Learning Techniques for Big Data Redundant features Group

Learning Techniques for Big Data • Redundant features – Group lasso – Feature selection Learning Techniques for Big Data 1

Online Learning for Group Lasso • Scenario: huge data with group features appear sequentially How to learn the decision function adaptively? Learning Techniques for Big Data 2

Solutions and Properties • Objective function • Three main steps • Close solution • Algorithm • Theoretical guarantee Learning Techniques for Big Data 3

Learning Techniques for Big Data • Redundant features – Group lasso – Feature selection • Insufficient labeled data – Multi-task learning – Unsupervised learning Learning Techniques for Big Data 4

Online Learning for Multi-task Feature Selection • Problems and Motivation – Learning multiple related tasks simultaneously to improve performance – Existing redundant or irrelevant features – Data occur sequentially • Challenges – How to adaptively update the models while selecting the important features? l a. MTFS – Any theoretical guarantee? Learning Techniques for Big Data 5

Solution and Properties • Objective function • Three main steps • Close-formed solution • Algorithm • Theoretical guarantee Learning Techniques for Big Data 6

Learning Techniques for Big Data • Redundant features – Group lasso – Feature selection • Insufficient labeled data – Multi-task learning – Unsupervised learning • Complicated decision function – Multiple kernel learning: level method speedup, generalization Learning Techniques for Big Data 7

Sparse Generalized Multiple Kernel Learning Labeled data: Horse/Donkey • Data characteristics – Multi-source – Heterogeneous • Learning Techniques for Big Data 8

Learning Techniques for Big Data • Redundant features – Group lasso – Feature selection • Insufficient labeled data – Multi-task learning – Unsupervised learning • Complicated decision function – Multiple kernel learning: level method speedup, generalization • Volume data – Online learning • Related work – ICML’ 10, CIKM’ 10 -11, IJCNN’ 10, IEEETNN’ 11, ACMTKDD’ 13 Learning Techniques for Big Data 9