ldn ml_1 inf_froze

-solutionism -probabilistic programming language expresses probabilistic models essentially as comp progs that generate data, as simulators

-as puao.

...

-infer what hidden states should be -all used to writing computer programs where you run it forward and generate some computer algorithms -if you want to do Bayes Rule on comp progs, say, -...have simulator - now this is actual data I observe, the real data

-so how should I set params of simul so params consistent with ... calls to RNGs that consistent with output data

ummmmm come on now

"Where'd you find this...?"

-"doing Bayes' rule on a computer program is the above"

Agree to disagree.

-Church: a truly universal probabilistic programming engine.

-Church: sort of based on Scheme-like Lisp-like ventures

TURING! .jl!

-HMM just modify a couple lines code and hit infer

-lovely tutorial by Frank Wood at NIPS 2015

Look for Turing on GitHub you'll find our language

-a Bayesian NN in Turing

-Bayesian opt of black-box functions that are expensive to evaluate

-don't think of as opt algos but as sequential decision problem under model uncertainty of the function -where do we eval next?

3

13

1