{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,2,10]],"date-time":"2026-02-10T16:06:05Z","timestamp":1770739565641,"version":"3.49.0"},"reference-count":84,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"3","license":[{"start":{"date-parts":[[2026,3,1]],"date-time":"2026-03-01T00:00:00Z","timestamp":1772323200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2026,3,1]],"date-time":"2026-03-01T00:00:00Z","timestamp":1772323200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,3,1]],"date-time":"2026-03-01T00:00:00Z","timestamp":1772323200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"name":"NTURSR"},{"name":"StartUpGrants"},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62225113"],"award-info":[{"award-number":["62225113"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["U23A20318"],"award-info":[{"award-number":["U23A20318"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62576083"],"award-info":[{"award-number":["62576083"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62025604"],"award-info":[{"award-number":["62025604"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62576364"],"award-info":[{"award-number":["62576364"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62276195"],"award-info":[{"award-number":["62276195"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100017607","name":"Shenzhen Municipal Fundamental Research Program","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100017607","id-type":"DOI","asserted-by":"publisher"}]},{"name":"Natural Science Foundation"},{"name":"Basic Research Key Project","award":["JCYJ20241202124430041"],"award-info":[{"award-number":["JCYJ20241202124430041"]}]},{"DOI":"10.13039\/501100012556","name":"Foundation for Innovative Research Groups of Hubei Province","doi-asserted-by":"publisher","award":["2024AFA017"],"award-info":[{"award-number":["2024AFA017"]}],"id":[{"id":"10.13039\/501100012556","id-type":"DOI","asserted-by":"publisher"}]},{"name":"Science and Technology Major Project of Hubei Province","award":["2024BAB046"],"award-info":[{"award-number":["2024BAB046"]}]},{"name":"Open Research Fund from Guangdong Laboratory of Artificial Intelligence and Digital Economy","award":["GML-KF-24-23"],"award-info":[{"award-number":["GML-KF-24-23"]}]},{"name":"NTURSR"},{"name":"StartUpGrants"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Pattern Anal. Mach. Intell."],"published-print":{"date-parts":[[2026,3]]},"DOI":"10.1109\/tpami.2025.3629605","type":"journal-article","created":{"date-parts":[[2025,11,6]],"date-time":"2025-11-06T18:53:01Z","timestamp":1762455181000},"page":"2329-2341","source":"Crossref","is-referenced-by-count":0,"title":["Efficient and Effective Weight-Ensembling Mixture of Experts for Multi-Task Model Merging"],"prefix":"10.1109","volume":"48","author":[{"ORCID":"https:\/\/orcid.org\/0000-0001-5659-3464","authenticated-orcid":false,"given":"Li","family":"Shen","sequence":"first","affiliation":[{"name":"Shenzhen Campus of Sun Yat-sen University, Shenzhen, China"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0576-8153","authenticated-orcid":false,"given":"Anke","family":"Tang","sequence":"additional","affiliation":[{"name":"Wuhan University, Wuhan, China"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-5419-5286","authenticated-orcid":false,"given":"Enneng","family":"Yang","sequence":"additional","affiliation":[{"name":"Shenzhen Campus of Sun Yat-sen University, Shenzhen, China"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-1709-5056","authenticated-orcid":false,"given":"Guibing","family":"Guo","sequence":"additional","affiliation":[{"name":"Northeastern University, Shenyang, China"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-2296-6370","authenticated-orcid":false,"given":"Yong","family":"Luo","sequence":"additional","affiliation":[{"name":"Wuhan University, Wuhan, China"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0542-2280","authenticated-orcid":false,"given":"Lefei","family":"Zhang","sequence":"additional","affiliation":[{"name":"Wuhan University, Wuhan, China"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-7141-708X","authenticated-orcid":false,"given":"Xiaochun","family":"Cao","sequence":"additional","affiliation":[{"name":"Shenzhen Campus of Sun Yat-sen University, Shenzhen, China"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0059-8458","authenticated-orcid":false,"given":"Bo","family":"Du","sequence":"additional","affiliation":[{"name":"Wuhan