Guest Speakers

摘要

Most of the machine learning methods are based on optimization models. Constructing the machine learning model by minimizing the loss function is a basic example to express the relation between machine learning and optimization. With the growing number of data and increase in dimension require sparse models in order to reduce the computation cost. While sparse models are computationally efective in practice, it can be obtained through zero norm or one norm approximations by non-convex optimization models. The challenge appears here at the non-convexity structure of the problem. Recently, learning algorithms based on such sparse models with nonconvexity approximations have gained an interest in literature. In this talk, I will give recent approximations and sparse learning models exist in literature specifically in feature selection, ensemble learning and deep learning. Since ensemble learning aims to aggregate the most accurate and diverse learners to produce the best final decision, selecting the best learning models out of the ensemble becomes a sparse learning similar to feature selection algorithms. The connections between deep learning and sparse learning in a same fashion will be explained throughout this talk. As an example to such methodologies, some ongoing biomedical projects at Bahcesehir University Computer Vision Laboratory on machine learning will be presented in the end of this talk.