Variational Inference Lecture I|Probabilistic Modelling|Machine Learning
6.6 هزار بار بازدید -
4 سال پیش
-
#MachineLearning
#MachineLearning #MachineLearningAlgorithms #DataScience #womeninmachinelearning #AI #empowerment #education #womeninAI #mathformachinelearning #math #deeplearning #deepneuralnetworks #convnets
In this lecture, we present an introduction to variational inference starting from scratch and assuming basic knowledge in probability distributions. We describe why Bayesian modelling is important and explain how one can deal with intractable integrals arriving from marginal distributions. Variational inference is one such technique that we dive into and derive the ELBO. With that achieved, we then detail an algorithm that can be used for fitting data in generative models. We go through the math step-by-step.
Topics we detail include:
0. Motivation on why we need Bayesian modelling
1. Marginalisation and Marginal Likelihood
2. Introducing Bayesian mixture of Gaussians model
3. ELBO derivation
4. A different form of writing the ELBO
5. The full derivation in the Bayesian mixture of Gaussians example
6. Update rules
7. CAVI and an implementation
8. Some results
Hope you like this video. Please note that we got the material from here:
1. Survey: https://arxiv.org/pdf/1601.00670.pdf
2. Code: https://zhiyzuo.github.io/VI/
There are more awesome lectures to come, where we will be detailing SVI, reparameterisation trick, and many other topics.
Some amazing authors doing work on VI & PM are -- many others not mentioned here:
David: http://www.cs.columbia.edu/~blei/
Shakir: http://scholar.google.co.uk/citations...
Zoubin: https://scholar.google.co.uk/citation...
Carl: https://scholar.google.co.uk/citation...
Marc: https://deisenroth.cc/
Stephen: https://scholar.google.co.uk/citation...
Best books on principled machine learning:
https://cds.cern.ch/record/998831/fil...
https://www.cs.ubc.ca/~murphyk/MLbook/
https://www.inference.org.uk/itprnn/b...
https://web.stanford.edu/~hastie/Pape...
http://www.gaussianprocess.org/gpml/c...
Kudos to:
Vincent Moens: https://www.linkedin.com/in/vincent-m...
https://scholar.google.com/citations?...
Rasul Tutunov: https://scholar.google.com/citations?...
Thanks for all the help.
Slides can be found here: https://docs.google.com/presentation/...
Please share, like and subscribe. For any questions please use: @hbouammar
In this lecture, we present an introduction to variational inference starting from scratch and assuming basic knowledge in probability distributions. We describe why Bayesian modelling is important and explain how one can deal with intractable integrals arriving from marginal distributions. Variational inference is one such technique that we dive into and derive the ELBO. With that achieved, we then detail an algorithm that can be used for fitting data in generative models. We go through the math step-by-step.
Topics we detail include:
0. Motivation on why we need Bayesian modelling
1. Marginalisation and Marginal Likelihood
2. Introducing Bayesian mixture of Gaussians model
3. ELBO derivation
4. A different form of writing the ELBO
5. The full derivation in the Bayesian mixture of Gaussians example
6. Update rules
7. CAVI and an implementation
8. Some results
Hope you like this video. Please note that we got the material from here:
1. Survey: https://arxiv.org/pdf/1601.00670.pdf
2. Code: https://zhiyzuo.github.io/VI/
There are more awesome lectures to come, where we will be detailing SVI, reparameterisation trick, and many other topics.
Some amazing authors doing work on VI & PM are -- many others not mentioned here:
David: http://www.cs.columbia.edu/~blei/
Shakir: http://scholar.google.co.uk/citations...
Zoubin: https://scholar.google.co.uk/citation...
Carl: https://scholar.google.co.uk/citation...
Marc: https://deisenroth.cc/
Stephen: https://scholar.google.co.uk/citation...
Best books on principled machine learning:
https://cds.cern.ch/record/998831/fil...
https://www.cs.ubc.ca/~murphyk/MLbook/
https://www.inference.org.uk/itprnn/b...
https://web.stanford.edu/~hastie/Pape...
http://www.gaussianprocess.org/gpml/c...
Kudos to:
Vincent Moens: https://www.linkedin.com/in/vincent-m...
https://scholar.google.com/citations?...
Rasul Tutunov: https://scholar.google.com/citations?...
Thanks for all the help.
Slides can be found here: https://docs.google.com/presentation/...
Please share, like and subscribe. For any questions please use: @hbouammar
4 سال پیش
در تاریخ 1398/11/19 منتشر شده
است.
6,693
بـار بازدید شده