Advances in Representation Learning and Multimodal Fusion

The field of representation learning is moving towards more robust and unified frameworks, with a focus on improving discriminative and generative capabilities. Recent developments have introduced new contrastive learning methods, such as those utilizing mutual information and adaptive view generation, which have shown promising results in various downstream tasks. Additionally, there is a growing interest in multimodal fusion, with approaches aimed at balancing modality contributions and capturing synergistic information between modalities. These advancements have the potential to enhance model robustness and performance in real-world scenarios. Noteworthy papers include: Contrastive Mutual Information Learning, which proposes a probabilistic framework for representation learning, and GenView++, which introduces a unified framework for adaptive view generation and quality-driven supervision. InfMasking is also notable for its contrastive synergistic information extraction method, and MIDAS for its misalignment-based data augmentation strategy. Furthermore, FairContrast has made significant contributions to enhancing fairness in representation learning on tabular data.

Sources

Contrastive Mutual Information Learning: Toward Robust Representations without Positive-Pair Augmentations

SSVIF: Self-Supervised Segmentation-Oriented Visible and Infrared Image Fusion

GenView++: Unifying Adaptive View Generation and Quality-Driven Supervision for Contrastive Representation Learning

Robust Multimodal Semantic Segmentation with Balanced Modality Contributions

CLASP: Adaptive Spectral Clustering for Unsupervised Per-Image Segmentation

InfMasking: Unleashing Synergistic Information by Contrastive Multimodal Interactions

MIDAS: Misalignment-based Data Augmentation Strategy for Imbalanced Multimodal Learning

Self-Supervised Representation Learning as Mutual Information Maximization

FairContrast: Enhancing Fairness through Contrastive learning and Customized Augmenting Methods on Tabular Data

Built with on top of