{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,17]],"date-time":"2026-07-17T03:28:41Z","timestamp":1784258921640,"version":"3.55.0"},"reference-count":71,"publisher":"IEEE","license":[{"start":{"date-parts":[[2025,10,19]],"date-time":"2025-10-19T00:00:00Z","timestamp":1760832000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,10,19]],"date-time":"2025-10-19T00:00:00Z","timestamp":1760832000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2025,10,19]]},"DOI":"10.1109\/iccv51701.2025.00339","type":"proceedings-article","created":{"date-parts":[[2026,4,29]],"date-time":"2026-04-29T19:45:49Z","timestamp":1777491949000},"page":"3552-3563","source":"Crossref","is-referenced-by-count":4,"title":["Learning to Inference Adaptively for Multimodal Large Language Models"],"prefix":"10.1109","author":[{"given":"Zhuoyan","family":"Xu","sequence":"first","affiliation":[{"name":"University of Wisconsin-Madison"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Khoi Duc","family":"Nguyen","sequence":"additional","affiliation":[{"name":"University of Wisconsin-Madison"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Preeti","family":"Mukherjee","sequence":"additional","affiliation":[{"name":"Purdue University"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Saurabh","family":"Bagchi","sequence":"additional","affiliation":[{"name":"Purdue University"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Somali","family":"Chaterji","sequence":"additional","affiliation":[{"name":"Purdue University"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yingyu","family":"Liang","sequence":"additional","affiliation":[{"name":"University of Wisconsin-Madison"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yin","family":"Li","sequence":"additional","affiliation":[{"name":"University of Wisconsin-Madison"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","doi-asserted-by":"publisher","DOI":"10.52202\/068431-1723"},{"key":"ref2","volume-title":"Anthropic. The system card: Claude opus 4 & claude sonnet 4","year":"2025"},{"key":"ref3","article-title":"Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond","author":"Bai","year":"2023","journal-title":"arXiv preprint"},{"key":"ref4","article-title":"Conditional computation in neural networks for faster models","author":"Bengio","year":"2015","journal-title":"arXiv preprint"},{"key":"ref5","article-title":"Flextron: Many-in-one flexible large language model","volume-title":"Forty-first International Conference on Machine Learning","author":"Cai","year":"2024"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.01487"},{"key":"ref7","article-title":"Shikra: Unleashing multimodal llm\u2019s referential dialogue magic","author":"Chen","year":"2023","journal-title":"arXiv preprint"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-73004-7_2"},{"key":"ref9","article-title":"Generating long sequences with sparse transformers","author":"Child","year":"2019","journal-title":"arXiv preprint"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.52202\/075280-2142"},{"key":"ref11","first-page":"5547","article-title":"Glam: Efficient scaling of language models with mixture-of-experts","volume-title":"International Conference on Machine Learning","author":"Du","year":"2022"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2017.194"},{"key":"ref13","article-title":"Mme: A comprehensive evaluation benchmark for multimodal large language models","author":"Fu","year":"2023","journal-title":"arXiv preprint"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2017.670"},{"key":"ref15","first-page":"458","article-title":"Speedboost: Anytime prediction with uniform near-optimality","author":"Grubb","year":"2012","journal-title":"Artificial Intelligence and Statistics"},{"key":"ref16","article-title":"Muprop: Unbiased backpropagation for stochastic neural networks","author":"Gu","year":"2015","journal-title":"arXiv preprint"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00380"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.1109\/TPAMI.2021.3117837"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v33i01.33013812"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.52202\/079017-2670"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2019.00686"},{"key":"ref22","article-title":"Categorical reparameterization with gumbel-softmax","volume-title":"International Conference on Learning Representations","author":"Jang","year":"2017"},{"key":"ref23","article-title":"Phi-2: The surprising power of small language models","author":"Javaheripi","year":"2023","journal-title":"Microsoft Research Blog"},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.1109\/TPAMI.2019.2959322"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2014.80"},{"key":"ref26","article-title":"The need for speed: Pruning transformers with one recipe","volume-title":"The Twelfth International Conference on Learning Representations","author":"Khaki","year":"2024"},{"key":"ref27","article-title":"Sparks of large audio models: A survey and outlook","author":"Latif","year":"2023","journal-title":"arXiv preprint"},{"key":"ref28","volume-title":"Introducing idefics: An open reproduction of state-of-the-art visual language model","author":"Lauren\u00e7on","year":"2023"},{"key":"ref29","article-title":"Seed-bench: Benchmarking multimodal LLMs with generative comprehension","author":"Li","year":"2023","journal-title":"arXiv preprint"},{"key":"ref30","article-title":"Llava-onevision: Easy visual task transfer","author":"Li","year":"2024","journal-title":"arXiv preprint"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR46437.2021.00609"},{"key":"ref32","first-page":"19730","article-title":"Blip-2: Bootstrapping language-image pretraining with frozen image encoders and large language models","volume-title":"International conference on machine