A series of colab / jupyter notebooks for demonstrating deep neural network architecture concepts.

Dependencies : Anaconda Tensorflow Jupyter numpy scikit-image keras scikit-learn Pandas Hidden Layer (experimental)

Tensorboard --logdir=./{directory}

Installation dependencies:

Anaconda: wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh bash Miniconda3-latest-Linux-x86_64.sh

Packages Tensorflow jupyter notebook numpy scikit-image scikit-learn keras (for importing images) Pandas

conda install tensorflow jupyter numpy scikit-image keras scikit-learn pandas hidden-layer

Tensorflow 1.X hosted on Google cloud Virtual Machine.

Mnist:

Recognize handwritten digits from 0 - 9.
1layer-dnn.pynb : a fully connected layer NN
linear -> 5 layer network example.pynb : 5 Layer CNN

Titanic:

Predict survivors on board the Titanic.
Feature engineering, data exploration, linear regression.

Housing:

Predict housing prices of cities from a dataset of features.  Feature selection, data exploration, feature engineering.

Airbus-ships:

Predict pixels which contain ships from satellite images. Image segmentation, Image classification.  Uses a 		hierarchical model of
Classifcation
Image segmentation

voice-recognition:

Predict speech out of 12 labels, up down left right...

subvocal:

Binary speech classification from EMG muscle readings.  Subvocalization predicts speech without the sound / voice signature using spectrogram translations.

Tensorflow 2.0 on google colab

lastname-country-identification:

Predict the country of origin of a persons lastname with LSTM.

reinforcement-value-policy iteration:

A simulation of value and policy iteration.
Qk+1(s,a)	= ∑s' P(s'|s,a) (R(s,a,s')+ γVk(s'))
for k ≥ 0
Vk(s)	= maxa Qk(s,a)  for k>0