ICML 2022 accepted papers on Vision-language and multimodal
17 of the 1,233 papers accepted at ICML 2022 (1.4%) are on Vision-language and multimodal, compared with 0.6% at ICML 2021. They fall into 3 topics.
See ICML 2022 in the interactive map
Submitting to ICML? Checklist →
Working on something in this area? Describe your idea in scime atlas to see the closest papers by meaning and how crowded the spot has become.
Topics and papers
Human and embodied modeling · 7 papers
- data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language (234 citations)
- Multi-Grained Vision Language Pre-Training: Aligning Texts with Visual Concepts (95 citations)
- The Unsurprising Effectiveness of Pre-Trained Vision Models for Control (21 citations)
- CtrlFormer: Learning Transferable State Representation for Visual Control via Transformer (2 citations)
- Generative Modeling for Multi-task Visual Learning (1 citation)
- Matching Structure for Dual Learning
- Parametric Visual Program Induction with Function Modularization
Cross-modal learning · 6 papers
- OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework (251 citations)
- Modality Competition: What Makes Joint Training of Multi-modal Network Fail in Deep Learning? (Provably) (22 citations)
- Characterizing and Overcoming the Greedy Nature of Learning in Multi-modal Deep Neural Networks (20 citations)
- Geometric Multimodal Contrastive Representation Learning (14 citations)
- VLMixer: Unpaired Vision-Language Pre-training via Cross-Modal CutMix (11 citations)
- Mitigating Modality Collapse in Multimodal VAEs via Impartial Optimization (6 citations)
CLIP and zero-shot · 4 papers
- BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (863 citations)
- Data Determines Distributional Robustness in Contrastive Language Image Pre-training (CLIP) (26 citations)
- IGLUE: A Benchmark for Transfer Learning across Modalities, Tasks, and Languages (24 citations)
- VLUE: A Multi-Task Multi-Dimension Benchmark for Evaluating Vision-Language Pre-training
