If anyone wants to jump into this, Josh Tenenbaum and Noah Goodman put together this amazing interactive book for learning probabilistic programming with Church: <a href="https://probmods.org/" rel="nofollow">https://probmods.org/</a>
> “It goes beyond image classification — the most popular task in computer vision — and tries to answer one of the most fundamental questions in computer vision: What is the right representation of visual scenes?<p>Can someone knowledgeable in graphics research explain the context that this question comes from?<p>If I am reading the question correctly, I infer that the question suggests that there exists a right way to reproduce the visual experience of reality. To me, this sounds like a question that is equally valid to have no answer (or many answers) in aesthetics, art, and philosophy, etc.
Yes, you can specify extremely powerful statistical models in only a few lines of code using probabilistic programming.<p>However, at this point, unless you design your program in a very specific way and use a lot of tricks, your sampler is very unlikely to converge, and you won't get any meaningful result without a gargantuan amount of computing power.
Also <a href="https://news.ycombinator.com/item?id=9363496" rel="nofollow">https://news.ycombinator.com/item?id=9363496</a> from yesterday.
No experience but looks like they are organizing a summer school on probabilistic programming languages. <a href="http://ppaml.galois.com/wiki/wiki/SummerSchools/2015/Announcement" rel="nofollow">http://ppaml.galois.com/wiki/wiki/SummerSchools/2015/Announc...</a>
Ah, so they wrote a DSL just for doing Bayesian inverse-vision models, and apparently they've now got it to accuracy rates competitive with most other major vision methods?<p>Good job!