University, Wuhan, China"}]},{"given":"Dacheng","family":"Tao","sequence":"additional","affiliation":[{"name":"Nanyang Technological University, Singapore"}]}],"member":"263","reference":[{"key":"ref1","first-page":"47778","article-title":"Merging multi-task models via weight-ensembling mixture of experts","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Tang"},{"issue":"8","key":"ref2","article-title":"Language models are unsupervised multitask learners","volume":"1","author":"Radford","year":"2019","journal-title":"OpenAI Blog"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52688.2022.01553"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.1007\/s11633-022-1410-8"},{"issue":"70","key":"ref5","first-page":"1","article-title":"Scaling instruction-finetuned language models","volume":"25","author":"Chung","year":"2024","journal-title":"J. Mach. Learn. Res."},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.1038\/s42256-024-00961-0"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1007\/s11633-023-1416-x"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1109\/tnnls.2025.3628666"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1007\/s11633-022-1398-0"},{"key":"ref10","article-title":"A survey on model MoErging: Recycling and routing among specialized experts for collaborative learning","author":"Yadav","year":"2024"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1145\/3787849"},{"key":"ref12","article-title":"Averaging weights leads to wider optima and better generalization","volume-title":"Proc. Conf. Uncertainty Artif. Intell.","author":"Izmailov"},{"key":"ref13","first-page":"23965","article-title":"Model soups: Averaging weights of multiple fine-tuned models improves accuracy without increasing inference time","volume-title":"Proc. 39th Int. Conf. Mach. Learn.","author":"Wortsman"},{"key":"ref14","first-page":"1","article-title":"Editing models with task arithmetic","volume-title":"Proc. 11th Int. Conf. Learn. Representations","author":"Ilharco"},{"key":"ref15","first-page":"7093","article-title":"TIES-merging: Resolving interference when merging models","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Yadav"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.52202\/079017-2691"},{"key":"ref17","first-page":"17703","article-title":"Merging models with Fisher-weighted averaging","volume-title":"Proc. 36th Int. Conf. Neural Inf. Process. Syst.","author":"Matena"},{"key":"ref18","first-page":"1","article-title":"Dataless knowledge fusion by merging weights of language models","volume-title":"Proc. 11th Int. Conf. Learn. Representations","author":"Jin"},{"key":"ref19","first-page":"1","article-title":"AdaMerging: Adaptive model merging for multi-task learning","volume-title":"Proc. 12th Int. Conf. Learn. Representations","author":"Yang"},{"key":"ref20","article-title":"Concrete subspace learning based interference elimination for multi-task model fusion","author":"Tang","year":"2023"},{"key":"ref21","first-page":"57755","article-title":"Language models are super mario: Absorbing abilities from homologous models as a free lunch","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Yu"},{"key":"ref22","first-page":"50268","article-title":"Localizing task information for improved model merging and compression","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Wang"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.52202\/079017-3900"},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1706.03762"},{"key":"ref25","first-page":"1","article-title":"An image is worth 16 x 16 words: Transformers for image recognition at scale","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Dosovitskiy"},{"key":"ref26","first-page":"525","article-title":"Multi-task learning as multi-objective optimization","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Sener"},{"key":"ref27","article-title":"Distilling the knowledge in a neural network","author":"Hinton","year":"2015"},{"key":"ref28","first-page":"1","article-title":"Pruning neural networks at initialization: Why are we missing the mark?","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Frankle"},{"key":"ref29","article-title":"Understanding neural networks through deep visualization","author":"Yosinski","year":"2015"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1109\/WACV57701.2024.00194"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1007\/s11263-024-02181-w"},{"key":"ref32","first-page":"8748","article-title":"Learning