Basics of Deep Learning

Improve neural networks by applying dimensionality reduction (pca), autoencoders, and careful data splitting (train, validate, test), balancing bias and variance with regularization and activation function choices.

Explore how convolutional networks process images through convolution and pooling to learn filters and feature maps, reduce input size, and end with a fully connected layer for classification.

Explore why recurrent neural networks capture context for sequences, and how long short term memory and gru gates overcome vanishing gradients to enable next-word prediction and time-series applications.

Explore word representations from one hot encoding to distributed word vectors, learn co-occurrence and PMI methods, and see how word2vec and skip gram capture semantics for vector arithmetic.