IML
Interpretable Machine Learning (CMPSCI 6390)
Spring 2026 - Hybrid (online and in-person)
Instructor & Contact Information
- Instructor (Facilitator): Badri Adhikari
- Email: adhikarib@umsl.edu
- Class meets: Wed 6:55 - 8:10 PM at Clark Hall 215
- Office hours: Wed 5:45 PM to 6:45 PM (in person). Please email me before visiting during my office hours to avoid any potential waiting.
You are welcome to the Interpretable Machine Learning course. We are among a handful of departments in the entire country to offer this course. Interpretable Machine Learning is also my research area, and it is a pleasure to be teaching a course on a topic that I am very passionate about.
Machine and deep learning have great potential for improving products, processes, and research. But computers usually do not explain their predictions. This course is about making machine and deep learning models and their decisions interpretable. After exploring the concepts of interpretability, we will learn about simple, interpretable models such as decision trees, decision rules, and linear regression. The course will introduce model-agnostic methods for interpreting black box models, such as feature importance and accumulated local effects, and explain individual predictions with Shapley values and LIME. After discussing the limitations of these methods, the course will survey modern interpretability methods, mainly focusing on understanding black-box deep neural network methods. Emerging and more promising interpretability methods are concept-based, which involve detecting human concepts from network activations.
In computer vision, for example, when predicting if a wood-block tower will fall, we may ask the network “What human-understandable concept in the input tower block is causing the output?” Some recently successful interpretability examples, including the chess-playing network AlphaZero, network dissection, testing with concept activation vectors, automatic concept-based explanations, and concept bottleneck models, will also be discussed in the course. You will also have opportunities to survey new literature and present it to the class. In addition, you will have the opportunity to work in groups, focus on interpreting specific domains/models, and write manuscripts for potential publication.
Your success in this class is essential to me. If you need official accommodations, you have the right to have these met. If aspects of this course prevent you from learning or exclude you, please get in touch with me as soon as possible. Together we’ll develop strategies to meet your needs and the course requirements.
Teaching Philosophy: In my classes, I see a student as a ‘learner’ and myself as a ‘facilitator.’ Computer science and technology is a discipline that is evolving rapidly, and hence many learners see it as a difficult discipline. To my learners, I suggest that an effective strategy to learn the fundamentals is to follow an iterative process of reading, analyzing, and coding. Sadly, many learners either like to analyze or code but not both. While some of us enjoy developing the skills for critically assessing concepts and algorithms, many others enjoy programming and love building things. An effective computer science course should balance (a) theoretical knowledge to understand computer technology and (b) implementation skills to test and execute the theories and algorithms. I design course contents and assignments so that learners can improve both skills: analytical and programming. Learners with a rich programming experience may find this balance slightly easier but will have a platform to explore further. For many others who do not consider themselves expert programmers, taking such a course can become a rewarding experience.
Course Description
Course Name
Interpretable Machine Learning (CMPSCI 6390)
UMSL Catalog Description
This research course discusses classical, modern, and advanced methods for machine and deep learning interpretability. It focuses on the application, analysis, and evaluation of model-agnostic methods to interpret shallow and deep neural network models and their predictions. [3 credit units].
Prerequisite
CMPSCI 4390/5390 (Deep Learning) or CMPSCI 4340/5340 (Machine Learning)
Learning Outcomes
Upon completing the course students will be able to:
- Understand interpretability fundamentals: importance, taxonomy, scope, evaluation, and properties of explanations.
- Apply various interpretable models such as linear/logistic regression, decision trees, and decision rules.
- Understand global model-agnostic methods, including partial dependence, local effects, and functional decomposition.
- Apply, analyze, and evaluate neural network interpretation via methods such as learned features, saliency maps, and concept detection.
- Apply algorithms to interpret various types of convolutional neural network (CNN) models and generative models, such as the ones generating images from text.
- Analyze and evaluate methods for interpreting various types of attention network and transformer network models.
- Understand the limitations of existing interpretability methods and future directions.
Textbook (required, not optional)
Second Edition of “Interpretable Machine Learning” by Christoph Molnar; the book is available for free.
Course Topics
These topics also correspond to the book’s chapters.
