Learning General Gaussian Mixtures With Efficient Score Matching

Watch a 35-minute research lecture from the Joint IFML/MPG Symposium at the Simons Institute where Vasilis Kontonis from UT Austin presents novel approaches to learning Gaussian mixtures through score matching techniques. Explore how diffusion models can be leveraged to transform distribution learning into a supervised learning problem, with a focus on approximating score functions of Gaussian mixtures using piecewise-polynomial functions. Discover groundbreaking theoretical guarantees for unsupervised learning tasks achieved through diffusion models, offering an alternative to traditional method of moments approaches.