haeffin

They said at ICCV that they will release the code after the CVPR paper submission deadline, so "soon".

How is that different from allowing non-axis-aligned splits?

Why do you say (3) does not exist? C++ has it (and it's one of the main reason why, at the moment, I wouldn't use Rust to do the things I do).

Using your search, at least for me, none of the snippets on the first page use normalization in the sense that you are in this context. So including that term just got you a lot of noise. And the only reference to whitening on that search page is not using it in an input pipeline, it is using ZCA to detect images that are modified to be adversarial.

If you want some better data than unreliable searches, go download pretrained models for popular architectures and popular frameworks and look at the input pipelines for them. You'll find that whitening is absolutely not common for image classification/detection today (yes, there are still some cases where it is used, but typically on smaller datasets where you can't get that invariance from data, which is the way you prefer it to be - if one class actually is more likely to be present in dark images, you don't want to kill that information).

"normalizing" is a bad search term, it can mean a lot of things. And whitening images is pretty much dead. What is done is subtracting mean colors from each pixel, but those are means over the whole database, not per-image, so that keeps brightness shifts intact.

> a common preprocessing step in computer vision (and NNs in general) is to whiten the image by standardizing or transforming pixels to a normal distribution; this would tend to wipe global brightness levels, promoting invariance to illumination

Is there anybody still doing this?

SC2 ticks way faster than SC1 though, which you don't notice much as a human, but you might change your strategy when designing the ai components.

That's a serious point of frustration for me. Having an option to use anaconda, fine. Forcing it on your users, meh. I already have a working system using virtualenv and pip, why force another on me?

> Curiosity was launched from Cape Canaveral on November 26, 2011, at 15:02 UTC aboard the MSL spacecraft and landed on Aeolis Palus in Gale Crater on Mars on August 6, 2012, 05:17 UTC.

You sure it's not 18 months total?

Illustration: https://www.youtube.com/watch?v=d-AYhMoLorI

They hired the author.

They put out a facebook post at about the same time that is something like a press release though.

> I like that we clearly include a patent license in our repo. IMO it would be nice if more companies would choose this route too.

I don't get this. The Apache2 license also has a section on patents ...

You can't get the things you get from the API from videos alone. If a player wants to see the amount of mana an enemy has, he has to click on the enemy, you don't see it usually. Same for items. There is an active component in information gathering in Dota (and those actions take time/clicks away from other things you could do at the same time).

SC2 ticks faster than SC1 - you only have 22ms. You don't need to tie everything to tick rate though, a strategy module could update way slower.

I don't know if I would label SC2 as continuous. I don't think anything happens to the game state at a finer granularity than tick level. So to me it seems that it's also discrete (but with the state changing 44.8 a second at default speed). I agree though that this looks more challenging for ML methods.

I haven't looked at if they limit the rate of commands that the AI can issue, otherwise this will be something that can be a very big advantage to the AI once it learns to micro ...

In efficient stores (e.g. Aldi in Germany and branches in Switzerland and Austria) the cashiers start taking the change into the hand the moment you say you pay cash. They do that up to the next bill and if you pay with a higher bill, picking up extra bills from the registry is really fast. Whereas with a credit card you put the card in, wait until the remote system decides if you should use your PIN or signature, then either put the pin in or wait until the cashier gives you the thing to sign and sign it. And if you're really unlucky the store has a system that looks you up in their customer database if you pay by card which adds another couple seconds.

The new contactless payment system is faster, unless the store does something stupid ... like wanting to do a customer database lookup before doing the actual transaction which one chain here in Austria unfortunately does.

Difficulty aside, gendered nouns make gender-neutral communication awkward. If you want to refer to a group in a gender-neutral way you have to refer to both the feminine plural and the masculine one, or invent some silly new forms.

Essential to create website traffic, maybe, but not for the engineers. These things have next to no additional value over official documentation, and are of lower quality than the official docs - like the "numpy cheat sheet" telling people that "np.float32" is a standard double-precision floating point dtype ... really?

I'm quite annoyed by announcements like this. To be complete, you have to provide a complete solution. It's really annoying with GCC that you have to check in multiple places to see which features a given release supports.

Mean absolute error is differentiable almost everywhere. Having objectives that are not differentiable, but are differentiable almost everywhere is very common - in a deep net, if you have rectified linear activations (very common) or L1 regularisation (not unheard of), you have an objective that is not differentiable everywhere ... but the methods still work.

Comparisons on AlexNet are not very useful now. I can get AlexNet-like quality a lot cheaper (at test time) now, and for the same computational cost I can get a lot better in quality of results ... or even better if I accept more cost. I can't think of a good reason to evaluate AlexNet nowadays, I'm more annoyed at the other guys(tm) that (exclusively) do, since that means to get meaningful datapoints I need to rerun the experiments myself.