superfx

I guess I was confused because the main TensorFlow page doesn't have that phrase. Made it sound like it won't be officially supported by Google.

Why does the GitHub page say this is not an official google project, yet it's on the google blog?

I believe the TensorFlow Fold paper came out before PyTorch.

Thanks for the clarification.

Does the upcoming XLA interact with this as well? I.e. compilation would be too costly for dynamic graphs, and so it would only make sense for static graphs?

How is adding dynamic graphs to TensorFlow "after the fact" while adding it to Torch isn't? (Torch is much older than TF).

Does anyone know where the video was filmed? It looks like it might be a neighborhood in the Bay Area.

Here's an official, publicly accessible, link to the article: http://rdcu.be/kXhV

Here's the full article for those interested: http://rdcu.be/iFZr

Can you explain by what is meant by a functional API? Functional as in functional programming?

Any news on what the next lab at YCR will be about?

A concerted, long-term (~10 year) effort to develop the models, theory, and tools necessary to simulate a prokaryotic cell. This is not the kind of thing that’s being funded by the NIH, given their translational shift, and it’s not the kind of thing that can be done by industry because of the time horizon. Nonetheless it’s foundational to building a quantitatively predictive biology. It’s the sort of thing that can enable entirely new applied sciences and industries. And I think it’s possible.

There’s a certain way to doing this which I believe is key to success. It can’t be about doing molecular dynamics from atoms on up, because of computational considerations, and because that’s useless anyway (one wouldn’t learn anything beyond the known physics and maybe better force fields and sampling techniques). It also can’t be like the kludgy attempts that have been made so far (e.g. Markus Covert’s paper in Cell from a few years back—it’s perfectly alright work, but won’t get us closer to solving this particular problem.) Instead, I think the key is to develop a new layer of abstraction for describing biological phenomena, much like say organic chemistry is a genuinely new abstraction built on top of physical chemistry (which is itself built on top of QM), that hides away the irrelevant details (e.g. most of QM) while capturing what’s salient about the biological phenomena of interest. It would require advances in the aforementioned areas of models, theory, and tools, that are made in concert with the aim of simulating a cell.

If $105 million (and counting) in dev costs and hollywood actors is not AAA what is?

I actually tried ordering but their website is not really working right now. Both credit cards and paypal seem not to work.

It's just like any other language. At first you're constantly looking at the docs to find the right function (perhaps library call is the better analogy here), but eventually you'll know what's important and where to look if you're trying to solve something new.

Does anyone know how this compares to Twitter's recently released torch autograd?

https://blog.twitter.com/2015/autograd-for-torch

EDIT: To be a bit more specific, autograd just requires that one writes the forward model, and it can involve any program logic like conditionals, etc, so long as the end result is differentiable. It then takes care of all the gradient calculations.

Link to the Science paper abstract (full paper is paywalled): http://www.sciencemag.org/content/350/6260/530.abstract

Couple questions regarding YCR:

1. Will the model be group-based where there's a head PI setting the research agenda for a group and researchers working under the PI? If so, what level of independence do you expect individual researchers within a group to have? I.e. is it more like the academic model with postdocs enjoying a fair degree of independence, or more akin to national labs where the research agenda is reasonably fixed and everyone is expected to contribute to the same research program? Or, are you coming up with an entirely different model?

2. In general how hands-on will YCR be in terms of intra-group management?

3. Do you expect to subscribe to all the major publishing houses so that YCR researchers have access to journals like they would at a top-tier university?

Will this be structured in a way where there are groups headed by "PIs" that set the research agenda, and researchers working under them, or will you be pursuing a different model?

I pay for workflowy.

I wish they didn't bundle a python installation with it. I'd rather just try the frontend without potentially messing up my python installation.

This question on CRISPR vs. deep learning on Quora is possibly relevant: http://www.quora.com/Which-of-the-following-will-have-a-more...

No pictures of the restored bridge?

They likely didn't spend millions, but got (hardware for) the robot for free from Boston Dynamics.

Link to original paper: http://rdcu.be/cJ0x

It's a little sad seeing how most of the old ornate buildings have been replaced with much more generic looking buildings. That's one thing that struck me about Berlin--it looks a lot more like a "typical" American city than a "typical" European city.

Why does Stanford need to "gain the respect of the East and the North"?

Ruling has already been amended unfortunately:

http://news.sciencemag.org/plants-animals/2015/04/judge-s-ru...

Genetically these bacteria are apparently not that small--on the order of a million DNA bases. We already know of smaller bacteria with just a few hundred thousand bases (e.g. M. genitalium.) And so from the perspective of "minimal life", these may not be all that special.

Here's a publicly-accessible link to the full paper:

http://rdcu.be/cdlg

The most damning aspect of the CIA report is that none of what you're claiming was true. There was no ticking time bomb scenario, and there was no critical information extracted. In fact, the information the CIA did obtain from prisoners was later shown to be false.