Machine Learning deployed at the edge

Table of contents

1. Tools to go deeper into ML knowledge
2. Toolkits for production AI application
3. Introduction to TinyML
4. Applications of TinyML
5. Deploying tiny ML in embedded systems
6. Image - Visual Wake Words context
7. Anomaly detection
8. Metrics
9. C++ Intro
10. Comunication protocols
11. Serial communication protocols
12. Debugging microcontrollers
13. Emmbeded frameworks
14. Board sensors documentation
15. Colabs
16. Extra

Tools to go deeper into ML knowledge

MIT Deep learning book - Recommended by OpenAI

Deep Reinforcement learning Tool - By OpenAi

Toolkits for production AI application

Coral AI

Machinery Anomaly detection with vibration

Open speech recording for keyword spoting apps

GAN: Generative Adversarial Networks AKA synthetic data generators

Real world project tutorials

Open Source Speech datasets. KeyWordSpot, FullSentenceSegments, etc.

Automatic hyperparameter optimization

Introduction to TinyML

how a Convolutional Neural Network can 'see' features in pictures

how to visualize what a CNN is learning?? Display the features the network extracts from its filters

Applications of TinyML

Saved models

Serving a model over HTTP

Tensorflow Lite

Tensorflow Lite Micro

Tensorflow JS

Running inference with saved models using tensorflow lite

Post training quantization

Quantization aware training

Model conversion to TF Lite

Model checkpoin files exploration - hands on

Model freezing in proto buffer format exploration - hands on

Deploying tiny ML in embedded systems

C++ Official documentation

C++ Simplified doc & tutorials

Google's C++ Style guide

C++ variables/functions/structures to control arduino

Full list of Arduino libraries

Serial communication protocols

Universal asynchronous receive transmit (UART)

I2C protocol

Serial peripheral interface (SPI)

Image - Visual Wake Words context

Depthwise Separable Convolutions - // Logic behind mobilenet

MobileNet Paper

Image preprocess and management in Tensorflow

buffered prefecth technique to improve data performance with TensorFlow API

Overfit and underfit

Data augmentation - improve model quality and diversity applying zoom and rotation to images

Pixel re-escaling layer

Functional layers API

Anomaly detection

MIMII Dataset paper: sound dataset for malfunctioning industrial machines

Link to MIMII Dataset

Variational autoencoders

Electro CardioGram dataset

Anomaly detection with k-means clustering

Building autoencoders in keras

Metrics

Confusion matrix

Accuracy

Precision

false positive rate

f1 score

cross validation

Colabs

Mask detection with transfer learning / mobileNet V1

Extra

Assigments, projects and more info from Harvard's onsite course

Harvard's maching learning community

Harvard's TinyML full course set

Board sensors documentation

Microphone: MP34DT05-A

Acceleromenter/gyroscope/magnetometer: LSM9DS1

Camera: OV7675

Emmbeded frameworks

Arduino

PlatformIO

ARM cortex - CMSIS

freeRTOS

mbed

STM32CubeMX