Prof. Shinto Eguchi, PhD
Director, Division of Statistical Learning and Inference. The Institute of Statistical Mathematics (ISM),Tokyo, Japan.
Keynote Speech Title: Two Paradigms in Statistical Prediction
Abstract: Statistical prediction is one of central issues for biosciences. We overview the present direction including support vector machine, adaptive boosting, deep learning and forth. Two paradigms are focused to discuss a future direction: One is a convex loss approach, which guarantees mathematically strong and sound properties reduced from the convexity. Such convex loss functions are shown to satisfy Bayes risk consistency. Typically there exists uniquely the optimal linear predictor under the assumption of linear model. The other is a non-convex loss approach, which offers flexible performance beyond that by linear predictors. In particular we discuss a quasi linear predictor to give the flexible performance in unsupervised and supervised manners with numerical demonstration.