New! New Scientist covered our work on human perception of chess move brilliance (ICCC paper). Also covered in a Chess.com Blog of the Month winner!
Photo of Michael Guerzhoy

Office: BA 2028

Email: guerzhoy@cs.toronto.edu

Phone: +1-416-978-7024

Address:
Division of Engineering Science
40 St. George Street, Room 2110
Toronto, Ontario M5S 2E4
Canada

NEW! I spoke to CBC Radio about Vibe Coding.

NEW! Automatically identifying chess moves humans perceive as brilliant (as opposed to merely strong), with Kamron Zaidi (EngSci 2T4). Paper: ICCC 2024, covered in New Scientist and chess.com.
NEW! A mysterious interaction between architectures and datasets when using the rotation pretext task in self-supervised pretraining, with Paul Yan (EngSci 2T6) and Amy Saranchuk (EngSci 2T5). Paper: NeurIPS 2025 Symmetry and Geometry in Neural Represetations workshop.
NEW! Learning a compact representation of microstructures using a custom VAE, with Andy Cai (EngSci 2T5), Sajjad Hashemi (Indy 2T4), and Noah Paulson (Argonne National Laboratory). Paper: ICCV 2025 Workshop on Efficient Computing Under Limited Resources.
NEW! How and whether modern vision architectures encode symmetry-based information in images, with Whitney (Zhi) Ji (EngSci 2T5). Paper: NeurIPS 2025 Symmetry and Geometry in Neural Representations Workshop.
NEW! A pipeline for gaining insights into complex diseases by training LLMs on social media data and analyzing their explanations, with Kexin Chen (MIE M.Eng. '24), Noelle Lim (EngSci 2T3), and Claire Lee (Princeton '20). Paper: Machine Learning for Health (ML4H) @ NeurIPS 2024.
NEW! Position information emerges in causal transformers without positional encodings via similarity of nearby embeddings, with Jason Zuo (EngSci 2T4) and Pavel Guerzhoy (UH Mānoa Math). COLING 2025, summary.
NEW! Testing the "Honour Culture" theory using social media data, with Juho Kim. Paper: SiCon wokrshop @ EMNLP 2024.

NΨ Nocturne 2T3: GPT is Coming to Class (as seen at NeurIPS Education Program 2025). NΨ Nocturne 2T4: Defying Gravity/Debugging Properly, Desafinado. NΨ Nocturne 2T3: re: Your Grades

About

I work in machine learning focusing on interpretability and representation learning, using interpretability to understand complex social phenomena, and AI applications, particularly in medicine. I teach computer science and machine learning in the Division of Engineering Science at the University of Toronto.

My last name is pronounced ger-JOY, with a hard "g", and with the "J" pronounced like the "s" in "measure."

Teaching

  • Data Science and Functional Programming in R Mini-Course Summer 2023
  • ECE324 — Machine Intelligence, Software, and Neural Networks Winter 2022, Winter 2023
  • ESC180 — Introduction to Computer Programming Fall 2009/2010/2014-2016, 2020-2025
  • ESC190 — Algorithms and Data Structures Winter 2021-2026
  • MIE490 — Mechanical and Industrial Engineering Capstone Fall-Winter 2022-2023
  • SML201 — Introduction to Data Science Spring 2020
  • SML480 — Pedagogy of Data Science (NEW!) Spring 2020
  • SML310 — Research Projects in Data Science Fall 2019, Fall 2018
  • SML201 — Introduction to Data Science Spring 2019
  • CSC411/CSC2515 — Machine Learning and Data Mining Winter 2018
  • CSC411 — Machine Learning and Data Mining Winter 2017
  • CSC180 — Introduction to Computer Programming Fall 2014/2015/2016/2020
  • STA303/STA1002 — Methods of Data Analysis II Summer 2016
  • C4M — Computing for Medicine Winter-Summer 2016
  • CSC321 — Introduction to Neural Networks and Machine Learning Winter 2016 (won the CSSU award for excellence in teaching)
  • CSC320 — Introduction to Visual Computing Winter 2015
  • CSC165 — Mathematical Expression and Reasoning for Computer Science Summer 2014

My Assignments Around the Web

I enjoy creating and sharing my assignments. I sometimes enjoy Googling my name to see who uses them.

