CS 539 (002), Fall 2020
Data Visualization for Machine Learning

School of Electrical Engineering and Computer Science
Oregon State University

Tue & Thu at 4:00 - 5:20pm
Milam Hall 234 (and/or Zoom)

Prof. Minsuk Kahng
Assistant Professor, School of Electrical Engineering and Computer Science
E-mail: [email protected]
Website: https://minsuk.com

Course Description

This course will introduce advanced state-of-the-art research on interactive data visualization for machine learning (ML). Data visualization has recently been used as a powerful way to understand and interact with ML models and systems. Students will learn how to design and develop interactive data visualization methods and tools for interpreting complex ML models (e.g., deep learning models), for a variety of users (e.g., ML researchers, practitioners like ML engineers and data scientists, non-expert learners). Students will also discuss how data visualization can help address several critical human-side issues in artificial intelligence (e.g., explainability, inclusiveness).

Here are some examples of work which we will study this term:

Learning Objectives

At the completion of this course, students will be able to:

  1. Understand fundamental and practical challenges of human factors in machine learning
  2. Design visual representations and human interactions for exploring and interpreting machine learning models and systems
  3. Create interactive data visualization tools for supporting various machine learning tasks
  4. Evaluate the usability and utility of data visualization for machine learning

Prerequisite

  • It is recommended that students have taken courses from at least one of the following three areas: machine learning (e.g., CS 534), information visualization (e.g., CS 458/519), or human-computer interaction (e.g., CS 565). It is not required, though. More specifically,
    • If your research area is AI and you are interested in applying data visualization for AI, you would be fine, even if you have not taken any visualization courses.
    • If your research area is visualization or HCI and you are interested in AI, you would also be fine.
    • If you are simply interested in AI and have not taken any ML, HCI, or visualization courses from your undergrad or at OSU, I suggest you take this course next time.
  • Grad standing in CS or permission of instructor is required.
  • This course involves some programming assignments. If you have no experience in writing computer programs (e.g., Python, JavaScript, Java), I suggest you take this course after you get more experience in programming.

Course Content and Tentative Schedule

We will first start with the basics of data visualization which covers the principles of visualization and implementation skills. After that, you will learn state-of-the-art research topics by reading relevant research papers, trying out existing systems, and discussing them in class.

Week Topic  
Week 1 Introduction of Data Visualization for Machine Learning
* Explainable AI (or ML), visual analytics, and challenges
 
Week 2 Data Visualization Basics (1)
* Principles of information visualization
* Implementation with D3.js or other libraries
 
Week 3 Data Visualization and Machine Learning Basics (2)
* Principles of information visualization
* Machine learning workflow
* Implementation with D3.js or other libraries
 
Week 4 Exploring ML Datasets
* Example-driven exploration and analysis
* Faceted exploration
 
Week 5 Model-agnostic Explanation and Analysis
* Global vs. local explanations (e.g., LIME)
* Embedding visualization (e.g., t-SNE)
 
Week 6 Model-specific Visualization (1)
* Image-based models (e.g., CNNs)
 
Week 7 Model-specific Visualization (2)
* Text-based models (e.g., RNNs)
 
Week 8 Human-in-the-Loop ML Workflow
* Model debugging, feature selection
* Interactive machine learning, AutoML
 
Week 9 ML Education, Responsible ML, etc.
* Bias and fairness
 
Week 10 Evaluation for Human-AI Interaction  
Week 11 Project Presentations  

How the course will be conducted

CS 539 is a hands-on course. You will be expected to actively participate in projects, in-class activities, and discussion. Because of this highly interactive nature, it is mandatory to attend classes.

The typical structure of every class will be as follow. You will read papers or materials before class. At the beginning of the class, you will conduct short quizzes (a.k.a. nano-quiz) on the readings, consisting of a few multiple-choice questions. A short lecture by the instructor will follow. For the rest of the class time, you will either work on in-class activities (e.g., visualization design exercises, discussion with other students, short programming exercises) or present papers.

Projects and Assignments

By the end of the term, you will write a short research paper on data visualization for machine learning, as a final product of your group project.

Your project will likely involve designing and implementating an interactive visualization tool. Though it is not required, we suggest you implement a tool that runs on web browsers and written in JavaScript, which is the state-of-the-art way to implement interactive visualization tools. The individual assignments will involve programming exercises that help you learn how to implement a web-based interactive tool. It is not required to be familiar with JavaScript, though. You would be fine if you have some programming experience.

Your project does not have to involve implementing a new tool. I encourage you to develop your own research topic based on your research, especially if you are a PhD student. For example, your project may primarily involve human subject studies, algorithms for explainable AI, or scalable data systems. You can discuss project topics with the instructor.

Grading

Your performance will be evaluated via projects, individual assignments, nano-quizzes, readings and presentations, and participation. There will be no exams. The distribution of grading will be as follow (subject to change):

  • Project: 35%
  • Individual Assignments: 20%
  • Nano-Quizes: 20%
  • Readings and Presentations: 20%
  • Participation: 5%

Textbooks

Required readings and papers will be provided. There is no required textbook. We will provide some optional readings later.

Announcements and Questions

We use Canvas for all announcements and submissions.

We use Zoom if our course is delivered remotely.

TA & Office Hours

The instructor and TA will hold office hours starting Week 2.

Classroom Policies

This class is our community. Every student should feel safe and welcome to contribute in this course, and it is all of our jobs to make sure this is the case. I will try to establish this tone whenever possible, but ultimately the responsibility for cultivating a safe and welcoming community belongs to the students---that means you! Fortunately, forming a safe and welcoming community is not too hard. A good place to start is to recognize (and continually remind yourself) of the following facts:

  • Your classmates come from a variety of cultural, economic, and educational backgrounds. Something that is obvious to you may not be obvious to them, and vice versa.
  • Your classmates are human beings with intelligence and emotions. This applies even when one or the other of you is posting anonymously. Rudeness and disrespect are unprofessional, and have no place in this course or in your career.
  • Your classmates are here to learn. They have the right to pursue their education without being distracted by others' disruptive behavior, or made uncomfortable by inappropriate jokes or unwanted sexual interest.

In addition, the OSU Expectations for Student Conduct apply.

In short, treat your classmates as respected colleagues, support each other when needed, have fun without spoiling it for anyone else, and everybody wins.

Students with Disabilities

Accommodations for students with disabilities are determined and approved by Disability Access Services (DAS). If you, as a student, believe you are eligible for accommodations but have not obtained approval please contact DAS immediately at 541-737-4098 or at http://ds.oregonstate.edu. DAS notifies students and faculty members of approved academic accommodations and coordinates implementation of those accommodations. While not required, students and faculty members are encouraged to discuss details of the implementation of individual accommodations.

Reach out for Success

University students encounter setbacks from time to time. If you encounter difficulties and need assistance, it’s important to reach out. Consider discussing the situation with an instructor or academic advisor. Learn about resources that assist with wellness and academic success at oregonstate.edu/ReachOut. If you are in immediate crisis, please contact the Crisis Text Line by texting OREGON to 741-741 or call the National Suicide Prevention Lifeline at 1-800-273-TALK (8255).