transferable visual models from natural language supervision","volume-title":"Proc. 38th Int. Conf. Mach. Learn.","author":"Radford"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2010.5539970"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1109\/ICCVW.2013.77"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.1109\/JPROC.2017.2675998"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.1109\/IGARSS.2018.8519248"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.2118\/18761-MS"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.1016\/j.neunet.2012.02.016"},{"key":"ref39","doi-asserted-by":"publisher","DOI":"10.1109\/5.726791"},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2014.461"},{"key":"ref41","article-title":"Fusionbench: A comprehensive benchmark of deep model fusion","author":"Tang","year":"2024"},{"issue":"70","key":"ref42","first-page":"1","article-title":"Scaling instruction-finetuned language models","volume":"25","author":"Chung","year":"2024","journal-title":"J. Mach. Learn. Res."},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/W18-5446"},{"key":"ref44","first-page":"1","article-title":"Benchmarking neural network robustness to common corruptions and perturbations","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Hendrycks"},{"key":"ref45","first-page":"1","article-title":"Stop wasting my time! Saving days of ImageNet and BERT training with latest weight averaging","volume-title":"Proc. Has Trained Yet? Conf. Neural Inf. Process. Syst.","author":"Kaddour"},{"key":"ref46","first-page":"37713","article-title":"$\\pi$\u03c0-Tuning: Transferring multimodal foundation models with optimal multi-task interpolation","volume-title":"Proc. 40th Int. Conf. Mach. Learn.","author":"Wu"},{"key":"ref47","first-page":"56332","article-title":"Representation surgery for multi-task model merging","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Yang"},{"key":"ref48","first-page":"2790","article-title":"Parameter-efficient transfer learning for NLP","volume-title":"Proc. 36th Int. Conf. Mach. Learn.","author":"Houlsby"},{"key":"ref49","first-page":"1","article-title":"LoRA: Low-rank adaptation of large language models","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Hu"},{"key":"ref50","first-page":"1","article-title":"Parameter-efficient multi-task model fusion with partial linearization","volume-title":"Proc. 12th Int. Conf. Learn. Representations","author":"Tang"},{"key":"ref51","article-title":"SurgeryV2: Bridging the gap between model merging and multi-task learning with deep representation surgery","author":"Yang","year":"2024"},{"key":"ref52","first-page":"66727","article-title":"Task arithmetic in the tangent space: Improved editing of pre-trained models","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Ortiz-Jimenez"},{"key":"ref53","first-page":"22045","article-title":"Model fusion via optimal transport","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Singh"},{"key":"ref54","first-page":"1","article-title":"Topology and geometry of half-rectified network optimization","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Freeman"},{"key":"ref55","first-page":"11615","article-title":"Uniform convergence may be unable to explain generalization in deep learning","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Nagarajan"},{"key":"ref56","first-page":"1309","article-title":"Essentially no barriers in neural network energy landscape","volume-title":"Proc. 35th Int. Conf. Mach. Learn.","author":"Draxler"},{"key":"ref57","first-page":"3259","article-title":"Linear mode connectivity and the lottery ticket hypothesis","volume-title":"Proc. 37th Int. Conf. Mach. Learn.","author":"Frankle"},{"key":"ref58","first-page":"1","article-title":"The role of permutation invariance in linear mode connectivity of neural networks","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Entezari"},{"key":"ref59","first-page":"196","article-title":"Convergent learning: Do different neural networks learn the same representations?","volume-title":"Proc. 1st Int. Workshop Feature Extraction: Modern Questions Challenges NIPS 2015","author":"Li"},{"key":"ref60","first-page":"15300","article-title":"Optimizing mode connectivity via neuron alignment","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Tatro"},{"key":"ref61","first-page":"1","article-title":"ZipIt! merging