learning","author":"Li","year":"2023"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.emnlp-main.20"},{"key":"ref34","article-title":"Moe-llava: Mixture of experts for large vision-language models","author":"Lin","year":"2024","journal-title":"arXiv preprint"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.52202\/075280-1516"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.02484"},{"key":"ref37","author":"Liu","year":"2024","journal-title":"Llavanext: Improved reasoning, ocr, and world knowledge"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-72658-3_13"},{"key":"ref39","doi-asserted-by":"publisher","DOI":"10.52202\/068431-0182"},{"key":"ref40","article-title":"The concrete distribution: A continuous relaxation of discrete random variables","volume-title":"International Conference on Learning Representations","author":"Maddison","year":"2017"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52688.2022.01199"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-58571-6_6"},{"key":"ref43","article-title":"OpenAI. GPT-4 technical report","year":"2023","journal-title":"arXiv preprint"},{"key":"ref44","first-page":"24898","article-title":"Iared2: Interpretability-aware redundancy reduction for vision transformers","volume":"34","author":"Pan","year":"2021","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref45","article-title":"Quadratic is not what you need for multimodal large language models","author":"Pham","year":"2024","journal-title":"arXiv preprint"},{"key":"ref46","article-title":"Dynamicvit: Efficient vision transformers with dynamic token sparsification","author":"Rao","year":"2021","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref47","article-title":"Mixture-of-depths: Dynamically allocating compute in transformer-based language models","author":"Raposo","year":"2024","journal-title":"arXiv preprint"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.acl-long.829"},{"key":"ref49","article-title":"Llava-prumerge: Adaptive token reduction for efficient large multimodal models","author":"Shang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref50","article-title":"Llava-mod: Making llava tiny via moe knowledge distillation","author":"Shu","year":"2024","journal-title":"arXiv preprint"},{"key":"ref51","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2019.00851"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1145\/3694715.3695964"},{"issue":"1","key":"ref53","first-page":"1929","article-title":"Dropout: a simple way to prevent neural networks from overfitting","volume":"15","author":"Srivastava","year":"2014","journal-title":"The journal of machine learning research"},{"key":"ref54","article-title":"Qwen2-vl: Enhancing vision-language model\u2019s perception of the world at any resolution","author":"Wang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-01261-8_25"},{"key":"ref56","article-title":"Not all images are worth 16 \u00d7 16 words: Dynamic transformers for efficient image recognition","author":"Wang","year":"2021","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1706.03762"},{"key":"ref58","article-title":"Small language model meets with reinforced vision vocabulary","author":"Wei","year":"2024","journal-title":"arXiv preprint"},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00919"},{"key":"ref60","doi-asserted-by":"publisher","DOI":"10.1145\/3492321.3519577"},{"key":"ref61","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52688.2022.00256"},{"key":"ref62","first-page":"1299","article-title":"The greedy miser: learning under test-time budgets","volume-title":"Proceedings of the 29th International Coference on International Conference on Machine Learning","author":"Xu","year":"2012"},{"key":"ref63","article-title":"Minicpm-v: A GPT4 V level MLLM on your phone","author":"Yao","year":"2024","journal-title":"arXiv preprint"},{"key":"ref64","article-title":"Mm-vet: Evaluating large multimodal models for integrated capabilities","author":"Yu","year":"2023","journal-title":"arXiv preprint"},{"key":"ref65","article-title":"Tinygpt-v: Efficient multimodal large language model via small backbones","author":"Yuan","year":"2023","journal-title":"arXiv preprint"},{"key":"ref66","article-title":"Llm inference unveiled: Survey and roofline model insights","author":"Yuan","year":"2024","journal-title":"arXiv preprint"},{"key":"ref67","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV51701.2025.01877"},{"key":"ref68","article-title":"Tinyllava: A framework of small-scale large multimodal models","author":"Zhou","year":"2024","journal-title":"arXiv preprint"},{"key":"ref69","article-title":"MiniGPT-4: Enhancing visionlanguage understanding with advanced large language models","volume-title":"The Twelfth International Conference on Learning Representations","author":"Zhu","year":"2024"},{"key":"ref70","first-page":"arXiv","article-title":"A comprehensive overhaul of multimodal assistant with small language models","author":"Zhu","year":"2024","journal-title":"arXiv e-prints"},{"key":"ref71","doi-asserted-by":"publisher","DOI":"10.1145\/3688863.3689575"}],"event":{"name":"2025 IEEE\/CVF International Conference on Computer Vision (ICCV)","location":"Honolulu, HI, USA","start":{"date-parts":[[2025,10,19]]},"end":{"date-parts":[[2025,10,25]]}},"container-title":["2025 IEEE\/CVF International Conference on Computer Vision (ICCV)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/11443115\/11443287\/11444101.pdf?arnumber=11444101","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,4,30]],"date-time":"2026-04-30T06:11:08Z","timestamp":1777529468000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11444101\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,10,19]]},"references-count":71,"URL":"https:\/\/doi.org\/10.1109\/iccv51701.2025.00339","relation":{},"subject":[],"published":{"date-parts":[[2025,10,19]]}}}