Probabilistic Programming: An Emerging Paradigm in Machine Learning

Probabilistic Programming is the next emerging technology in machine learning, after Deep Learning and Big Data Analytics. By combining general purpose programming with probabilistic modelling, probabilistic programming makes it possible to perform complex statistical reasoning with minimum programming efforts and high computational efficiency.

The main idea is:

1. Use a computer program augmented with statistical operators,
2. formulate a suitable probabilistic model for a given data set and
3. perform automated statistical inference by executing the program.

This has been made possible by recent breakthroughs in automated inference (including advanced sampling methods and variational inference) and numerical computing (including libraries like PyTorch and TensorFlow). Probabilistic programming will make it possible to infer Deep Learning models that deliver reliable estimates of uncertainty from relatively small data sets. By decoupling statistical modelling (i.e., formulating the questions) from statistical inference (i.e., answering the questions), probabilistic programming is expected to dramatically extend the scope of probabilistic reasoning in technology, science, finance, society and medicine.

Speaker:
Thomas Hamelryck is associate professor at the University of Copenhagen. He has a shared position at the Department of Computer Science (DIKU) and the Department of Biology. His main interests are bioinformatics (in particular, the protein folding problem) and Bayesian machine learning, including probabilistic programming.