models from different tasks without training","volume-title":"Proc. 12th Int. Conf. Learn. Representations","author":"Stoica"},{"key":"ref62","first-page":"13857","article-title":"Deep neural network fusion via graph matching with applications to model ensemble and federated learning","volume-title":"Proc. 39th Int. Conf. Mach. Learn.","author":"Liu"},{"key":"ref63","first-page":"1","article-title":"Git re-basin: Merging models modulo permutation symmetries","volume-title":"Proc. 12th Int. Conf. Learn. Representations","author":"Ainsworth"},{"key":"ref64","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.emnlp-main.102"},{"key":"ref65","doi-asserted-by":"publisher","DOI":"10.1038\/s42256-024-00975-8"},{"key":"ref66","first-page":"1","article-title":"Model merging by uncertainty-based gradient matching","volume-title":"Proc. 12th Int. Conf. Learn. Representations","author":"Daheim"},{"key":"ref67","first-page":"1","article-title":"Sparse model soups: A recipe for improved pruning via model averaging","volume-title":"Proc. 12th Int. Conf. Learn. Representations","author":"Zimmer"},{"key":"ref68","first-page":"1","article-title":"Localize-and-stitch: Efficient model merging via sparse task arithmetic","author":"He","year":"2024","journal-title":"Trans. Mach. Learn. Res."},{"key":"ref69","article-title":"Della-merging: Reducing interference in model merging through magnitude-based sampling","author":"Deep","year":"2024"},{"key":"ref70","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-73226-3_16"},{"key":"ref71","first-page":"27011","article-title":"Task-specific skill localization in fine-tuned language models","volume-title":"Proc. 40th Int. Conf. Mach. Learn.","author":"Panigrahi"},{"key":"ref72","first-page":"78905","article-title":"Twin-merging: Dynamic integration of modular expertise in model merging","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Lu"},{"key":"ref73","article-title":"Soft merging of experts with adaptive routing","author":"Muqeeth","year":"2023"},{"key":"ref74","first-page":"1","article-title":"Sparse upcycling: Training mixture-of-experts from dense checkpoints","volume-title":"Proc. 11th Int. Conf. Learn. Representations","author":"Komatsuzaki"},{"key":"ref75","article-title":"Merging vision transformers from different tasks and domains","author":"Ye","year":"2023"},{"key":"ref76","first-page":"36829","article-title":"Learning to route among specialized experts for zero-shot generalization","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Muqeeth"},{"key":"ref77","doi-asserted-by":"publisher","DOI":"10.1162\/neco.1991.3.1.79"},{"key":"ref78","article-title":"Mixtral of experts","author":"Jiang","year":"2024"},{"key":"ref79","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.acl-long.70"},{"key":"ref80","first-page":"7103","article-title":"Mixture-of-experts with expert choice routing","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Zhou"},{"issue":"120","key":"ref81","first-page":"1","article-title":"Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity","volume":"23","author":"Fedus","year":"2022","journal-title":"J. Mach. Learn. Res."},{"key":"ref82","first-page":"6265","article-title":"BASE layers: Simplifying training of large, sparse models","volume-title":"Proc. 38th Int. Conf. Mach. Learn.","author":"Lewis"},{"key":"ref83","article-title":"A review of sparse expert models in deep learning","author":"Fedus","year":"2022"},{"key":"ref84","first-page":"3320","article-title":"How transferable are features in deep neural networks?","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Yosinski"}],"container-title":["IEEE Transactions on Pattern Analysis and Machine Intelligence"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/34\/11372200\/11230232.pdf?arnumber=11230232","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,2,9]],"date-time":"2026-02-09T21:06:06Z","timestamp":1770671166000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11230232\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,3]]},"references-count":84,"journal-issue":{"issue":"3"},"URL":"https:\/\/doi.org\/10.1109\/tpami.2025.3629605","relation":{},"ISSN":["0162-8828","2160-9292","1939-3539"],"issn-type":[{"value":"0162-8828","type":"print"},{"value":"2160-9292","type":"electronic"},{"value":"1939-3539","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,3]]}}}