DDCM Research Reports

Web view (GitHub Pages): https://azwanazamuddin.github.io/reports/

A wrapper for my PhD research reports on Dynamic Discrete Choice Models (DDCM) for activity-based travel demand.

The folder follows the HIIRM lab meeting template: each lab meeting has (1) a rolling thesis draft as the research overview document and (2) a per-meeting discussion doc in meetings/<semester>/.

Folder layout

reports/
├── README.md                              ← this file (landing page)
├── templates/                             ← HIIRM meeting templates (English)
│   ├── instructions.md                       how the meeting process works
│   ├── meeting-template.md                   blank meeting-day document
│   └── thesis-template.md                    blank thesis draft
├── thesis/                                ← rolling Master thesis draft
│   └── thesis-draft.md                       single source of truth — update before each meeting
├── slides/                                ← interactive presentation decks (Slidev, built static HTML)
│   └── ddcm/                                 main DDCM research deck — rebuild with: slidev build --base /reports/slides/ddcm/
├── ddcm/                                  ← standalone HTML reports
│   └── work_timing_analysis.html
├── meetings/                              ← per-meeting discussion docs
│   ├── 2025-fall/                            M1 Fall 2025 (archive)
│   │   ├── *.md                              meeting docs
│   │   └── presentations/*.pdf               slide exports for that semester
│   └── 2026-spring/                          M2 Spring 2026 (current)
│       ├── *.md                              meeting docs
│       ├── figures/*.png                     figures used in meeting docs
│       └── presentations/*.pdf               slide exports for that semester

Updating the slides deck: The slides/ddcm/ folder is a built Slidev output (not source). To rebuild after editing 4 - Projects/ddcm/ddcm_slides/slides.md:

cd "4 - Projects/ddcm/ddcm_slides"
npx slidev build --base /reports/slides/ddcm/
rsync -a --delete dist/ "../../../3 - Permanent Notes/reports/slides/ddcm/"

Then commit and push the updated slides/ddcm/ folder in the reports repo.

Two documents per meeting (HIIRM convention):

  1. Research overview → always thesis/thesis-draft.md. Do not duplicate per meeting — update the single draft, Git history preserves versions.
  2. Meeting-day discussion doc → a new file in meetings/<semester>/, based on templates/meeting-template.md. Concise: what to discuss, current situation, stuck points, confirmations.

The big picture

I am building a computational framework that makes activity-based DDCM estimation tractable at city scale, and re-specifying its utility function so that activity timing emerges from preferences rather than from hard-coded rules.

Two things have been bottlenecks in activity-based demand modelling:

  1. State-space explosion — a single city has ~145 million states when you multiply time, zone, activity, duration, mode, vehicle, and history. Estimating anything on that is a 1000-CPU-day / 6.7 TB problem.
  2. Timing is usually assumed, not estimated — classic specs bolt hard time windows onto the model (e.g. “work happens 6 am–10 pm”). So behaviour is partly an input, not an output.

My contribution has three pieces:

# Piece What it does
1 DDCM as a DAG Time only moves forward, so the state-transition graph has no cycles. Backward induction is a reverse topological sort over a DAG — which means a GPU can process each time level in one batched kernel.
2 Reachability pruning A forward BFS from feasible starts, with Hägerstrand space-time-prism constraints, keeps only states an agent can actually reach. 145M → 1.5M (~1%), no approximation.
3 μ(t) utility Each activity gets a time-varying marginal-utility profile μₐ(t). Trip-making emerges from the gradient. Timing becomes an output.

End-to-end result so far: full pipeline ~69 hours → 105 seconds (~2,400×), memory 6.7 TB → 6.5 GB, on a 144-zone Higashi-Hiroshima city network.

Current semester — M2 Spring 2026

Per-meeting discussion docs. Update thesis/thesis-draft.md before each meeting; add a new meeting doc per session.

Date Meeting Doc Slides Event / context
2026-05-11 Research Progress (M2, May 11, 2026) — Weekly progress. K=10 L-BFGS-B convergence (LL=−19,187, BHHH SEs), work timing gap analysis, 11 new lit reviews, JSPS outline finalized.
2026-04-27 Research Progress (M2, Apr 27, 2026) — Weekly progress. Gradient bug fixes (forbidden_masks, zone ID), c_change ridge diagnosis, cumulative utility diagnostic planned; JSPS/thesis writing updates.
2026-04-19 Master Thesis Progress (M2, April 2026) April_Seminar_Presentation.pdf M2 mid-semester progress presentation. Framework, μ(t), results, estimation diagnosis.

Thesis draft (research overview document): thesis/thesis-draft.md.

Last semester — Archive (M1 Fall 2025)

Older reports kept for reference. These predate the DAG / reachability / μ(t) reframing and are not the current state of the work.

Date Report Topic
2025-10-08 PUMCM for Activity-Based Models Not used. Early brainstorm on PUMCM.
2025-10-15 Ideas on TD Estimation on DDCM TD-learning ideas for DDC estimation.
2025-11-04 Readings & Implementation Plans Lit-review notes + implementation roadmap.
2025-11-09 SMASO-X Presentation Summary Summary of the SMASO-X talk.
2025-11-26 Tensor-Based Summary Update Early tensor-graph framework write-up.
2025-12-15 Documentation & Re-framework Algorithmic re-framework progress.
2025-12-24 Universal Graph & BI Optimization Shared-graph structure + BI tuning.
2026-01-07 RMDP Theory & Implementation Relational MDP theory + lifted solvers.
2026-01-27 Progress on Validation Plan Seven-layer validation framework.

Slide decks and PDFs from last semester: meetings/2025-fall/presentations/.

How to prepare for a meeting (HIIRM workflow)

  1. Update the thesis draft (thesis/thesis-draft.md) to reflect current progress. Mark unwritten parts with TODO:.
  2. Write a meeting-day doc — copy templates/meeting-template.md into meetings/<semester>/YYMMDD - <short title>.md and fill in:
    • What you want to discuss today (1–3 items)
    • Current situation (what you did / what slipped)
    • Definitions of any new terms
    • Your own thoughts on each discussion item
    • Where you are stuck
    • Things to confirm
    • Computational results, each with a link back to the generating code
  3. Commit and push to azwanazamuddin/reports (this repo).
  4. In the meeting, spend the first few minutes reading silently — no verbal explanation needed up front.
  5. After the meeting, fill in the “Action Items for Next Meeting” block at the bottom of the meeting doc.

Full instructions: templates/instructions.md.

  • Master thesis detailed outline — 3 - Permanent Notes/research-plan/MASTER_THESIS_OUTLINE.md
  • Research roadmap — 3 - Permanent Notes/research-plan/RESEARCH_PLAN.md
  • DDCM map of content — 5 - Indexes/index-ddcm.md
  • Active implementation — 4 - Projects/ddcm/
  • Editable slide source (Slidev) — 4 - Projects/ddcm/ddcm_slides/slides.md