David Blei

Within machine learning, David Blei has made important and numerous contributions to probabilistic graphical models, approximate posterior inference, causal inference, information retrieval, text processing, and Bayesian nonparametric statistics. He is best known for his seminal work on topic models (also known as latent Dirichlet allocation models). He has brought these techniques to the attention of a wide audience in computer science, in areas such as information retrieval, Web search, recommender systems, and computer vision. This work has also gained (and has retained) widespread attention in research areas outside computer science, particularly in the digital humanities and in computational social science. Topic models are, for example, routinely used to classify newspaper articles, as well as to perform sentiment analysis, and categorize emails.

David Blei has also had a major impact on the field of statistics. His research has brought approximate inference ideas from machine learning to the forefront, allowing statisticians to think about inference in a much broader context than in the past. In particular, he has promoted the use of optimization-based variational inference methods in addition to the usual methods such as Monte Carlo sampling. This has led to great gains in inference efficiency that has enabled innovative and widespread applications in research and industry in areas ranging from astrophysics, such as modelling the orbits of exoplanets, to financial forecasting.