We have the second Cambridge Julia Meetup.
This time with a planned speaker, as well as time for adhoc short talks after.
Come for a chill meetup, drink some coffee and meet other Julia users.
Pizza will be provided at 7:00pm,
and we will aim to start the main talk at 7:45pm.
For the short adhoc talks they will be occuring shortly after our main speaker (details on that below).
Feel welcome to give short demonstrations of anything you think is cool.
There will be plenty of time for anyone who wishes to give a 5-10 minute half-planned talk, if one is so inclined.
## Title: You have data and I have distributions: a talk on Turing.jl and Bijectors.jl
You have data? What a coincidence; I have distributions! I know distributions. I have the best distributions.
If you come to this talk I'll tell you a bunch of different ways to get a distribution for your data. Some distributions will be (asymptotically) exact and some approximate. Nonetheless, if you want a distribution for you data this is definitively the place to be.
We'll look at using
- Bayesian inference to obtain distributions over latent/"hidden" variables given data,
- mean-field variational inference to _approximately_ do the same,
- neural networks (a particular class called normalising flows) to approximately do the same but _even better_, and more.
Then we'll finish off with a ludicrous display of using learned normalising flows inside a generative model which we will then run exact Bayesian inference on! It will be wild.
No idea what I'm talking about? Don't worry; we'll decode this alphabet-soup together.
This will also be a nice example of how Julia allow composable and performant implementations without sacrificing convenience. If you thought Julia was awesome before this talk, be prepared to have your opinion be confirmed!
## Our speaker: Tor Fjelde
Hey. My name is Tor, like the onion browser or the norse god. Neither are fitting; I'm not a man of many layers nor am I long-haired and muscular. I'm currently working as a research assistant in the Machine Learning group at University of Cambridge where I a lot of my time on developing variational inference for the probabilistic programming language Turing.jl. In the more recent weeks my work has been on a different package focused on transformations of distributions, called Bijectors.jl. It's all related to probabilities, and it's all good fun.
And like you, I of course loooove Julia!
If you are thinking about visiting from London, Invenia Labs is a 20 minute walk from the train-station.
This meetup is just a day after the London Meetup.
Do both, it will be great.
Claim the event and start manage its content.I am the organizer