the decision blog

We create a three layer neural net that can identify handwritten numbers and use a regularization technique called dropout to prevent overfitting

we can best understand how neural nets work by studying a small, simplified example

unsupervised learning algorithms such as Isomap and K-means can produce intuitive categorizations for unlabeled data

we can use distance-preserving maps to simplify and get a useful overview of our data

by rotating a noisy line we can better understand the meaning of principal component decomposition