- What is interpretability?
- Types of interpretable models
- Model-Agnostic methods
- Example-Based Explanations
- Global Model-Agnostic Methods
- Local Model-Agnostic Methods
- Neural Network Interpretation
Academic Honesty
Any academic dishonesty in this class will result in an F for the semester, and the case will be referred to the provost’s office for possible further disciplinary action, regardless of how trivial it is. Please don’t use another student’s assignment (or a solution on the Internet) to complete your assignment. Discussing the material is OK, but please do your work independently. You should complete the assignments alone, not together, and not in a group. If you have any questions about the lessons or the assignments, please get in touch with me, and I will point you in the right direction. Please read UMSL’s policy and keep yourself out of plagiarism. In your reports, all sources must be clearly cited. You should not copy-paste any content from the Internet without citing it. “If you didn’t write it (or create it), cite it.” Lastly, you are not allowed to print any of the quiz, mid-term, or final exam questions, or post them anywhere outside of the class.
Programming Language
Python3 is the programming language for this course; you are expected to use Python3 for all your classroom activities, assignments, and projects. You are welcome to use Google colab or your own hosted Jupyter Notebook to run your programs.
Course Resources and Expected Expenses
Students should plan for possible expenses up to $200 for the entire semester for professional versions of AI agents (such as ChatGPT Plus, Claude Pro, or other AI tools) and/or access to online deep learning platforms. While free alternatives exist, professional subscriptions may enhance your learning experience and provide access to more powerful tools for completing coursework effectively.
Assignments, Quizzes, and Exams
Assignments
This course will have several types of assignments: chapter readings, regular course assignments, project-related assignments, and drawing concept maps as chapter summaries. All assignments should be submitted via Canvas. I ignore any assignments submitted to me via email. If you need help finding the right place to submit in Canvas, ask me by sending an email instead of emailing me your assignments.
Citation Format
You can choose any standard format for citations, such as MLA and APA.
Quizzes
Some classes may begin with a short surprise quiz (5 to 10 minutes). Questions will be mainly from the book chapter readings, classroom lectures, and assignments. Please bring your laptops to the classroom so you can take the quiz on your computer. If for some reason we switch to an online setting, the quizzes will be proctored.
Feedback and Grading Timeline
You can expect to receive assignment and quiz feedback within a week after the due date. Sometimes it may take me longer during certain weeks when I travel to conferences. However, you are welcome to email me and request quicker feedback for a particular submission.
Exams
The mid-term exam and final exam will be in-class computer-based exams. You are expected to come to the exam after reading and having an understanding of all the previous lectures and assignments. The exams will be open book, open internet, open AI, open generative AI. You are basically allowed to use any resource to complete the exam. However, you will be expected to share your working process as a part of the exam.
Final exam: 13 May, 7:45-9:45 PM.
Exam Eligibility Requirements
To be eligible to appear for the mid-term exam, you must complete all assignments due before the mid-term. Similarly, to be eligible for the final exam, you must complete all assignments and projects due before the final exam. You may still physically appear for the exam if you have not met these requirements, but you will not receive a grade for that exam.
No Grading
Throughout the semester, I will provide you feedback on the quizzes, assignments, and projects. However, I will not assign grades to individual components as they are completed. You will receive your first official grade after the midterm exam and your final course grade after the final exam.
This approach is intentional. I want to reduce the tendency to treat learning as a transactional process (i.e., where work is completed primarily in exchange for points), rather than as an opportunity for genuine intellectual growth. By postponing formal grades, the course emphasizes understanding, experimentation, revision, and critical thinking over point accumulation.
You are expected to use feedback actively to improve your work and deepen your understanding throughout the semester. If you are uncertain about your progress at any point, you are encouraged to discuss it with me during office hours or via email.
Learning to Use AI Agents for Coding and Professional Work
This course explicitly encourages the responsible and skilled use of AI agents for coding, problem solving, research, and synthesis. I subscribe to the ideas of co-active emergence (Bryan P. Sanders) and co-intelligence (Ethan Mollick): when used well, AI systems can meaningfully augment human thinking rather than replace it.
