Tired of waiting forever for MCMC chains to converge? We experimented with using Pathfinder VI to initialize HMC and get early model diagnostics. mlcolab.org/public-events/… #bayesian #probprog #probml

@AndrewC_ML @willarvey @ValentinDeBort1 @tom_rainforth @ArnaudDoucet1 I am down to discuss its connections to graphical models, simulator based inference/ #probprog compilers such as github.com/plai-group/dap… also over a coffee. Feel free to reach out 😀

Are there any #probprog/#bayesian folks with a @Garmin watch or similar who've done interesting modeling work with the data?

Are you interested in composable Bayesian inference, especially in Julia? It would help me to have more active discussions about this for Tilde. Things like •MCMC as iteration •nested models •models as latent+pushforward #bayes #probprog #julialang github.com/cscherrer/Tild…

1997: Started using #Python 2008: ditched cable, got a Roku box and Netflix 2010: Started doing #probprog 2015: started using #julialang

Sacha @sachaker

3 years ago

1997: Started using #Python 2008: ditched cable, got a Roku box and Netflix 2010: Started doing #probprog 2015: started using #julialang

146 5 197 0 14

Way more interesting for me were the research interviews. These are very subjective, problem-based and topic-focused. This year mine were focused around #causality and #probprog and boy was that fun!

#probprog hot take: For trace-based inference with stochastic control flow, branch traces should be in a sum type. That way, simulating the latents can be strongly typed and type-stable. Partly hot because I haven't fully convinced myself yet 🔥

#probprog book!

Jan-Willem van de Meent @jwvdm

4 years ago

#probprog book!

6 150 709 0 231

Download Image

International Conference on Probabilistic Programming 2021 (#ProbProg) is online Wed Oct 20 2021 - Fri Oct 22 2021. Here's the schedule: probprog.cc/schedule/ #ProbabilisticProgramming

Looks like big tech interest in probabilistic programming is picking up #probprog

A hierarchical model cannot ‘compress’ training data for efficient inference on future observations. Is there a way to learn from data in probabilistic programs which is both Bayesian in nature and computationally efficient? dtolpin.github.io/posts/h2typ/ #probprog #bayes

Gen (gen.dev) is quite good, and it works! However, the opposite of ‘flexible‘ inference is ‘black-box’ (rather than ’stiff’). Stan's inference is ‘very black-box’. Gen's inference is ‘very flexible‘. Either is good, for different purposes. #probprog #stan #gen

@roydanroy That's very surprising, makes me curious about their view on #probprog in particular. Maybe they just see it as a thin wrapper over the numerical algorithms?

@Martin_Trapp "Publications" in academia, "inertia" outside it. Same problem for #julialang , same problem for #probprog

Really impressed so far with @zulip. Like @SlackHQ but - open source (no scrolling into the void) - LaTeX-friendly - better syntax highlighting - better threading model means more bandwidth per channel See you in the #julialang #probprog channel julialang.zulipchat.com

Research data is an essential foundation for scientific work. How can #probprog, #neurosymbolic, and #AI help to understand this rich data? Happy to be appointed to the working group "Handling of Research Data" of the Wissenschaftliche Kommission Niedersachsen. @TUDarmstadt

Style file and guide: it's here! #CallforPapers #CallForSubmissions #ProbProg Web and email update with this info to follow shortly. probprog.cc/2020/probprog-…

"Big tech" investment in probabilistic programming: - @GoogleAI (@TensorFlow probability, Edward) - @FacebookAI (Prophet, HackPPL) - @UberAILabs (Pyro) Others? #probprog

Fun or awful #probprog challenge: Derive a random forest (or similar) as a posterior sample. Prior is a single decision tree on a single bootstrap of the rows. Has anyone done this? Sounds like a good one for @TuringLang

"Probabilistic programming" is misleading - there is little 'probabilistic' in probabilistic programming. Should be "inferential programming" instead - a programming paradigm which embraces inference. The term is taken (en.wikipedia.org/wiki/Inferenti…) but lays unused. #probprog