{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,8]],"date-time":"2026-06-08T20:05:27Z","timestamp":1780949127509,"version":"3.54.1"},"reference-count":78,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"7","license":[{"start":{"date-parts":[[2026,7,1]],"date-time":"2026-07-01T00:00:00Z","timestamp":1782864000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2026,7,1]],"date-time":"2026-07-01T00:00:00Z","timestamp":1782864000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,7,1]],"date-time":"2026-07-01T00:00:00Z","timestamp":1782864000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"name":"Hong Kong Research Grant Council-General Research Fund Scheme","award":["17202422"],"award-info":[{"award-number":["17202422"]}]},{"name":"Hong Kong Research Grant Council-General Research Fund Scheme","award":["17212923"],"award-info":[{"award-number":["17212923"]}]},{"name":"Hong Kong Research Grant Council-General Research Fund Scheme","award":["17215025"],"award-info":[{"award-number":["17215025"]}]},{"name":"Theme-based Research","award":["T45-701\/22-R"],"award-info":[{"award-number":["T45-701\/22-R"]}]},{"name":"Strategic Topics","award":["STG3\/E-605\/25-N"],"award-info":[{"award-number":["STG3\/E-605\/25-N"]}]},{"name":"JC STEM Lab of Robotics for Soft Materials"},{"name":"Hong Kong Jockey Club Charities Trust"},{"DOI":"10.13039\/501100001809","name":"NSFC","doi-asserted-by":"crossref","award":["62306046"],"award-info":[{"award-number":["62306046"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"crossref"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Pattern Anal. Mach. Intell."],"published-print":{"date-parts":[[2026,7]]},"DOI":"10.1109\/tpami.2026.3670471","type":"journal-article","created":{"date-parts":[[2026,3,4]],"date-time":"2026-03-04T20:49:46Z","timestamp":1772657386000},"page":"8378-8394","source":"Crossref","is-referenced-by-count":1,"title":["Understanding Data Influence With Differential Approximation"],"prefix":"10.1109","volume":"48","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-6721-2468","authenticated-orcid":false,"given":"Haoru","family":"Tan","sequence":"first","affiliation":[{"name":"Department of Electrical and Electronic Engineering, University of Hong Kong, Pokfulam, Hong Kong"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sitong","family":"Wu","sequence":"additional","affiliation":[{"name":"Chinese University of Hong Kong, Sha Tin, Hong Kong"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiuzhe","family":"Wu","sequence":"additional","affiliation":[{"name":"Stanford University, Stanford, CA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0001-7114-2546","authenticated-orcid":false,"given":"Wang","family":"Wang","sequence":"additional","affiliation":[{"name":"University of Hong Kong, Pokfulam, Hong Kong"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-2732-8618","authenticated-orcid":false,"given":"Bo","family":"Zhao","sequence":"additional","affiliation":[{"name":"Shanghai Jiao Tong University, Shanghai, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4766-435X","authenticated-orcid":false,"given":"Zeke","family":"Xie","sequence":"additional","affiliation":[{"name":"Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-7660-6090","authenticated-orcid":false,"given":"Gui-Song","family":"Xia","sequence":"additional","affiliation":[{"name":"School of Artificial Intelligence, Wuhan University, Wuhan, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-4285-1626","authenticated-orcid":false,"given":"Xiaojuan","family":"Qi","sequence":"additional","affiliation":[{"name":"Department of Electrical and Electronic Engineering, University of Hong Kong, Pokfulam, Hong Kong"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","article-title":"Language models are few-shot learners","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Brown"},{"key":"ref2","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV51070.2023.00371"},{"key":"ref3","article-title":"DataInf: Efficiently estimating data influence in loRA-tuned LLMs and diffusion models","author":"Kwon","year":"2023"},{"key":"ref4","article-title":"Dataset pruning: Reducing training data by examining generalization influence","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Yang"},{"key":"ref5","article-title":"LESS: Selecting influential data for targeted instruction tuning","author":"Xia","year":"2024"},{"key":"ref6","article-title":"Studying large language model generalization with influence functions","author":"Grosse","year":"2023"},{"key":"ref7","article-title":"Training data attribution for diffusion models","author":"Dai","year":"2023"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1201\/chmonstaapp"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1111\/j.2517-6161.1986.tb01398.x"},{"key":"ref10","article-title":"Understanding black-box predictions via influence functions","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Koh"},{"key":"ref11","article-title":"On second-order group influence functions for black-box predictions","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Basu"},{"key":"ref12","article-title":"On the accuracy of influence functions for measuring group effects","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Koh"},{"key":"ref13","article-title":"Influence functions in deep learning are fragile","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Basu"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.02483"},{"key":"ref15","article-title":"Estimating training data influence by tracing gradient descent","author":"Pruthi","year":"2020"},{"key":"ref16","article-title":"Capturing the temporal dependence of training data influence","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Wang"},{"key":"ref17","first-page":"18251","article-title":"Data pruning via moving-one-sample-out","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Tan"},{"key":"ref18","first-page":"4213","article-title":"Data cleansing for models trained with SGD","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Hara"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1162\/neco.1994.6.1.147"},{"key":"ref20","doi-asserted-by":"crossref","DOI":"10.52202\/075280-0258","article-title":"GEX: A flexible method for approximating influence via geometric ensemble","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Kim"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v36i8.20791"},{"key":"ref22","first-page":"19523","article-title":"Beyond neural scaling laws: Beating power law scaling via data pruning","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Sorscher"},{"key":"ref23","article-title":"Knowledge removal in sampling-based Bayesian inference","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Fu"},{"key":"ref24","article-title":"The LLAMA 3 herd of models","author":"Grattafiori","year":"2024"},{"key":"ref25","article-title":"Training verifiers to solve math word problems","author":"Cobbe","year":"2021"},{"key":"ref26","article-title":"Moderate coreset: A universal method of data selection for real-world data-efficient deep learning","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Xia"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1007\/s10994-023-06495-7"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.1515\/9781400881970-018"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.52202\/068431-0948"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1007\/s11263-015-0816-y"},{"key":"ref31","article-title":"LAION-5B: An open large-scale dataset for training next generation image-text models","volume-title":"Proc. 36th Conf. Neural Inf. Process. Syst. Datasets Benchmarks Track","author":"Schuhmann"},{"key":"ref32","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR46437.2021.00356"},{"key":"ref33","first-page":"2881","article-title":"What neural networks memorize and why: Discovering the long tail via influence estimation","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Feldman"},{"key":"ref34","first-page":"192","article-title":"The loss surfaces of multilayer networks","volume":"38","author":"Choromanska","year":"2015","journal-title":"J. Mach. Learn. Res."