Your success in this course—including on quizzes and exams—will depend on your ability to use AI tools effectively, critically, and transparently. This includes, but is not limited to, using AI agents to:
- write and debug code,
- explore alternative solution strategies,
- summarize and synthesize readings and research papers,
- connect concepts across lectures, assignments, and the literature,
- evaluate claims, detect errors, and verify results.
You are encouraged to actively use AI tools such as ChatGPT, Claude, Perplexity, Gemini, and similar systems. However, using AI does not reduce your responsibility. You must understand, justify, and be able to explain any solution, code, or analysis you submit. On exams, you will be expected to clearly document your reasoning process and how AI tools were used as part of that process.
Blindly copying AI-generated output (without understanding, verification, or attribution) will be treated the same as any other form of academic dishonesty.
Attendance, Due Dates, Late Policy, and Discussion Board Assignments
- Attendance will be recorded in almost every class.
- All assignments in this course will be due at midnight on Wednesdays. If you need special accommodations (for example, medical), please email me as soon as possible.
- Quizzes, assignments, and projects have their respective due dates (see Canvas).
- You can request a maximum two-day extension on any assignment or project submissions - for up to two submissions. If you email me a few hours before a deadline and I don’t reply to you immediately, you can assume that the extension is granted automatically (if you have not used your two-day extensions).
- Once you use up the extension days, subsequent late submissions will receive no points.
- Usually, in discussion boards you will need to submit your answer before seeing others’. You are disallowed from submitting ‘something dummy’ before posting your actual answer. When grading, your first submission (even if it is a dummy answer) will be graded as your answer.
Submitting Assignments
Please adhere to the guidelines below when submitting your assignments via Canvas. Non-compliance may lead to a deduction of up to 100% in your grade.
Complete Submissions Required
Merely providing links (e.g., to Google Docs or Google Colab) is insufficient. You must submit your work in a universally accessible format, such as PDF, DOCX, or HTML. Links may be included as additional, supporting material.
Human-Readable Formats
For submissions that include multiple files, if you are compressing them (e.g., in a ZIP file), ensure you also upload key documents (such as PDFs) separately for easier access.
Consolidate Images
When submitting multiple images, consolidate them into a single document, either as a PDF or in a Word document. This helps streamline the review process.
Zoom Meetings
If we ever need to Zoom (for class meetings or office hours), say due to severe weather, you are required to keep your video on.
Grade Composition
Your final course grade will be determined holistically at the end of the semester, taking into account all aspects of your work and engagement throughout the course. The approximate weight of each component is as follows:
| Scoring aspect | Approximate weight (%) |
|---|---|
| Assignments | 25 |
| Projects | 25 |
| Mid-term Exam | 25 |
| Final Exam | 25 |
Note: You are required to submit the course evaluation survey at the end of the semester to receive your final grade.
Grading Scheme
| Points (%) | Grade |
|---|---|
| 94 to 100 | A |
| 90 to 94 | A- |
| 87 to 90 | B+ |
| 84 to 87 | B |
| 80 to 84 | B- |
| 77 to 80 | C+ |
| 74 to 77 | C |
| 70 to 74 | C- |
| 67 to 70 | D+ |
| 64 to 67 | D |
| 61 to 64 | D- |
| 0 to 61 | F |
Additional Policies and Resources
- Here are additional course policies and student resources.
- If you feel sick (or if you test positive for COVID-19, etc.) please do not come to the class and inform me when you can.
Recommended Study Strategy
This is an active and intensive course. Please expect a minimum of 10 hours per week of work in this course. Some weeks may require much longer hours of effort. In addition to the time required to attend lectures, the course requires you to read the book chapters, watch several videos, and complete the assignments and projects. Every week, you should plan on reading the book chapters as soon as they are assigned. The best day to finish the reading assignments is the day after the class. Please connect with other students in the class and form study groups to discuss the assignment questions and solution strategies. For assignment deadlines, keep an eye on Canvas. To share your findings and explore what others have shared, please use the classroom discussion board. To help you, most weeks, I will send announcements to remind you of important deadlines and announce upcoming topics.
Note: This syllabus will be subject to change at the instructor’s discretion. Important changes will be announced to the class.