Course Schedule — Autumn 2026

ESS 469/569, University of Washington. Class meets Monday, Wednesday, Friday, 10:00–11:20 in SIG 227 (Sieg Hall). Bring a charged laptop — the room has tablet-arm seating and limited power; lab sessions run the full 80 minutes. Instruction runs September 30 – December 11, 2026; there is no class on Wednesday November 11 (Veterans Day) or Friday November 27 (Native American Heritage Day), and Thanksgiving is Thursday November 26. Final examination week is December 12–18.

This page is the course-delivery layer on top of the book: which sessions cover which sections, and when graded work is due. The book carries the full depth; in-class delivery selects from it (see the 469/569 differentiation notes in each assignment).

Two design choices shape the quarter. First, the agent thread runs through the whole course rather than waiting for the end: what agents are and how we use them (week 2), turning your data skills on an AI’s claims (week 4), and building eval sets for agents (week 7) — so the course’s distinctive material is taught early and applied often, and the capstone gets a month of runway. Second, December contains no new core material: only application, clinics, one deliberate buffer session, and presentations, so nothing important is lost if a week slips.

The paper-pulse rhythm. From week 3 through week 10, most sessions open with two 4-minute student talks — one paper from the speaker’s reading-arc literature review, dissected against the class’s rubric-in-progress. Every student presents once (sign-ups open Mon Oct 5 on Canvas) and everyone in the room scores every talk through the standing peer-feedback survey: five rubric ratings plus one strength and one improvement, anonymized to the presenter, completion counted as participation. Pulse talks skip the two 4.5 lab days and check-in #1; the same survey instrument returns for the final presentations.

Graded-work calendar at a glance¶

Chapter 7 has no quiz: its outcomes (audience translation, downstream impact) are assessed directly by the final-project deliverables. Quizzes are timed, auto-graded on Canvas, open for the window listed, and drawn from scenario banks — they test judgment about situations, not recall.

Week by week¶

Week 1 — Open, reproducible science (2 sessions)

• Wed Sep 30 — Course introduction; why ML in the geosciences; open reproducible science (1.1). HW1 assigned; setup help continues in the week-1 install clinic (office hours), not in lecture. · slides
• Fri Oct 2 — Workbench lab: version control and environments (1.2–1.5, with 1.9 as the self-serve walkthrough); the pull-request dry run. · slides

Week 2 — Agents, then data (Ch 1 quiz Tue–Thu)

• Mon Oct 5 — Working with agents: policy and mechanism (1.8 + 6.1): what an agent is, where errors enter the loop, the badge system, disclosure. Reading arc stage 1 assigned (topic: opportunities and challenges of AI in your subfield, with the two anchor readings); paper-pulse sign-ups open. · slides
• Wed Oct 7 — Meet the data: definitions and formats (2.1–2.2), the data gallery, and a first real dataset (1.6–1.7). · slides
• Fri Oct 9 — Tables: pandas and dataframe preparation (2.3–2.4). · slides

Week 3 — Signals (HW1 due Mon)

• Mon Oct 12 — Arrays and gridded data (2.5); resampling and irregular data (2.6). · slides
• Wed Oct 14 — Statistical considerations (2.7); spectral transforms (2.8). · slides
• Fri Oct 16 — Filtering, gaps, timing errors: repairing real records (2.9). · slides

Week 4 — AI-ready data, then interrogate the AI (arc stage 1 due Wed)

• Mon Oct 19 — Synthetic data and the STA/LTA detection floor (2.10); feature engineering (2.11). · slides
• Wed Oct 21 — Dimensionality reduction (2.12); the AI-ready checklist and raster-to-station joins (2.13). · slides
• Fri Oct 23 — Critical-evaluation lab (6.2) — Chapter 2’s capstone: turn your new data skills on an AI’s claims and verify against the data, not against plausibility. · slides

Week 5 — Classic ML begins (Ch 2 quiz Mon–Wed; proposals due Fri)

• Mon Oct 26 — Supervision concepts; classification vs regression (3.1–3.2). · slides
• Wed Oct 28 — Clustering (3.3). · slides
• Fri Oct 30 — Binary classification and imbalance (3.4). Project proposals due. · slides

Week 6 — Classification, honestly (arc stage 2 due Wed)

