Fundamentals of tinyML

 Fundamentals of tinyML Harvard course

Always on ML use cases

Wake-word and keyword spotting

Knowledge distillation

Feature selection and swiss roll projection technique

tinyML FPGA implementation talks about paring down to Accumulator, removing FPU which are some of the challenges of ML in tinyML.

Learning Rate = stepsize

Loss function

tensor flow in google colab

Gradient Descent in tensorflow and tinyML google colab

Types of neural networks

Fun to run this program

had to add 

import tensorflow as tf

import keras

import numpy as np

It took it as

import tensorflow as tf

import numpy as np

from tensorflow import keras

2.2

flatten

activation

relu

softmax

dense

2.3 Exploring Machine Learning Scenarios

CNN

Recurrent Layers

LSTM

Pooling

Feature Extraction

Yoga pose classification using Deep learning

DNN

2.4 Building computer vision model

tensorflow datasets

Dropout to prevent overfitting

tinyML colabs

Recognizing handwritten MNIST digits on an Arduino with 2KB RAM using the LogNNet neural network: https://t.co/24RhP9tZRE #tinyML pic.twitter.com/QUwcmKuU8n

— Edge Impulse (@EdgeImpulse) May 18, 2021

Visualising and understanding convolutional Networks

Visualizing what convnets learn

Recap

Neural Network

Gradient Descent

Loss Function

Kernels, Filters

CNNs Vs DNNs

Training

Inference

Features

Overfitting

Data Augmentation

Preprocessing

Training Data, Validation Data, Test Data

Classification, Regression

Quantisation