The Benefits of Balance: From Information Projections to Variance Reduction
This provides a theoretical insight for improving data balancing in contrastive multimodal learning and self-supervised clustering, though it appears incremental as it builds on existing methods like CLIP and DINO.
The paper tackles the problem of data balancing across modalities and sources in foundation models, showing that it reduces variance, with a non-asymptotic statistical bound quantifying this effect.
Data balancing across multiple modalities and sources appears in various forms in foundation models in machine learning and AI, e.g. in CLIP and DINO. We show that data balancing across modalities and sources actually offers an unsuspected benefit: variance reduction. We present a non-asymptotic statistical bound that quantifies this variance reduction effect and relates it to the eigenvalue decay of Markov operators. Furthermore, we describe how various forms of data balancing in contrastive multimodal learning and self-supervised clustering can be better understood, and even improved upon, owing to our variance reduction viewpoint.