Explore latent stochastic differential equations for irregularly-sampled time series in this comprehensive seminar on theoretical machine learning. Delve into topics such as ordinary differential equations, latent variable models, ODE latent-variable models, and stochastic transition dynamics. Learn about the Poisson Process Likelihoods, limitations of Latent ODEs, and continuous-time backpropagation. Discover the intricacies of running SDEs backwards, time and memory costs, and variational inference. Gain insights from speaker David Duvenaud of the University of Toronto as he presents advanced concepts in machine learning and data analysis for handling irregularly-timed datasets.
Latent Stochastic Differential Equations for Irregularly-Sampled Time Series - David Duvenaud
Institute for Advanced Study via YouTube
Overview
Syllabus
Intro
Summary . We generalized the adjoint sensitivity method to
Motivation: Irregularly-timed datasets
Ordinary Differential Equations
Latent variable models
ODE latent-variable model
Physionet: Predictive accuracy
Poisson Process Likelihoods
Limitations of Latent ODES
Stochastic transition dynamics
How to fit ODE params?
Continuous-time Backpropagation
Need to store noise
Brownian Tree Code
What is running an SDE backwards?
Time and memory cost
Variational inference
Taught by
Institute for Advanced Study
