CS-E4830 - Kernel Methods in Machine Learning D, Lecture, 2.3.2022-2.6.2022

Topic outline

• Select topic General

  General
  Not available unless: Your User account contains (use: aalto.fi) is aalto.fi

  
  

  Course logistics

  • Lectures: Wed, 12:15pm (see also section "Zoom")
    
  • Exercise sessions: Please come to TU1 Saab Auditorium. Won't be recorded.
    
  • Discussions take place in Zulip: https://kmml22.zulip.aalto.fi/join/sxhncjfkzttuozcrzrn6g5bm/
  • Assignments: To be submitted in MyCourses (Pen & Paper) and online in Jupyter Hub (Programming).
  • Exam: 02.06.2022, Check Sisu for location.

  
  

  Schedule

  
  

  Date (dd.mm.yyyy)	Activity	Location
  02.03.2022 (12:15)	Lecture 1	Zoom
  09.03.2022 (12:15)	Lecture 2	Zoom
  10.03.2022 (16:15)	Python Tutorial	TU1 Saab Auditorium
  11.03.2022	Assignment 1 release
  16.03.2022 (12:15)	Lecture 3
  17.03.2022 (16:15)	Assignment 1 exercise session	TU1 Saab Auditorium
  21.03.2022 (16:00)	Assignment 1 due
  23.03.2022 (12:15)	Lecture 4	Zoom
  25.03.2022	Assignment 2 release
  30.03.2022 (12:15)	Lecture 5	Zoom
  31.03.2022 (16:15)	Assignment 2 exercise session	TU1 Saab Auditorium
  06.04.2022 (12:15)	Lecture 6	Zoom
  06.04.2022 (16:00)	Assignment 2 due
  20.04.2022 (12:15)	Lecture 7	Zoom
  22.04.2022	Assignment 3 release
  27.04.2022 (12:15)	Lecture 8	Zoom
  28.04.2022 (16:15)	Assignment 3 exercise session	TU1 Saab Auditorium
  04.05.2022 (12:15)	Lecture 9 (by Dr. Riikka Huusari)	Zoom
  04.05.2022 (16:00)	Assignment 3 due
  11.05.2022 (12:15)		Self study of review slides      (No zoom lecture)
  02.06.2022 (09:00)	Exam	Check Sisu for location

  
  

  Contents

  Inner product spaces, Kernels, Reproducing kernels, and RKHS. Introductory learning theory and Generalization. Empirical Risk Minimization, Uniform Convergence and Rademacher Complexity. Kernel Ridge Regression and Logistic Regression. Optimization and Duality. Margin-based methods and Support vector machines. Unsupervised learning including clustering, PCA and their kernel variants.

  Course personnel

  • Lecturer: Rohit Babbar
  • Course assistants: Petrus Mikkola, Adrian Müller, Mohammadreza Qaraei

  
  

  Grading

  The course can be completed by two alternative ways:

  • Assignments (max. 50 points) + Exam (max. 50 points) giving a grade 0..5. The assignment part has a weightage of 50% and exam of 50%. In aggregate, lowest passing points total is 40. 85 points will give the grade of 5.
  • Exam only (max. 50 points), giving a grade 0..5. 20 points will give the grade 1, 45 points will give the grade of 5.

  
  

  The better of the resulting two grades will be taken into account.

  Prerequisite knowledge

  • The course Machine Learning: Supervised Methods (CS-E4710), or equivalent knowledge, strongly recommended
  • Basics of probability and linear algebra
    
  • Python programming skills are recommended (We will use Python, but there will be an introductory python tutorial.)
    

  
  

  Learning Outcomes

  After attending the course, the student knows the basics of kernels, positive-definiteness, and RKHS. The courses also formally introduces the notions of generalization in machine learning by studying the principle of Empirical Risk Minimization and its consistency. The student knows how convex optimization methods can be used to efficiently train kernel-based and large-scale linear models. It is also discussed how to apply kernel based methods for unsupervised learning such as PCA.

  Language of Instruction: English

  Course Material

  • Lecture slides and exercises are the examined content
  • Further references will be provided during respective lectures

  Course position and Prerequisites 
  

  Course is MSc level in Machine learning, targeted to 1st/2nd year MSc students in Machine Learning and Computer science. Also suitable for PhD studies.