{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,11,17]],"date-time":"2026-11-17T12:18:20Z","timestamp":1794917900604,"version":"4.9.1"},"publisher-location":"California","reference-count":0,"publisher":"International Joint Conferences on Artificial Intelligence Organization","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,9]]},"abstract":"<jats:p>Current Parameter-Efficient Fine-Tuning (PEFT) methods typically operate under an implicit assumption: once a target module is selected, every token passing through it contributes equally to the downstream task and requires a parameter update.  In this paper, we challenge this convention by revealing a pervasive token-level redundancy in the fine-tuning of large models (LMs). We propose TS-PEFT, a theoretical framework utilizing proximal optimization that acts as a dynamic probe to identify token-level redundancy during the fine-tuning process.  Extensive experiments demonstrate that indiscriminately updating all tokens is not only computationally superfluous but often introduces optimization noise.  Surprisingly, by discarding 30%-70% of token updates, TS-PEFT consistently matches or exceeds the performance of dense baselines such as LoRA, DoRA. Our in-depth analysis shows that the learned token-level sparsity is a superior indicator of module importance compared to traditional weight criteria, providing a novel data-driven perspective on the intrinsic adaptation mechanism of LMs.<\/jats:p>","DOI":"10.24963\/ijcai.2026\/514","type":"proceedings-article","created":{"date-parts":[[2026,9,16]],"date-time":"2026-09-16T22:01:13Z","timestamp":1789596073000},"page":"4616-4624","source":"Crossref","is-referenced-by-count":0,"title":["TS-PEFT: Unveiling Token-Level Redundancy in Parameter-Efficient Fine-Tuning"],"prefix":"10.24963","author":[{"given":"Dabiao","family":"Ma","sequence":"first","affiliation":[{"name":"Qifu Technology"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ziming","family":"Dai","sequence":"additional","affiliation":[{"name":"Tianjin University"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhimin","family":"Xin","sequence":"additional","affiliation":[{"name":"Qifu Technology"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shu","family":"Wang","sequence":"additional","affiliation":[{"name":"Qifu Technology"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jian","family":"Yang","sequence":"additional","affiliation":[{"name":"Qifu Technology"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Haojun","family":"Fei","sequence":"additional","affiliation":[{"name":"Qifu Technology"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"10584","event":{"name":"Thirty-Fifth International Joint Conference on Artificial Intelligence {IJCAI-26}","theme":"Artificial Intelligence","location":"Bremen, Germany","acronym":"IJCAI-2026","number":"35","sponsor":["International Joint Conferences on Artificial Intelligence Organization (IJCAI)"],"start":{"date-parts":[[2026,8,15]]},"end":{"date-parts":[[2026,8,21]]}},"container-title":["Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence"],"original-title":[],"deposited":{"date-parts":[[2026,11,17]],"date-time":"2026-11-17T11:30:15Z","timestamp":1794915015000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.ijcai.org\/proceedings\/2026\/514"}},"subtitle":[],"proceedings-subject":"Artificial Intelligence Research Articles","short-title":[],"issued":{"date-parts":[[2026,9]]},"references-count":0,"URL":"https:\/\/doi.org\/10.24963\/ijcai.2026\/514","relation":{},"subject":[],"published":{"date-parts":[[2026,9]]}}}