[Pycon] [new paper] "Sandeep Saurabh" - Probabilistic Programming and Bayesian Deep Learning

[info a pycon.it]

Title: Probabilistic Programming and Bayesian Deep Learning
Duration: 45 (includes Q&A)
Q&A Session: 15
Language: en
Type: Talk

Abstract: Bayesian deep learning has combined the benefits of probabilistic programming and deep learning to generate scalable model for uncertainty estimates.This approach is a leap towards making artificial intelligence truly intelligent alike humans

While creating model for solution of any real life scenario modelling uncertainty is really important. Traditionally, machine learning provides some approach such as Gaussian Processes for modelling uncertainty but these approaches can't scale to high dimensional inputs like images and videos. On the other hand deep learning algorithms are great for high dimensional inputs but struggles to model uncertainty.

The solution to the problem described above is the topic of my talk -Bayesian Deep Learning .The talk will be about how probabilistic programming, which has been good in modelling uncertainty and deep neural network, which can scale to high dimensional inputs like images and videos, can be combined into Bayesian Deep Learning which provides a deep learning framework and can also model uncertainty.The talk will cover different types of uncertainties and how to model them with Bayesian approach using python libraries(Edward and ZhuSuan) to create safe A.I.

I'll utilise the time of my talk as follows:

0-15 mins: Introduction to the problem in hand, of modelling uncertainty and why deep learning is not good enough for the problem. What is traditional probabilistic approach to the problem and it's shortcomings for high dimensional input, leading to the introduction of Bayesian Deep Learning.

15-30 mins: Explanation of basic mechanism and mathematics behind Bayesian deep learning with some examples.Description of various types of uncertainties such as Epistemic and Aleatoric which need to be modelled.

30-50 mins: Introduction to Edward python library and it's usefulness for creating Bayesian deep learning frameworks. Some example problems to show the working of this library . Introduction to ZhuSuan python library and how it is useful in creating Bayesian deep learning framework along with some examples to show it's workings. 

50-60 mins : Q&A

Tags: [u'ComputerVision', u'neural network', u'Statistical Learning', u'image-processing', u'scipy', u'mathematical-modelling', u'machine-learning', u'data']