Coursera course beta testers

[Kevin Webster]

Hi everyone,

For the last few months I've been developing an online course for probabilistic deep learning with TensorFlow, which will be launched on Coursera in August (it will be third part of a three-part specialisation, the first two parts of which can be found here and here). This third course makes heavy use of the TFP library, and so is also an introduction to the library (at least, some parts of it).

As part of the launch process, this course will need to go through a beta testing phase prior to launch. This phase will take place in late July/early August and will last for a couple of weeks. 

I'm posting this message to ask if anyone would be interested in beta testing the course, please get in touch. It would be great to have the course tested by people who are already familiar with the TFP library. Testers can do as much or as little of the course as they have time for, and any feedback received is appreciated. Please feel free to email me directly at kevin....@imperial.ac.uk if you are interested.

Thanks!

Kevin

[Tridib dutta]

I know it is a bit late to respond to this post. But I am going to do it anyway. 

I am actually looking at the course you mentioned currently and needless to say, I learned a lot about using TFP. However, for beginners, some of the stuff could have been explained a bit more. For example, in the 1st Week's reading on VAE, in the section "A LOWER-VARIANCE ESTIMATOR OF ELBO", it is mentioned that E_{Z ~ q(z|x)}[log p(x|z)]  can be estimated by MCMC. But in the 4th week's tutorial, showing how to maximize ELBO, the tutor grossed over by taking just 1 sample. When I tried to actually take multiple samples and estimate this reconstruction error, I couldn't do it. The program is throwing errors because of the error in tensor shapes. I am desperately trying to fix it. So far couldn't. 

Also, the decoder distribution is taken as IndependentBernoulli(), but if I want a gaussian distribution instead, that is, 

if we assume that p(x|z) ~ N(mu(z), sigma**2I), ie the sigma is learnable but a shared parameter, independent of z, how to do I go about implementing it? 

But overall, I learned many things.

Thanks.

[Mohan Radhakrishnan]

Hello,

                      After finishing the course on Probabilistic graphical models I have been looking for a way to code that using TensorFlow. Will this

course help ? Nonetheless this course seems to be very useful.

Thanks,

Mohan

--
You received this message because you are subscribed to the Google Groups "TensorFlow Probability" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tfprobabilit...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/tfprobability/800e81d3-2e45-4bfe-ac12-76588a465135n%40tensorflow.org.