공대생 도전 일지

MEMENTO: Teaching LLMs to Manage Their Own Context

https://arxiv.org/abs/2604.09852 MEMENTO: Teaching LLMs to Manage Their Own ContextReasoning models think in long, unstructured streams with no mechanism for compressing or organizing their own intermediate state. We introduce MEMENTO: a method that teaches models to segment reasoning into blocks, compress each block into a memento, i.e.arxiv.org 마소 논문입니다. https://github.com/microsoft/memento Gi..

https://aclanthology.org/2021.findings-emnlp.161/ Span Pointer Networks for Non-Autoregressive Task-Oriented Semantic ParsingAkshat Shrivastava, Pierce Chuang, Arun Babu, Shrey Desai, Abhinav Arora, Alexander Zotov, Ahmed Aly. Findings of the Association for Computational Linguistics: EMNLP 2021. 2021.aclanthology.org21년 emnlp findings 였네요. 입력 문장을 다시 생성하지 말고 스팬의 처음 끝 위치만 출력하도록 진행했습니다. 여기서 디코더는 A..

https://arxiv.org/abs/2309.03057 Hide and Seek (HaS): A Lightweight Framework for Prompt Privacy ProtectionNumerous companies have started offering services based on large language models (LLM), such as ChatGPT, which inevitably raises privacy concerns as users' prompts are exposed to the model provider. Previous research on secure reasoning using multi-party carxiv.org 엄청 옛날 논문인데 LLM서버에 민감한 개인 ..

https://arxiv.org/abs/2412.12928 Truthful Text Sanitization Guided by Inference AttacksText sanitization aims to rewrite parts of a document to prevent disclosure of personal information. The central challenge of text sanitization is to strike a balance between privacy protection (avoiding the leakage of personal information) and utility prearxiv.org후배가 전해준 논문인데 이 것도 prompt의 의미를 지키면서 프라이버시는 제거해 ..

https://arxiv.org/abs/2602.20743 Adaptive Text Anonymization: Learning Privacy-Utility Trade-offs via Prompt OptimizationAnonymizing textual documents is a highly context-sensitive problem: the appropriate balance between privacy protection and utility preservation varies with the data domain, privacy objectives, and downstream application. However, existing anonymization mearxiv.org ACL 2026 fi..

https://aclanthology.org/2026.acl-long.4/ Your Inference Request Will Become a Black Box: Confidential Inference for Cloud-based Large Language ModelsChung-ju Huang, Huiqiang Zhao, Yuanpeng He, Lijian Li, Wenpin Jiao, Zhi Jin, Peixuan Chen, Leye Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.aclanthology.org2026 ACL에 붙었..

https://aclanthology.org/2024.findings-naacl.171/ LatticeGen: Hiding Generated Text in a Lattice for Privacy-Aware Large Language Model Generation on CloudMengke Zhang, Tianxing He, Tianle Wang, Lu Mi, Niloofar Mireshghallah, Binyi Chen, Hao Wang, Yulia Tsvetkov. Findings of the Association for Computational Linguistics: NAACL 2024. 2024.aclanthology.org 24 NAACL이네요. 생각보다 오래된 논문입니다. 기존 클라우드 서비스 ..

https://aclanthology.org/2025.findings-acl.197/ RemoteRAG: A Privacy-Preserving LLM Cloud RAG ServiceYihang Cheng, Lan Zhang, Junyang Wang, Mu Yuan, Yunhao Yao. Findings of the Association for Computational Linguistics: ACL 2025. 2025.aclanthology.org Cloud RAG에서 사용자의 쿼리를 숨기면서 완전한 암호화 방식처럼 엄청난 계산 비용을 지불하지 않고 정확한 검색을 할 수 있는가 라는 문제를 해결하는 것이 핵심입니다. 기존 쿼리 임베딩을 클라우드에 보내는 것이 아닌 DP로 교란한 임베딩을 보내 후보 문서 집..

https://arxiv.org/abs/2410.11366 LargePiG: Your Large Language Model is Secretly a Pointer GeneratorRecent research on query generation has focused on using Large Language Models (LLMs), which despite bringing state-of-the-art performance, also introduce issues with hallucinations in the generated queries. In this work, we introduce relevance hallucinatiarxiv.org기존 LLM을 학습 없이 Pointer Generator처럼..

https://aclanthology.org/2024.findings-emnlp.983/ PositionID: LLMs can Control Lengths, Copy and Paste with Explicit Positional AwarenessNoah Wang, Feiyu Duan, Yibo Zhang, Wangchunshu Zhou, Ke Xu, Wenhao Huang, Jie Fu. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024.aclanthology.org EMNLP 2024 findings네요 llm이 생성 중에 현재 위치를 명시적으로 추적하지 못하는 것을 positional awareness 부족으로 해석..