Scalable methods for high-dimensional inverse problems using neural networks and implicit method of moments

Description:  I will talk about two recent projects of mine in that concern high-dimensional inverse problems. In the first project, we use machine learning tools for discovering laws and equations that govern high-dimensional datasets. Specifically, I will present two methods we developed, one to discover partial differential equations on data, and the other to recover latent stochastic differential equations. On the second part of the talk, I will mention an application of the method of moments to learn parameters of a Gaussian mixture model. Specifically, we propose numerical methods for implicit computations with the higher-order moments of the data. Our approach reduces the computational and storage costs, and opens the door to the competitiveness of the method of moments as compared to expectation maximization methods. Date:  2022-12-02 Start Time:   14:30 Speaker:  João M. Pereira (IMPA, Rio de Janeiro, Brazil) Institution:  IMPA Place:  Sala 5.5, DMUC Research Groups: -Numerical Analysis and Optimization See more:   <Main>