Nifty Assignments at SIGCSE 2017
Used in CPSC231 at the University of Calgary (2017) and in CS2230 at the University of Iowa (2017). Originally designed for CSC180 (2010-2016), with Jackie C.K. Cheung and François Pitt.
Model AI Assignments at EAAI 2018
Used in CSC321 at the University of Toronto (2017) and at Hacettepe University in CMP722 (2017) and CMP784 (2018). Originally designed for CSC321 (2016), with Renjie Liao.
Model AI Assignments at EAAI 2018
Used in part at Udacity as part of the Self-Driving Car Engineer Nanodegree (2016-). Some materials used at the Hebrew University of Jerusalem (2017) and at Yale University (2018). Used at Hacettepe University in BBM406, CMP722, and CMP784 (2017-2018). Used at the University of Toronto in ECE521 (2018). Used by the Intel AI Academy (Week 6). Originally designed for CSC321 and CSC411 (2016-2017).
Nifty Assignments at SIGCSE 2018
Posted on NCWIT's EngageCSEdu. Used in part in the Deep Learning and Artificial Intelligence program at LMU Munich (2019). Originally designed for CSC180 and for contests at the University of Toronto (2014-2016). Used in COMP 202 at McGill University (2020).
Model AI Assignments at EAAI 2019
Used in part in CSC411/2515 at the University of Toronto (Fall 2018). Originally designed for CSC411/2515 (Winter 2018), with Lisa Zhang.
Model AI Assignments at EAAI 2020
Used at NYU 2024-. Originally designed for SML 201 (Spring 2019), with Stephen Keeley.
Tips, Techniques, and Courseware at ITiCSE 2020
Originally designed for SML 201 (Spring 2019), with Claire S. Lee and Jeremy Du.
ITiCSE 2025
Exercises for learning C pointers. Originally designed for ESC190.

Conference & contest organization

Co-chair, Nifty Assignments at SIGCSE TS (2023-2027)

International Scientific Committee, International Olympiad on Artificial Intelligence (IOAI) (2024-)

Co-chair, Symposium on Educational Advances in Artificial Intelligence (EAAI) 2021-2023, co-located with AAAI. Steering committee member, 2023-

(Inaugaural) Program Committee: Toronto Machine Learning Summit (2017-2018)

Media

TV appearances around Geoffrey Hinton's Nobel Prize win, including CTV News, Global TV, CityTV, and CBC Television (2024).

Interviewed by CBC Radio on vibe coding (follow-up article) (2025).

New Scientist covered our work on human perception of chess brilliance (2024), also featured in a Chess.com Blog of the Month winner.

Quoted in MIT Technology Review (2017) on teaching machine learning with TensorFlow; also in Business Insider.

Quoted in The Varsity (2017) on AI for literature search and careers in ML/data science; quoted in The Cannon on grading policies.

Internal UofT/Princeton stuff: UofT News on Computing for Medicine (2016), the RL for medication tuning project (2024), and the perception of chess move brilliance project (2024). Princeton CSML articles on the data visualization contest (2019) and on publishing peer-reviewed pedagogical materials (2020). Welcome pieces from when I started at Princeton CSML in 2018 and UofT Engineering in 2020.

Miscellania

Just for Fun

Back in grad school, I used to co-ordinate the weekly CSGSBS cookie breaks.

Derandomizing Bogosort: A Very Serious Webpage.

Re: Your Grades at the NΨ 2017 Nocturne Talent Show (lyrics).

How Deep is Your Love with Gradient Descent (note the lyrics: "How deep is your love? I really need to learn.") Source code.