},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.5555\/2969033.2969154"},{"key":"ref36","article-title":"Revisiting inverse Hessian vector products for calculating influence functions","author":"Klochkov","year":"2024"},{"key":"ref37","article-title":"Revisit, extend, and enhance Hessian-free influence functions","author":"Yang","year":"2024"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.52202\/068431-1305"},{"key":"ref39","article-title":"\u201cWhat data benefits my classifier?\u201d Enhancing model performance and interpretability through influence-based data selection","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Chhabra"},{"key":"ref40","article-title":"Adam: A method for stochastic optimization","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Kingma"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-022-31384-3"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-662-48986-4_301626"},{"key":"ref43","article-title":"Datamodels: Predicting predictions from training data","author":"Ilyas","year":"2022"},{"key":"ref44","article-title":"Achieving fairness at no utility cost via data reweighing with influence","author":"Li","year":"2022"},{"key":"ref45","first-page":"7081","article-title":"HYDRA: Hypergradient data relevance analysis for interpreting deep neural networks","volume-title":"Proc. AAAI Conf. Artif. Intell.","author":"Chen","year":"2022"},{"key":"ref46","first-page":"2242","article-title":"Data shapley: Equitable valuation of data for machine learning","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Ghorbani"},{"key":"ref47","doi-asserted-by":"publisher","DOI":"10.14778\/3342263.3342637"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR46437.2021.00814"},{"key":"ref49","article-title":"What is your data worth to GPT? LLM-scale data valuation with influence functions","author":"Choe","year":"2024"},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-021-87762-2"},{"key":"ref51","article-title":"Outlier gradient analysis: Efficiently identifying detrimental training samples for deep learning models","author":"Chhabra","year":"2024"},{"key":"ref52","article-title":"Resolving training biases via influence-based data relabeling","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Kong"},{"key":"ref53","doi-asserted-by":"publisher","DOI":"10.1145\/3366423.3380072"},{"key":"ref54","first-page":"27075","article-title":"Exploring example influence in continual learning","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Sun"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV48922.2021.00914"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.2307\/j.ctvcm4g18.8"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"ref58","first-page":"2488","article-title":"How does batch normalization help optimization?","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Santurkar"},{"key":"ref59","first-page":"6391","article-title":"Visualizing the loss landscape of neural nets","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Li"},{"key":"ref60","article-title":"Learning multiple layers of features from tiny images","author":"Krizhevsky","year":"2009"},{"key":"ref61","article-title":"Tiny imageNet visual recognition challenge","author":"Le","year":"2015","journal-title":"CS 231N"},{"key":"ref62","article-title":"Learning transferable visual models from natural language supervision","author":"Radford","year":"2021"},{"key":"ref63","article-title":"Automated cleanup of the ImageNet dataset by model consensus, explainability and confident learning","author":"Kert\u00e9sz","year":"2021"},{"key":"ref64","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR42600.2020.00932"},{"key":"ref65","first-page":"3832","article-title":"Certified data removal from machine learning models","volume-title":"Proc. 37th Int. Conf. Mach. Learn.","author":"Guo"},{"key":"ref66","article-title":"SSSE: Efficiently erasing samples from trained machine learning models","author":"Peste","year":"2021"},{"key":"ref67","article-title":"DivideMix: Learning with noisy labels as semi-supervised learning","author":"Li","year":"2020"},{"key":"ref68","article-title":"Early-learning regularization prevents memorization of noisy labels","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Liu"},{"key":"ref69","doi-asserted-by":"publisher","DOI":"10.1145\/3664647.3680664"},{"key":"ref70","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2015.7298885"},{"key":"ref71","article-title":"WebVision database: Visual learning and understanding from web data","author":"Li","year":"2017"},{"key":"ref72","first-page":"5907","article-title":"SELFIE: Refurbishing unclean samples for robust deep learning","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Song"},{"key":"ref73","article-title":"Active learning for convolutional neural networks: A coreset approach","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Sener"},{"key":"ref74","article-title":"BLIP: Bootstrapping language-image pre-training for unified vision-language understanding and generation","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Li"},{"key":"ref75","article-title":"BLIP-2: Bootstrapping language-image pre-training with frozen image encoders and large language models","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Li"},{"key":"ref76","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2015.303"},{"key":"ref77","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00745"},{"key":"ref78","article-title":"EfficientNet: Rethinking model scaling for convolutional neural networks","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Tan"}],"container-title":["IEEE Transactions on Pattern Analysis and Machine Intelligence"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/34\/11552636\/11421676.pdf?arnumber=11421676","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,8]],"date-time":"2026-06-08T19:51:33Z","timestamp":1780948293000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11421676\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,7]]},"references-count":78,"journal-issue":{"issue":"7"},"URL":"https:\/\/doi.org\/10.1109\/tpami.2026.3670471","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":[[2026,7]]}}}