English

i-Code: An Integrative and Composable Multimodal Learning Framework

Machine Learning 2022-05-06 v2 Artificial Intelligence Computation and Language Computer Vision and Pattern Recognition Audio and Speech Processing

Abstract

Human intelligence is multimodal; we integrate visual, linguistic, and acoustic signals to maintain a holistic worldview. Most current pretraining methods, however, are limited to one or two modalities. We present i-Code, a self-supervised pretraining framework where users may flexibly combine the modalities of vision, speech, and language into unified and general-purpose vector representations. In this framework, data from each modality are first given to pretrained single-modality encoders. The encoder outputs are then integrated with a multimodal fusion network, which uses novel attention mechanisms and other architectural innovations to effectively combine information from the different modalities. The entire system is pretrained end-to-end with new objectives including masked modality unit modeling and cross-modality contrastive learning. Unlike previous research using only video for pretraining, the i-Code framework can dynamically process single, dual, and triple-modality data during training and inference, flexibly projecting different combinations of modalities into a single representation space. Experimental results demonstrate how i-Code can outperform state-of-the-art techniques on five video understanding tasks and the GLUE NLP benchmark, improving by as much as 11% and demonstrating the power of integrative multimodal pretraining.

Keywords

Cite

@article{arxiv.2205.01818,
  title  = {i-Code: An Integrative and Composable Multimodal Learning Framework},
  author = {Ziyi Yang and Yuwei Fang and Chenguang Zhu and Reid Pryzant and Dongdong Chen and Yu Shi and Yichong Xu and Yao Qian and Mei Gao and Yi-Ling Chen and Liyang Lu and Yujia Xie and Robert Gmyr and Noel Codella and Naoyuki Kanda and Bin Xiao and Lu Yuan and Takuya Yoshioka and Michael Zeng and Xuedong Huang},
  journal= {arXiv preprint arXiv:2205.01818},
  year   = {2022}
}
R2 v1 2026-06-24T11:06:32.668Z