Write your awesome label here.
Understand where a model’s energy, carbon and water footprint is decided, from dataset to retirement.
Ten modules on the training side: six routes to a working model, sizing and data efficiency, the training run, search and evaluation, hardware-aware training, when and where to run, compression, and lifecycle decisions, each amortized over the model’s serving life.
7h 30m
10
30 questions
80%
Certification
Green AI for ML Practitioners
How the course works
One equation, ten modules
(E × I + M) ÷ R, with each module naming the term or multiplier it moves and amortization deciding which move is worth it.
Patterns rather than tips
More than thirty catalog entries, each stating the term it moves and whether it saves once or on every inference.
The arithmetic is shown
Break-even volumes and payback periods, including the cases where a technique does not repay.
What this course helps with
Know which decision moves which term
One expression runs through the course, and each module names the term or multiplier it changes, so the choices are comparable.
Compare a one-off saving with a recurring one
Amortization over expected serving volume, which is what decides whether a reduction is worth making.
Count the work that is usually missed
Search, evaluation and synthetic generation frequently exceed the training run itself and are rarely in the figure.
Who this is for
- ML engineers, applied scientists and research engineers
- Data scientists who train, fine-tune and retrain models
- ML platform teams setting defaults for others
- Prerequisite: SCI for AI Practitioner, or the ability to produce a measured figure
Related courses:
Assumes you can already produce a measured figure. SCI for AI Practitioner covers that first.
What you will work with
The six routes
Reuse, retrieve, merge, adapt, fine-tune or train, with the calculation that ranks them for your volume.
The multiplier chapter
Search and evaluation budgets, where much of the training-side compute actually goes.
Placement
Time- and region-shifting, and why an intensity reduction is reported differently from an efficiency gain.
What you will be able to do
Course Lessons
The full course
01
Why Water, and Why You
Software feels dry, but the buildings it runs in use water for cooling, the electricity it consumes was made using water, and the hardware was manufactured with it. This module shows those three connections and closes the gap between the people who can see the water and the people who change the software.
module-01
02
The Four Layers: building, electricity, hardware, and the stress conversion
This module gives the three ways software uses water their proper names, as the draft SWI specification uses them, and adds a fourth layer that converts litres into how much each litre matters. It ends with the only line of arithmetic in the course.
module-02
03
Consumption and Withdrawal: used up vs borrowed
Water figures come in two kinds, and confusing them is the most common mistake in reading them. This module teaches the distinction between water used up and water borrowed and returned, and the rule that the two are never added together.
module-03
04
Reading a Water Number
This module teaches no calculations. It teaches how to look at any water figure, whether a provider report, a news headline or a colleague’s slide, and know what it does and does not say.
module-04
05
Stress-Adjusted Water: a liter is not a liter
The same liter counts for more where water is scarce and for less where it is plentiful. This module turns that idea into arithmetic using a published factor per place, giving the conversion layer its full treatment at foundation depth.
module-05
06
Small Things That Save Water
The earlier modules built a vocabulary; this one turns it into actions small enough to take this week, organised by role. None of them needs a water meter, a budget or permission to run a project.
module-06
07
Final Assessment
A final assessment of 12 questions - 80% to pass, 30 minutes, retake available.
Final Assessment
Assessment & Certification
Time to complete