/Unsupervised-Learning

We look at the structure of data and not its corresponding output Y. By looking at the structure of data we can better understand the given data. We can do this using various unsupervised machine learning algorithms, but in this case I am using an autoencoder to model the structure of the inputs.

Primary LanguageJupyter Notebook

Homework-3-Machine-Learning-Unsupervised-Learning

We look at the structure of data and not its corresponding output value, just the data matrix. By looking at the structure of data we can better understand it. In this project, I am reducing a 30 dimensional data set into K dimensions, and then back into its original 30 dimensions. This is game-changing! If I am able to reduce a dataset into K dimensions and I am able to blow it back up to 30 dimensions, then I have successfully compressed the data.

How am I compressing the data?

I am coding a neural network (auto-encoder) from scratch. This neural network has 3 layers: input layer, hidden layer, and an output layer. These layers are fully connected and the weights of each neuron is updated using stochastic gradient descent.

Other ways to compress the data

In addition to an auto-encoder you can also use PCA (principal component analysis) to reduce the dimensions of a dataset.