{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/scaling-vision-with-sparse-mixture-of-experts","title":"Scaling Vision with Sparse Mixture of Experts","arxiv_id":"2106.05974","date":"2021-06-10","proceeding":"NeurIPS 2021 12","authors":["Carlos Riquelme","Joan Puigcerver","Basil Mustafa","Maxim Neumann","Rodolphe Jenatton","André Susano Pinto","Daniel Keysers","Neil Houlsby"],"abstract":"Sparsely-gated Mixture of Experts networks (MoEs) have demonstrated excellent scalability in Natural Language Processing. In Computer Vision, however, almost all performant networks are \"dense\", that is, every input is processed by every parameter. We present a Vision MoE (V-MoE), a sparse version of the Vision Transformer, that is scalable and competitive with the largest dense networks. When applied to image recognition, V-MoE matches the performance of state-of-the-art networks, while requiring as little as half of the compute at inference time. Further, we propose an extension to the routing algorithm that can prioritize subsets of each input across the entire batch, leading to adaptive per-image compute. This allows V-MoE to trade-off performance and compute smoothly at test-time. Finally, we demonstrate the potential of V-MoE to scale vision models, and train a 15B parameter model that attains 90.35% on ImageNet.","url_abs":"https://arxiv.org/abs/2106.05974v1","url_pdf":"https://arxiv.org/pdf/2106.05974v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"scaling-vision-with-sparse-mixture-of-experts","repo_url":"https://github.com/google-research/vmoe","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"jax","reach":null}],"tasks":[{"task_slug":"few-shot-image-classification","task_name":"Few-Shot Image Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"mixture-of-experts","task_name":"Mixture-of-Experts"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"},{"method_slug":"vision-transformer","method_name":"Vision Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-1-1","task":"Few-Shot Image Classification","dataset":"ImageNet - 1-shot","model":"ViT-MoE-15B (Every-2)","rank_in_archive_order":1,"of":8,"metrics":{"Top 1 Accuracy":"68.66"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-1-1","task":"Few-Shot Image Classification","dataset":"ImageNet - 1-shot","model":"V-MoE-H/14 (Every-2)","rank_in_archive_order":3,"of":8,"metrics":{"Top 1 Accuracy":"63.38"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-1-1","task":"Few-Shot Image Classification","dataset":"ImageNet - 1-shot","model":"V-MoE-H/14 (Last-5)","rank_in_archive_order":4,"of":8,"metrics":{"Top 1 Accuracy":"62.95"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-1-1","task":"Few-Shot Image Classification","dataset":"ImageNet - 1-shot","model":"V-MoE-L/16 (Every-2)","rank_in_archive_order":5,"of":8,"metrics":{"Top 1 Accuracy":"62.41"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-1-1","task":"Few-Shot Image Classification","dataset":"ImageNet - 1-shot","model":"VIT-H/14","rank_in_archive_order":6,"of":8,"metrics":{"Top 1 Accuracy":"62.34"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-10","task":"Few-Shot Image Classification","dataset":"ImageNet - 10-shot","model":"ViT-MoE-15B (Every-2)","rank_in_archive_order":2,"of":7,"metrics":{"Top 1 Accuracy":"84.29"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-10","task":"Few-Shot Image Classification","dataset":"ImageNet - 10-shot","model":"V-MoE-H/14 (Every-2)","rank_in_archive_order":5,"of":7,"metrics":{"Top 1 Accuracy":"80.33"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-10","task":"Few-Shot Image Classification","dataset":"ImageNet - 10-shot","model":"V-MoE-H/14 (Last-5)","rank_in_archive_order":6,"of":7,"metrics":{"Top 1 Accuracy":"80.1"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-10","task":"Few-Shot Image Classification","dataset":"ImageNet - 10-shot","model":"VIT-H/14","rank_in_archive_order":7,"of":7,"metrics":{"Top 1 Accuracy":"79.01"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-5","task":"Few-Shot Image Classification","dataset":"ImageNet - 5-shot","model":"ViT-MoE-15B (Every-2)","rank_in_archive_order":1,"of":8,"metrics":{"Top 1 Accuracy":"82.78"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-5","task":"Few-Shot Image Classification","dataset":"ImageNet - 5-shot","model":"V-MoE-H/14 (Every-2)","rank_in_archive_order":5,"of":8,"metrics":{"Top 1 Accuracy":"78.21"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-5","task":"Few-Shot Image Classification","dataset":"ImageNet - 5-shot","model":"V-MoE-H/14 (Last-5)","rank_in_archive_order":6,"of":8,"metrics":{"Top 1 Accuracy":"78.08"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-5","task":"Few-Shot Image Classification","dataset":"ImageNet - 5-shot","model":"V-MoE-L/16 (Every-2)","rank_in_archive_order":7,"of":8,"metrics":{"Top 1 Accuracy":"77.1"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-image-classification-on-imagenet-5","task":"Few-Shot Image Classification","dataset":"ImageNet - 5-shot","model":"VIT-H/14","rank_in_archive_order":8,"of":8,"metrics":{"Top 1 Accuracy":"76.95"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"V-MoE-H/14 (Every-2)","rank_in_archive_order":52,"of":1060,"metrics":{"Number of params":"7200M","Top 1 Accuracy":"88.36%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"VIT-H/14","rank_in_archive_order":61,"of":1060,"metrics":{"Number of params":"656M","Top 1 Accuracy":"88.08%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"V-MoE-L/16 (Every-2)","rank_in_archive_order":88,"of":1060,"metrics":{"Number of params":"3400M","Top 1 Accuracy":"87.41%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-jft-300m","task":"Image Classification","dataset":"JFT-300M","model":"V-MoE-H/14 (Every-2)","rank_in_archive_order":1,"of":4,"metrics":{"prec@1":"60.62"},"uses_additional_data":true},{"leaderboard":"/sota/image-classification-on-jft-300m","task":"Image Classification","dataset":"JFT-300M","model":"V-MoE-H/14 (Last-5)","rank_in_archive_order":2,"of":4,"metrics":{"prec@1":"60.12"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-jft-300m","task":"Image Classification","dataset":"JFT-300M","model":"V-MoE-L/16 (Every-2)","rank_in_archive_order":3,"of":4,"metrics":{"prec@1":"57.65"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-jft-300m","task":"Image Classification","dataset":"JFT-300M","model":"VIT-H/14","rank_in_archive_order":4,"of":4,"metrics":{"prec@1":"56.68"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2106.05974","atlas_url":"https://app.syntology.ai/?focus=2106.05974","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2106.05974"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/google-research/vmoe","reach":null}],"summary":{"ran_honours":1},"by_repo_kind":{"official":{"samples":1,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"b7c6719771038ec2","entry":"get_mixup_concentration","repo":"google-research/vmoe","repo_kind":"official","path":"vmoe/configs/vmoe_paper/common.py","file_url":"https://github.com/google-research/vmoe/blob/HEAD/vmoe/configs/vmoe_paper/common.py","link_basis":"first_harvest_node","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":"well_formed","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"b7c6719771038ec2"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}