A new class of estimators for the Sliced Wasserstein (SW) distance has been developed, leveraging cumulative distribution functions (CDFs) instead of quantile functions. This approach allows for massive dataset parallelism and avoids the need for sorting projected samples, making it more computationally efficient. The method is particularly well-suited for scenarios involving mixtures of Gaussians and is compatible with federated learning, as CDFs can be computed and aggregated locally without sharing raw data. AI
IMPACT This research could improve the efficiency of distance metric calculations in machine learning, particularly for distributed and federated learning scenarios.
RANK_REASON The cluster contains an arXiv preprint detailing a new research methodology in statistics and machine learning.
- arXiv
- Christophe Vauthier
- Cumulative Distribution Functions
- federated learning
- Gaussian mixture model
- Quantile Functions
- Sliced-Wasserstein distance
- Wasserstein metric
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
