 |
Center for Nonlinear Studies |
Nowadays, many problems of artificial intelligence are formulated as probabilistic inference. Even though, standard inference algorithms are available, a lot of hand-crafting is required to turn a probabilistic model into code. Probabilistic programming is a powerful tool to specify probabilistic models directly in terms of a computer program. This can either be achieved by designing a specialized programming language for expressing probabilistic models or extending the semantics of an existing language.
In this talk, I will shortly explain the semantics underlying a probabilistic computation and discuss the implementation of different inference algorithms. Finally, I will present my Clojure library for probabilistic programming and show some examples of Bayes nets and Gaussian mixture models expressed as probabilistic programs.
Host: David Wolpert