. arXiv preprint arXiv:2601.21468

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  30. [30]

    Weinan Sun, Madhu Advani, Nelson Spruston, Andrew Saxe, and James E Fitzgerald. 2023. Organizing memories for generalization in complementary learning systems. Nature neuroscience, 26(8):1438--1448

    2023

  31. [31]

    Cheng Tan, Jingxuan Wei, Linzhuang Sun, Zhangyang Gao, Siyuan Li, Bihui Yu, Ruifeng Guo, and Stan Z Li. 2024. Retrieval meets reasoning: Even high-school textbook knowledge benefits multimodal reasoning. arXiv preprint arXiv:2405.20834

    work page arXiv 2024

  32. [32]

    Mamatha Thota, Dewei Yi, and Georgios Leontidis. 2023. Lleda—lifelong self-supervised domain adaptation. Knowledge-Based Systems, 279:110959

    2023

  33. [33]

    Bin Wang, Fan Wang, Pingping Wang, Jinyu Cong, Yang Yu, Yilong Yin, Zhongyi Han, and Benzheng Wei. 2026. Agentic unlearning: When llm agent meets machine unlearning. arXiv preprint arXiv:2602.17692

    work page arXiv 2026

  34. [34]

    Haoran Wei, Yaofeng Sun, and Yukun Li. 2025. Deepseek-ocr: Contexts optical compression. arXiv preprint arXiv:2510.18234

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  35. [35]

    Biao Wu, Yanda Li, Zhiwei Zhang, Yunchao Wei, Meng Fang, and Ling Chen. 2024. https://doi.org/10.48550/arXiv.2411.02006 Foundations and recent trends in multimodal mobile agents: A survey . arXiv preprint arXiv:2411.02006

    work page doi:10.48550/arxiv.2411.02006 2024

  36. [36]

    Jiajun Xu, Zhiyuan Li, Wei Chen, Qun Wang, Xin Gao, Qi Cai, and Ziyuan Ling. 2024. On-device language models: A comprehensive review. arXiv preprint arXiv:2409.00088

    work page arXiv 2024

  37. [37]

    Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. 2026. A-mem: Agentic memory for llm agents. Advances in Neural Information Processing Systems, 38:17577--17604

    2026

  38. [38]

    Shengyuan Ye, Bei Ouyang, Tianyi Qian, Liekang Zeng, Mu Yuan, Xiaowen Chu, Weijie Hong, and Xu Chen. 2025. Venus: An efficient edge memory-and-retrieval system for vlm-based online video understanding. arXiv preprint arXiv:2512.07344

    work page arXiv 2025

  39. [39]

    Zhongkai Yu, Shengwen Liang, Tianyun Ma, Yunke Cai, Ziyuan Nan, Di Huang, Xinkai Song, Yifan Hao, Jie Zhang, Tian Zhi, Yongwei Zhao, Zidong Du, Xing Hu, Qi Guo, and Tianshi Chen. 2024. https://doi.org/10.1109/MICRO61859.2024.00108 Cambricon-llm: A chiplet-based hybrid architecture for on-device inference of 70b llm . In 2024 57th IEEE/ACM International Sy...

    work page doi:10.1109/micro61859.2024.00108 2024

  40. [40]

    Delong Zeng, Yuexiang Xie, Yaliang Li, and Ying Shen. 2025. Enhancing multimodal retrieval via complementary information extraction and alignment. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 22092--22105

    2025

  41. [41]

    Zeyu Zhang, Rui Li, Xiaoyan Zhao, Yang Zhang, Wenjie Wang, Xu Chen, and Tat-Seng Chua. 2026. Nextmem: Towards latent factual memory for llm-based agents. arXiv preprint arXiv:2603.15634

    work page arXiv 2026

  42. [42]

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, and Zheyan Luo. 2024. https://doi.org/10.18653/v1/2024.acl-demos.38 Llamafactory: Unified efficient fine-tuning of 100+ language models . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), pages 400--410

    work page doi:10.18653/v1/2024.acl-demos.38 2024

  43. [43]

    Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. 2024. Memorybank: Enhancing large language models with long-term memory. In Proceedings of the AAAI conference on artificial intelligence, volume 38, pages 19724--19731

    2024

  44. [44]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  45. [45]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...