{"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/alternating-gradient-descent-and-mixture-of","title":"Alternating Gradient Descent and Mixture-of-Experts for Integrated Multimodal Perception","arxiv_id":"2305.06324","date":"2023-05-10","proceeding":"NeurIPS 2023 11","authors":["Hassan Akbari","Dan Kondratyuk","Yin Cui","Rachel Hornung","Huisheng Wang","Hartwig Adam"],"abstract":"We present Integrated Multimodal Perception (IMP), a simple and scalable multimodal multi-task training and modeling approach. IMP integrates multimodal inputs including image, video, text, and audio into a single Transformer encoder with minimal modality-specific components. IMP makes use of a novel design that combines Alternating Gradient Descent (AGD) and Mixture-of-Experts (MoE) for efficient model and task scaling. We conduct extensive empirical studies and reveal the following key insights: 1) Performing gradient descent updates by alternating on diverse modalities, loss functions, and tasks, with varying input resolutions, efficiently improves the model. 2) Sparsification with MoE on a single modality-agnostic encoder substantially improves the performance, outperforming dense models that use modality-specific encoders or additional fusion layers and greatly mitigates the conflicts between modalities. IMP achieves competitive performance on a wide range of downstream tasks including video classification, image classification, image-text, and video-text retrieval. Most notably, we train a sparse IMP-MoE-L variant focusing on video tasks that achieves new state-of-the-art in zero-shot video classification: 77.0% on Kinetics-400, 76.8% on Kinetics-600, and 68.3% on Kinetics-700, improving the previous state-of-the-art by +5%, +6.7%, and +5.8%, respectively, while using only 15% of their total training computational cost.","url_abs":"https://arxiv.org/abs/2305.06324v2","url_pdf":"https://arxiv.org/pdf/2305.06324v2.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":[],"tasks":[{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"mixture-of-experts","task_name":"Mixture-of-Experts"},{"task_slug":"text-retrieval","task_name":"Text Retrieval"},{"task_slug":"video-classification","task_name":"Video Classification"},{"task_slug":"video-text-retrieval","task_name":"Video-Text Retrieval"},{"task_slug":"zero-shot-action-recognition","task_name":"Zero-Shot Action Recognition"},{"task_slug":"zero-shot-environment-sound-classification","task_name":"Zero-Shot Environment Sound Classification"},{"task_slug":"zero-shot-learning","task_name":"Zero-Shot Learning"},{"task_slug":"zero-shot-transfer-image-classification","task_name":"Zero-Shot Transfer Image Classification"},{"task_slug":"image-classification","task_name":"image-classification"}],"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"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/zero-shot-action-recognition-on-hmdb51","task":"Zero-Shot Action Recognition","dataset":"HMDB51","model":"IMP-MoE-L","rank_in_archive_order":5,"of":29,"metrics":{"Top-1 Accuracy":"59.1"},"uses_additional_data":true},{"leaderboard":"/sota/zero-shot-action-recognition-on-kinetics","task":"Zero-Shot Action Recognition","dataset":"Kinetics","model":"IMP-MoE-L","rank_in_archive_order":2,"of":20,"metrics":{"Top-1 Accuracy":"76.8"},"uses_additional_data":true},{"leaderboard":"/sota/zero-shot-action-recognition-on-ucf101","task":"Zero-Shot Action Recognition","dataset":"UCF101","model":"IMP-MoE-L","rank_in_archive_order":2,"of":35,"metrics":{"Top-1 Accuracy":"91.5"},"uses_additional_data":true},{"leaderboard":"/sota/zero-shot-transfer-image-classification-on-1","task":"Zero-Shot Transfer Image Classification","dataset":"ImageNet","model":"IMP-MoE-L","rank_in_archive_order":8,"of":23,"metrics":{"Accuracy (Private)":"83.9"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=2305.06324","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}