Embedded probabilistic programming

The paper and slides for Oleg and my presentation on embedded probabilistic programming are online.

I hope that I managed to convey why our DSL for probabilistic inference is useful and what it is like to use it. I didn't get to any detail about how the language is implemented, but it actually is closely related to the paper "J is for JavaScript" by Olivier Danvy, Ian Zerny, and me, presented at the beginning of the conference. Our probabilistic DSL is yet another example of a language whose interpretation requires unconventional (non-sequential) control structure. We prefer to embed this language in as uncreative and homomorphic a way as possible, and delimited control operators indeed reduce the creativity required of the implementors (and the overhead imposed by the implementation).