Teaching Guide
How to Use These Slides
Each slide deck is designed for 50–75 minutes of classroom teaching. The decks follow the textbook chapter structure exactly, so students can read the corresponding chapter before or after class.
Speaker Notes
Every slide has speaker notes containing:
- Timing guidance (
[N min]) for pacing the lecture - Teaching tips for explaining difficult concepts
- Common student errors to address proactively
- Discussion prompts for engaging the class
- Transition cues connecting each slide to the next
To view speaker notes in the PDF, use your PDF reader’s presenter mode. In the LaTeX source, notes appear as \note{...} commands on each frame.
Active Learning
Every deck includes 8–11 active learning moments, distributed across these types:
- Predict: Students predict the outcome of a scenario before seeing the answer. Builds intuition about system behavior.
- Exercise: A short calculation or analysis task (2–5 minutes). Often involves the Iron Law, roofline analysis, or memory estimation.
- Discussion: Open-ended question for pair or group discussion (3–5 minutes). Explores trade-offs and design decisions.
- Retrieval Practice: Students recall key concepts from earlier in the lecture or from previous chapters without looking at notes.
- Muddiest Point: End-of-lecture check where students identify the concept they found most confusing.
These moments are marked with distinct frame titles and styling so they are easy to find in the deck.
Customizing the Slides
Source Format
The slides are Beamer LaTeX, not locked PDFs. The full source is open and designed for customization.
slides/
assets/beamerthememlsys.sty # Theme (colors, fonts, layout)
vol1/<chapter>/
<chapter>.tex # Slide source
images/ # SVGs, photos, generated PDFs
vol2/<chapter>/
<chapter>.tex
images/
Makefile # Build system
Building
To build a single deck (PDF):
cd slides
make ch01 # Builds vol1/01_introduction PDFTo build everything (PDF):
make allTo build PowerPoint (PPTX) from PDFs:
make pptx-ch01 # Single chapter PPTX
make pptx-vol1 # All Vol1 PPTX
make pptx # All PPTX (both volumes)PPTX conversion requires poppler (brew install poppler on macOS, apt install poppler-utils on Ubuntu) and python-pptx + Pillow (pip install python-pptx Pillow). The PPTX files are image-based at 300 DPI — visually identical to the PDF, suitable for presenter mode and annotations, but not editable text.
Output goes to _build/vol1/ and _build/vol2/ (PDFs and PPTX side by side).
Editing Workflow
- Fork the repository
- Edit the
.texfile for the chapter you want to change - Rebuild with
make chXX - Review the PDF output in
_build/
SVG Figures
All diagrams are authored as SVG files in each chapter’s images/ directory. SVGs are the source of truth; the build system converts them to PDF via Inkscape for embedding in Beamer.
To modify a diagram:
- Edit the
.svgfile in any vector editor (Inkscape, Illustrator, Figma) - Follow the SVG style guide for colors and typography
- Rebuild the deck; the Makefile handles SVG-to-PDF conversion
Theme Customization
The file assets/beamerthememlsys.sty controls the global look:
- Accent color: Deep Rose (
#BE185D) used throughout - Header bar: Section navigation with centered labels
- Footer: Book logo + chapter context (left), page number + institution (right)
- Fonts: Metropolis base with custom overrides
Change the accent color or institution branding in the theme file and all decks update on rebuild.
Suggested Semester Plans
16-Week Plan: Volume I (Foundations)
| Week | Chapter | Topic | ~Time |
|---|---|---|---|
| 1 | 0 | Course Overview | 53 min |
| 2 | 1 | Introduction | 74 min |
| 3 | 2 | ML Systems | 70 min |
| 4 | 3 | ML Workflow | 79 min |
| 5 | 4 | Data Engineering | 66 min |
| 6 | 5 | Neural Network Computation | 68 min |
| 7 | 6 | Network Architectures | 63 min |
| 8 | Midterm / Review | ||
| 9 | 7 | ML Frameworks | 73 min |
| 10 | 8 | Model Training | 68 min |
| 11 | 9 | Data Selection | 63 min |
| 12 | 10 | Model Compression | 64 min |
| 13 | 11 | Hardware Acceleration | 66 min |
| 14 | 12–13 | Benchmarking + Model Serving | 142 min |
| 15 | 14–15 | ML Ops + Responsible Engineering | 130 min |
| 16 | 16 | Conclusion + Final Review | 51 min |
16-Week Plan: Volume II (At Scale)
| Week | Chapter | Topic | ~Time |
|---|---|---|---|
| 1 | 0 | Course Overview: ML Systems at Scale | 49 min |
| 2 | 1 | Introduction | 76 min |
| 3 | 2 | Compute Infrastructure | 72 min |
| 4 | 3 | Network Fabrics | 68 min |
| 5 | 4 | Data Storage | 69 min |
| 6 | 5 | Distributed Training Systems | 70 min |
| 7 | 6 | Collective Communication | 64 min |
| 8 | Midterm / Review | ||
| 9 | 7 | Fault Tolerance and Reliability | 68 min |
| 10 | 8 | Fleet Orchestration | 74 min |
| 11 | 9 | Inference at Scale | 71 min |
| 12 | 10 | Performance Engineering | 72 min |
| 13 | 11–12 | Edge Intelligence + Ops at Scale | 132 min |
| 14 | 13–14 | Security/Privacy + Robust AI | 142 min |
| 15 | 15–16 | Sustainable AI + Responsible AI | 131 min |
| 16 | 17 | Conclusion + Final Review | 47 min |
32-Week Combined Plan
For a two-semester sequence, teach Volume I in the first semester and Volume II in the second. Volume I provides the single-machine foundations that Volume II builds upon for distributed systems. The combined plan covers the full ML systems stack from data engineering on a single accelerator to fleet-scale governance.
Audio Lectures
Coming Soon: Audio Lectures Coming Soon
Each slide deck will have a companion audio lecture generated from the speaker notes written by the textbook author. These recordings will be:
- Synchronized with the slide deck for self-paced playback
- Based on author notes, not generic AI narration
- Designed for flipped classrooms: students listen before class, class time is for active learning
- Downloadable as MP3 for offline use