Research map / Vision-language and multimodal / Cross-modal learning
OSKAR: Omnimodal Self-supervised Knowledge Abstraction and Representation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes OSKAR, a multimodal foundation model pretrained by predicting bootstrapped latent features of masked tokens across modalities, without negatives or hand-crafted augmentations.
Problem
Building multimodal foundation models that learn without generative pixel reconstruction, negative pairs, memory banks, or hand-crafted augmentations.
Approach
Bootstrapped latent feature prediction across modalities: given masked tokens from multiple modalities, predict a subset of missing tokens per modality, avoiding memorization of unnecessary details such as pixels.
Closest papers by meaning
- Unity by Diversity: Improved Representation Learning for Multimodal VAEs (NeurIPS 2024 · similarity 0.88)
- OmniSegmentor: A Flexible Multi-Modal Learning Framework for Semantic Segmentation (NeurIPS 2025 · similarity 0.86)
- OmniBind: Large-scale Omni Multimodal Representation via Binding Spaces (ICLR 2025 · similarity 0.86)
- Matryoshka Multimodal Models (ICLR 2025 · similarity 0.86)
- Omni-Interactive Universal Embedder (NeurIPS 2026 · similarity 0.86)
- Emu: Generative Pretraining in Multimodality (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.