Outline
The following schedule outlines the different modules and topics under each module that we would discuss in this course over a span of 14 weeks (28 lectures). Relevant reading material and resources will be provided following each lecture.
Modules | Topics | Lecture content |
Module 1 | Geometric dimensionality reduction | Principal Component Analysis (PCA) |
| | Cannonical Correlation Analysis (CCA) |
| | Maximum Correlation, Discrete CCA and Kernel CCA for real-valued data, |
| | Alternating Conditional Expectation algorithm (ACE) and Non-linear dimensionality reduction methods |
| | |
| | |
Module 2 | Probabilistic dimensionality reduction | Generative models: non-parametric |
| | Generative models: mixtures of Gaussians ( EM algorithm and method of moments) |
| | Discriminative models: Mixture models (method of moments and tensor decomposition algorithms) |
| | Discriminative models: Mixture of Experts |
| | |
Module 3 | Neural networks | Neural networks (Have a look at the zoo!) |
| | Training of neural networks: Backpropagation algorithm and SGD |
| | Variational Autoencoders |
| | Generative Adversarial Networks |
| | |
Module 4 | Representation methods: NLP | Language models (KN and GT smoothing) |
| | Word2vec (Skipgram and CBOW) and
GloVe |
| | Doc2vec, Sentence2vec, Skip-thought |
| | Polysemy, Compositionality |
| | Deep learning pipeline for NLP |
Module 5 | Representation methods: Others | Graph2Vec |
| | Program2Vec |
| | Interpretable ML
|
|