中央研究院 資訊科學研究所

Gaussian process regression (GPR) and classification (GPC) are non-parametric Bayesian approaches to learning a regression or classification function.  GPR and GPC have several properties that are desirable, such as robust learning on small training sets, probabilistic predictions, effective non-linear representations using kernel functions, and principled estimation of these kernel hyperparameters through maximum marginal likelihood.  In this talk, we propose a generalized Gaussian process model (GGPM), which is a unifying framework that encompasses many existing Gaussian process models, such as GP regression, classification, and counting. In the GGPM framework, the observation likelihood of the GP model is itself parameterized using the exponential family distribution. By deriving approximate inference algorithms for the generalized GP model, we are able to easily apply the same algorithm to all other GP models. Novel GP models are created by changing the parameterization of the likelihood function, which greatly simplifies their creation for task-specific output domains. We also derive a closed-form efficient Taylor approximation for inference on the model, and draw interesting connections with other model-specific closed-form approximations. Finally, using the GGPM, we create several new GP models and show their efficacy in building task-specific GP models for computer vision.

Antoni B. Chan received the BS and MEng degrees in electrical engineering from Cornell University in 2000 and 2001, respectively, and the PhD degree in electrical and computer engineering from University of California, San Diego (UCSD), in 2008. From 2001 to 2003, he was a visiting scientist in the Vision and Image Analysis Lab at Cornell University, and in 2009, he was a postdoctoral researcher in the Statistical Visual Computing Lab at UCSD. In 2009, he joined the Department of Computer Science at the City University of Hong Kong, as an assistant professor. From 2006 to 2008, he was the recipient of a US National Science Foundation (NSF) IGERT Fellowship. His research interests are in computer vision, machine learning, pattern recognition, and music analysis.