Erik B. Sudderth, Statistical Computation & Perception
I am an Associate Professor of Computer Science at the University of California, Irvine. I previously spent seven great years on the faculty at Brown University, where I remain an Adjunct Associate Professor of Computer Science. My Learning, Inference, & Vision Group develops statistical methods for scalable machine learning, with applications in artificial intelligence, vision, and the natural and social sciences. Particular areas of expertise include:
- Machine Learning
- graphical models, Bayesian nonparametrics, approximate inference
- Computer Vision
- object recognition & scene understanding, segmentation, motion & tracking
- Signal Processing
- nonlinear dynamical systems, image & video analysis, multiscale models
See my CVPR tutorial for an overview of Bayesian nonparametrics in computer vision. For a tutorial introduction to probabilistic modeling and approximate inference, see the background chapter of my doctoral thesis, advised by Professors Alan Willsky and William Freeman at MIT EECS. My postdoctoral research at Berkeley EECS was advised by Professors Michael Jordan and Stuart Russell.
For more information: bio · curriculum vitæ · research projects & code · publications & lectures
- At AISTATS 2018, our framework for prediction-constrained training of probabilistic models leads to improved semi-supervised learning of topic models, with applications to the analysis of documents and electronic health records. This work received the SoCal NLP Symposium best paper award.
- Our work on latent support surfaces for semantic 3D scene understanding, which uses contextual cues to detect objects in RGB-D images, appears at CVPR 2018.
- Work with BrainGate on multiscale semi-Markov dynamics for improved brain-computer interfaces appeared at NIPS 2017. A supplemental video demonstrates accurate, interactive control of a computer cursor by a clinical trial participant with tetraplegia.
- I gave a talk at the 2017 SoCal Machine Learning Symposium about our diverse particle max-product algorithm, which gives state-of-the-art predictions of continuous protein side-chain conformations. Code available.
- Our open source toolbox BNPy: Bayesian Nonparametric clustering for Python implements scalable, stochastic and memoized variational inference algorithms. Read about applications to HDP topic models at AI & Statistics 2015, to HDP hidden Markov models at NIPS 2015, and to patch-based models for natural image processing at ICML 2017.
- The 2014 ISBA Mitchell Prize for Bayesian analysis of an important applied problem goes to our NET-VISA system for global seismic monitoring, learned from data provided by the comprehensive nuclear-test-ban treaty organization (CTBTO). For details see the Brown University news article.
- An NSF CAREER Award supports my work on large-scale Bayesian nonparametric learning.
- Weiss & Pearl introduce our review article on Nonparametric Belief Propagation for the CACM.
- Associate editor for IEEE Transactions on Pattern Analysis and Machine Intelligence.
- Member of advisory committee for the NIPS Workshops on Practical Bayesian Nonparametrics.
- Area chair for ICML 2015, CVPR 2015, ICCV 2015, NIPS 2016, & ICML 2017.
- Editor, IEEE PAMI Special Issue on Bayesian Nonparametrics, Feb. 2015. (editorial)
- Organizer, ICERM Workshop & Tutorials on Bayesian Nonparametrics, Sept. 2012. (group photo)
- Editor, IEEE Signal Processing Magazine special issue on Recent Advances & Emerging Developments of Graphical Models, Nov. 2010. (editorial)
- Editor, IEEE PAMI Special Issue on Probabilistic Graphical Models in Computer Vision, Oct. 2009. (editorial)