Harvard CS181 Machine Learning: Your 2026 Roadmap Through 7 Homeworks (With a 4-Year Comparison)

TL;DR CS181 2026 is A3 with hw0–6 as the weekly clock (no public recordings); 2025 adds a practical, 2024 has two midterms, 2023 was taught by Weiwei Pan. Start with HW0, then follow hw1→hw6.

#harvard #cs181 #machine-learning #learning-path #homework #practical #textbook

Table of Contents

TL;DR
Four-year comparison (one table)
Prerequisites and the HW0 gate
How the homework chain runs (2026 main)
Materials and the weekly clock
How to read this series
References

Series: Harvard CS181 Weekly Guides (0 / 3)

Harvard CS181 HW0: Do These 4 Problems First — They Tell You What to Patch

🌏 中文版

⚠️ Edition: 2026 is primary from the CS181 2026 site and s26 homeworks; 2025/2024/2023 are对照 via cs181-web-2025/2024/2023 and cs181-s25/s24/s23-homeworks. Google Sheet schedules are deleted/empty for all four years — this series uses homework numbers as weeks.

TL;DR

Can I self-study? CS181 2026 is A3 (hw0-6 + notes + sections + textbook, all learning will be in-person with no public recordings; Gradescope/Ed require enrollment), as rated in the Harvard AI/ML Course Map.
How to follow: Do the HW0 readiness check first (due 2026-02-02, 4%), patch the weakest of the four problems, then follow hw1→hw6; use the 2025 practical as the capstone backup.
Four-year delta: 2025 adds practical 6%, 2024 has two midterms and no final, 2023 was solo-taught by Weiwei Pan.

Four-year comparison (one table)

Year	Site	Instructors	Grading	Homework dir (verified via `api.github.com`)	Note
2026	cs181-web	Alvarez-Melis / Du + Head TFs Russell Li / Elvin Lo	syllabus: `hw0 4% + hw1-6 11% each + midterm 15% + final 15%`	s26 homeworks: `hw0-6` 7 dirs	`hw6` = Sequential/MDP/RL + Autoregressive
2025	cs181-web-2025	Doshi-Velez / Alvarez-Melis + Preceptor Papon	`hw0 4% + hw1-6 10% each + practical 6% + midterm/final 15% each`	s25 homeworks: `hw0-6 + practical` 8 dirs	hw3–5 reshuffle before
2024	cs181-web-2024	Doshi-Velez / Alvarez-Melis + Head TFs Badrinath/Cai	`hw0 4% + hw1-6 11% each + two midterms 15% each` (no final)	s24 homeworks: `hw0-6` 7 dirs	Two-midterm year, `hw0 due Jan26,2024`
2023	cs181-web-2023	Weiwei Pan solo `TTh 2:15 SEC 1.321`	see `cs181-s23` practical grading	s23 homeworks: `hw0-5 + practical1`	Only 6 hws, `T*_TestCases.py` naming

Prerequisites and the HW0 gate

Four years share the same prereqs (syllabus): beyond CS50 Python + STAT 110 + calculus + linear algebra (AM 22a / Math 21b), STAT 111 / CS 51 not required. HW0 warns During the term, the staff will be prioritizing support for new material... it might be prudent to postpone. Use the HW0 guide: linear y=Xw (x1≠x2 ↔ X invertible), calculus, probability, and OLS code — the slowest problem is your week-1 patch (MML Book, STAT 110).

How the homework chain runs (2026 main)

Verified by reading hw0/hw1/hw6 (30-80 lines) and hw2-5 titles + Intro snippets from s26 homeworks:

HW0 Modeling Linear Trends (due 2026-02-02) — four-in-one review
HW1 Regression — NN/kernel/linear regression, earth_temperature_sampled_train/test.csv (800k-year ice core)
HW2 Classification and Bias-Variance — plus uncertainty quantification
HW3 Neural Networks and Kernels — kernel ridge → weighted sum, Section 4
HW4 Representation Learning, Transformers, Non-parametric — Attention(Q,K,V)=softmax(QK^T/√dk)V, variance 1 after /√dk
HW5 Clustering, PCA, SSL — cluster centers vs PCA images
HW6 Sequential Models and Decision Making — Kalman, Gridworld policy/value iteration, Swingy Monkey Q-learning, autoregressive KV cache / speculative decoding (due 2026-05-01)

2025's hw3 Bayesian / hw4 SVM / hw5 EM are the对照; the missing practical (Kaggle-style) lives in cs181-s25-homeworks/practical and cs181-s19-practicals.

Materials and the weekly clock

Textbook: cs181-textbook (senior thesis, 370 stars, 13 chapters, Textbook.pdf 3.59 MB)
Sections: syllabus says flipped classroom, section cycle restarts each Tuesday, solutions will be posted, but cs181-section only has s17-19 — 2025/2026 are not in version control; schedule only shows S0 Math review, Google Sheet deleted for four years — this series uses homework numbers as weeks, not calendar weeks.
Submission: two Gradescope entries per homework (writeup PDF with assigned pages + LaTeX/code), per homework page.

How to read this series

Read this overview to choose the 2026 main line
Do HW0 and patch prerequisites
Follow HW1→HW6 weekly — each post maps to hw*_release.tex/pdf/ipynb + data and marks 2026/2025 deltas
Use the 2025 practical as the capstone; for CS182 see the CS182 historical disclaimer (2026/2025 are A0, only F22 22-lecture is writable)