Reproduce “Uncertainty Voting Ensemble for Imbalanced Deep Regression” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuchang Jiang, Vivien Sainte Fare Garnot, Konrad Schindler, Jan Dirk Wegner
UVOTE, a deep learning ensemble method, addresses imbalanced regression by dynamically aggregating predictions from specialized expert models based on their predicted aleatoric uncertainty. The system demonstrates state-of-the-art performance, particularly in data-scarce regions, while providing well-calibrated uncertainty estimates across diverse datasets.
Key insights
- Aleatoric uncertainty estimation provides a principled and computationally efficient mechanism for dynamically aggregating expert predictions in an ensemble, directly integrating confidence into the aggregation process.
- Training diverse expert models with varying degrees of inverse-frequency weighting (controlled by a rebalancing parameter p_m) effectively encourages specialization across different data density regions.
- A dynamic learning strategy, which initially prioritizes robust latent representation learning and gradually phases in imbalance-focused experts, is crucial for balancing overall performance with targeted improvements in sparse regions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.