Machine Learning Scientist Reinforcement Learning - Probabilistic Modeling

Joined October 2019
Missed the Machine Learning in Science conference? Catch the recordings now on our YouTube channel! youtube.com/playlist?list=PL…
4
9
412
Happy to announce I passed my viva for my thesis "Dimensionless Bayesian Model-Based Reinforcement Learning" I thank the examiners Dr Christos Anagnostopoulos and Prof Carl Edward Rasmussen for their comments and insightful discussions.
2
4
1
15
706
And many thanks to my supervisors @MurraySmithRod @ssteinresearch @bjornsandjensen for all your help and patience during these PhD years, I could not have done it without you.
1
3
106
Some teaser before the thesis is published: 2 RL agents trying to solve the cartpole problem with a shifted pole length. Both are trained in the same single environment (with length of 1 meter) but only the one trained in a dimensionless MDP (right fig) transfers succesfully.
1
99
Thank you to our speakers, Dr Paul Kirkland (@PaulKirkland87) and Dr Giulia Marcucci (@GiuliasNLWorld), for their fascinating talks today and thanks to all the members of audience as well! Stay tuned for more talks to follow! youtube.com/watch?v=QlYuGPda…
3
1
6
795
Sample efficient model-based RL with Gaussian Processes does not get old:
9
551
Valentin Charvet retweeted
This is the dream team (plus some online that couldn't make it) that made the first ever @MLinScience conference happen. Huge thanks to speakers and poster presenters for presenting their work and the sponsors that helped fund it!
1
3
23
We hope everyone is enjoying the conference! Here's a few of the photos we've snapped over the course of the day 📸
4
8
Starting the last keynote talk of @MLinScience conference with @MichelBesserve
1
4
Our main talks are split up into three applications of machine learning in science: • 🏥 Medical Applications • 🔬 Physics Applications • 🖥️ Theoretical Machine Learning & Neuroscience
1
1
1
Very excited to see our first conference set up for tomorrow! We are sold out but you can join online from here uofglasgow.zoom.us/j/8801059…
The conference starts in just over 14 hours⌛️ We're looking forward to welcoming you all tomorrow for #mlis22 at @UofGARC! An outline of the schedule tomorrow with all of our exciting talks: mlisconf.org/schedule.pdf
6
#dallemini Brian Jones doing Gangnam style in Disneyland
1
Valentin Charvet retweeted
👀 Sneak a peek at the deepest & sharpest infrared image of the early universe ever taken — all in a day’s work for the Webb telescope. (Literally, capturing it took less than a day!) This is Webb’s first image released as we begin to #UnfoldTheUniverse: nasa.gov/webbfirstimages/
4,618
77,816
21,170
317,337
We finally have the first look at our Milky Way black hole, Sagittarius A*. It’s the dawn of a new era of black hole physics. Credit: EHT Collaboration. #OurBlackHole #SgrABlackHole Link: eventhorizontelescope.org/bl…
436
9,319
2,810
26,955
When Monte Carlo Tree Search diverges
Anyone else experience analysis paralysis?
1
Today in @nature, with @EPFL, the first deep reinforcement learning system that can keep nuclear fusion plasma stable inside its tokamaks, opening new avenues to advance nuclear fusion research. Paper: dpmd.ai/fusion-paper
52
1,164
387
4,430
Valentin Charvet retweeted
An important intuition about Stochastic Gradient Descent (SGD) is that when you are far from the solutions, individual stochastic gradients will more or less point in the same direction, but at optimum, since their average is 0, they will all point in different directions !
4
56
6
450
Great to see the work out at last @gw_hunter ! Well done to Francesco Tonoloni, @CMessengerGW, & Ik Siong Heng! A great collaboration between CS & Physics! @IDAglasgow @GlasgowCS @UofGPhysAstro @QuantIC_QTHub
I'm beyond excited to announce work which was a large part of my thesis has now been published in @NaturePhysics! We show for the first time that a form of #ML can provide estimates on #GW parameters in a fraction of the time of traditional methods. nature.com/articles/s41567-0…
1
4
9
Fortunately we have Monte Carlo methods
- dériver : faire sortir le dentifrice du tube - intégrer : faire rentrer le dentifrice dans le tube
1
Thanks once again to our speakers, @JEinsle and @rory_morrison, for presenting at our first in-person seminar in almost two years! If you have a university email, the recordings for these talks will soon be available to view online.