Event
Vahid Partovi Nia, Huawei Technologies and Ecole Polytechnique Montreal
Linear Unsupervised and Active Learning
This talk is composed of two parts, linear unsupervised learning, and linear active learning. Part 1: Unsupervised machine learning, or clustering, divides a heterogeneous data into homogenous subsets. Here we develop a clustering algorithm for linear regressions, with direct application in clustering shapes. Looking at physical shapes as a closed surface, and employing this algorithm allows us to treat clustering shapes through mathematical functions. This new view extends the Bayesian information criterion for clustering purpose. Part 2: Active learning is concerned about requesting specific data points to increase prediction power, combining machine learning with design of experiments. I develop linear active learning, and will discuss the challenges of applying the method in practice on empirical modelling of optical fibre amplifiers.