Generative vs discriminative learning

Create to understand.

1h20 introductory course

What I cannot create, I do not understand - Richard Feynman.

From Naive Bayes

xyz

To latent variables and models

Intro PGM

Gaussian Mixture Models (GMM)

xyz

Theory: Probability review. Bayesian learning.

Hidden Markov Models (HMM)

Latent Dirichlet Allocation (LDA)

xyz

Usecase: Melody harmonisation. Ableton. Music.

Variational Auto Encoders (VAE)

VAE can be used to disentangle representations (style vs semantics), and consequently measure similarity between pairs, such as questions.

Learning to repeat, reformulate. Visualisation of learned representations after dimensionality reduction with PCA.
Learning to repeat, reformulate. Visualisation of learned representations after dimensionality reduction with PCA.

Summary (models, hypothesis, limits, link with information theory, probabilities, algebra)

Reference

Francis Bach. Introduction to Probabilistic Graphical Models. 2018.

Michel Deudon. Learning semantic similarity in a continuous space. Advances in neural information processing systems. vol 31. 2018.

Previous
Next