• Mon Nov 2 — Multiclass classification (3.5); how the leaderboard and hidden test sets work. Classification leaderboard opens. · slides
• Wed Nov 4 — Logistic regression from scratch; are your probabilities honest? (3.6). HW-CML assigned. · slides
• Fri Nov 6 — Trees, forests, ensembles (3.7 + 3.9); reading importances without fooling yourself. 3.10 assigned as flipped reading (checked by the Ch 3 quiz). · slides

Week 7 — The two evaluation lectures (short week — no class Wed Nov 11; Ch 3 quiz Tue–Thu; arc stage 3 due Fri)

• Mon Nov 9 — Robust training (3.8), full session: why random splits lie; temporal, grouped, and spatial splits; the split ladder on data with planted truth. · slides
• Fri Nov 13 — Build an eval set (6.3) + disclosure and norms (6.4): the same fair-evaluation move, applied to agents. Reading arc stage 4 assigned — four weeks of runway. · slides

Week 8 — Check-in, then deep learning (Ch 6 quiz Mon–Wed; HW-CML due Fri)

• Mon Nov 16 — Project check-in #1: data-audit studio. Each team, five minutes: your data against the 2.13 checklist, your baseline, your split design — fresh from 3.8.
• Wed Nov 18 — Perceptrons to MLPs (4.0–4.2). · slides
• Fri Nov 20 — CNNs: images, waveforms, and the classical baseline (4.3). · slides

Week 9 — Sequence models and the training lab (short week — Thanksgiving)

• Mon Nov 23 — Sequence models: RNN, LSTM, attention (4.4). HW-DL assigned. · slides
• Wed Nov 25 — Model training lab, session I: data curation and label quality (4.5 sections 1–3). · slides

Week 10 — Uncertainty, forecasting, flipped workflows (Ch 5 quiz Mon–Thu; classification leaderboard closed Nov 24)

• Mon Nov 30 — Model training lab, session II: architecture, uncertainty, calibration, out-of-range behavior (4.5 sections 4–5). Ch 5 (5.1–5.5) assigned as flipped reading. · slides
• Wed Dec 2 — Time-series forecasting shootout, probabilistic forecasts, skill horizons (4.10). Forecasting leaderboard opens. Autoencoders and PINNs (4.6–4.7) as enrichment (569 required, 469 optional). · slides
• Fri Dec 4 — Ch 5 discussion (30 min, from the flipped reading) + communicating your science: audiences and downstream impact (7.1–7.2). HW-DL due. · slides

Week 11 — Clinics, buffer, presentations (Ch 4 quiz closes Mon; forecasting leaderboard closes Wed; arc stage 4 due Thu)

• Mon Dec 7 — Project check-in #2: presentation dry-runs + review-agent clinic. Rehearse; run your stage-4 agent on another team’s draft.
• Wed Dec 9 — Buffer session. Absorbs any slippage from the quarter; if nothing slipped, a synthesis discussion of the quarter’s question — opportunities and challenges of AI for geoscience, argued from the class’s own measurements (agent evals, peer-score agreement statistics) — or a deep dive by class vote (live agent-eval demo, PINNs, data-at-scale).
• Fri Dec 11 — Final presentations, session I.

Finals week — Presentations session II in the registrar-assigned exam slot (check the Time Schedule); final report and repository due Wednesday December 16.

Notes on pacing¶

• The agent thread is deliberately early: verification culture (6.1/6.2) is installed before homework habits form, 6.3 lands the week after 3.8 so fair evaluation reads as one idea applied twice, and stage 4 of the reading arc has a month of runway. The Ch 6 quiz follows 6.3/6.4 in mid-November.
• Notebook 4.5 is split across two lab sessions at its marked checkpoint; do not attempt it in one sitting.
• Flipped chapters (3.10, Ch 5) are examinable through their quizzes; the in-class time they would have used funds the check-ins and the buffer.
• Check-in #1 replaces a written midterm report: earlier feedback, at the moment it can still change the project.
• 469 students follow the diluted path noted in each assignment: Apply-level expectations on the uncertainty and agent outcomes, 4.6/4.7 optional, and assisting rather than leading final projects.
• Sections not lectured are not optional reading for 569 unless marked so; lectures select, the book carries the depth.