Semester V · Professional Core Elective · AI255TBD · 45 classes in five units · decks go up as the course moves
The model was trained in a data centre. It has to run on something that fits in your hand.
Edge AI is everything that has to happen in between.
Each class below is a slide deck I use in the room. The decks go up as the course moves; the questions further down are where we are headed.
Start here
What the course is, the arcs, and how it is marked
Unit I · Classes 1 to 9
Where did the AI run? Not in a data centre: in a box stuck on the machine.
01What is Edge AI? Key termsopen deck 02Why Edge AI? Edge versus cloudopen deck 03Applications of Edge AIopen deck 04Hardware and software componentsopen deck 05Architecture and design patternsopen deck 06The data pipeline: ingestion, processing, storageopen deck 07Embedded ML and TinyMLopen deck 08Digital signal processingopen deck 09Challenges and opportunitiesopen deckUnit II · Classes 10 to 18
Open the box: the sensors that feed a model and the processors that run it.
10Sensors, signals and sources of dataopen deck 11Acoustic and vibration sensorsopen deck 12Visual and scene sensorsopen deck 13Motion, position, force and tactile sensorsopen deck 14Optical, electromagnetic, environmental, biological and chemical sensorsopen deck 15Hardware architecture, microcontrollers and DSPsopen deck 16System-on-chip and deep learning acceleratorsopen deck 17FPGAs, ASICs and edge serversopen deck 18Multi-device architectures, devices and workloadsopen deckUnit III · Classes 19 to 27
From a data stream to a model that fits, and how to tell whether you should build it at all.
Unit IV · Classes 28 to 36
The model does not fit. Make it fit without making it worse.
Unit V · Classes 37 to 45
Scope it, build it, ship it, and prove it works in the real world.
Background · optional
For anyone who has not met a neural network yet: from a straight line to a neuron.
P1From a line to a neuronopen deck P2Will I passopen deck