Group items tagged

love this one ... it seems to take physicist to explain to the AI crowd what they are actually doing ... Deep learning is a broad set of techniques that uses multiple layers of representation to automatically learn relevant features directly from structured data. Recently, such techniques have yielded record-breaking results on a diverse set of difficult machine learning tasks in computer vision, speech recognition, and natural language processing. Despite the enormous success of deep learning, relatively little is understood theoretically about why these techniques are so successful at feature learning and compression. Here, we show that deep learning is intimately related to one of the most important and successful techniques in theoretical physics, the renormalization group (RG). RG is an iterative coarse-graining scheme that allows for the extraction of relevant features (i.e. operators) as a physical system is examined at different length scales. We construct an exact mapping from the variational renormalization group, first introduced by Kadanoff, and deep learning architectures based on Restricted Boltzmann Machines (RBMs). We illustrate these ideas using the nearest-neighbor Ising Model in one and two-dimensions. Our results suggests that deep learning algorithms may be employing a generalized RG-like scheme to learn relevant features from data.

One of the authors of the STL (C++'s Standard Template Library) explains generic programming and slams Java.

"I'd rather not even *think* about developing a web application/webservice/web-whatever in standard C++... is it actually possible?? Perhaps with some weird Microsoft solutions... I bet your bank online services are written in Java. Certainly not in PHP+MySQL :)" Doing in C++ would be awesomely crazy, I agree :) But as I see it there are lots of huge websites that operate on PHP, see for instance Facebook. For the banks and the enterprise market, as a general rule I tend to take with a grain of salt whatever spin comes out from them; in the end behind every corporate IT decision there is a little smurf just trying to survive and have the back covered :) As they used to say in the old times, "No one ever got fired for buying IBM". "Industry has chosen Java not because of industrial-corporate marketing bullshit, but because of economics... it enables you develop robustly, reliably, error-prone, modular, well integrated etc... software. And the costs? Well, using java technologies you can set-up enterprise-quality web application servers, get a fully featured development environment (which is better than ANY C/C++/whatever development environment I've EVER seen) at the cost of exactly 0 (zero!) USD/GBP/EUR... Since many years now, the central issue in software development is not implementing algorithms, it's building applications. And that's where Java outperforms many other technologies." Apart from the IDE considerations (on which I cannot comment, since I'm not a IDE user myself), I do not see how Java beats the competition in this regard (again, Python and the huge software ecosystem surrounding it). My impression is that Java's success is mostly due to Sun pushing it like there is no tomorrow and bundling it with their hardware business.

OK, I think there is a bit of everything, wrong and right, but you have to acknowledge that Python is not always the simplest. For info, Facebook uses Java (if you upload picture for instance), and PHP is very limited. So definitely, in company, engineers like you and me select the language, it is not a marketing or political thing. And in the case of fb, they come up with the conclusion that PHP, and Java don't do everything but complement each other. As you say Python as many things around, but it might be too much for simple applications. Otherwise, I would seriously be interested by a study of how to implement a Python-like system on-board spacecrafts and what are the advantages over mixing C, Ada and Java.

NASA Releases Tool Enabling Citizen Scientists to Examine Asteroid Vesta Vesta Trek is a free, web-based application that provides detailed visualizations of Vesta, one of the largest asteroids in our solar system. NASA's Dawn spacecraft studied Vesta from July 2011 to September 2012. Data gathered from multiple instruments aboard Dawn have been compiled into Vesta Trek's user-friendly set of tools, enabling citizen scientists and students to study the asteroid's features. The application includes: -- Interactive maps with the ability to overlay a growing range of data sets including topography, mineralogy, abundance of elements and geology, as well as analysis tools for measuring the diameters, heights and depths of surface features and more. -- 3-D printer-exportable topography so users can print physical models of Vesta's surface. -- Standard keyboard gaming controls to manoever a first-person visualization of "flying" across the surface of the asteroid. "There's nothing like seeing something with your own eyes, but these types of detailed data-visualizations are the next best thing," said Kristen Erickson, Director, Science Engagement and Partnerships at NASA Headquarters in Washington DC.

Researchers have demonstrated a spinning liquid mirror design that could be used for the main mirror of a huge infrared telescope based on the moon.

Luca Turin presents a theory according to which the sense of smell is aroused by molecular vibrations instead of the shape-based compatibility with receivers on the proteins.

Almost in time for the workshop last week! This new Nature paper (e-mail me for full paper) claims training and usage of neural network implemented with metal-oxide memristors, without selector CMOS. They used it to implement a delta-rule algorithm for classification of 3x3 pixel black and white letters. Very impressive work!!!!

For those not that much into the topic, see the Nature's News and View section www.nature.com/nature/journal/v521/n7550/full/521037a.html?WT.ec_id=NATURE-20150507 where they feature this article.

Finally the thing will have a little more artificial intelligence than just "bump into wall, turn a bit, repeat". We've known for a while now that iRobot has been developing robots with wireless integration along with intelligent navigation capability based on VSLAM (Vision Simultaneous Localization and Mapping). We've know this for enough of a while that it's been a little bit frustrating to see iRobot's most recent Roomba upgrades come out without those neat features.

What I don't get is how the measure the extent of matching: how "close", or realistic is the modelisation they achieve with different methods? And moreover, if weights cannot adapt and there are no direct connections between neurons and layers of neurons, isnt that a very arbitrary matching?

impressive, particularly if it works like it does in the video the whole time. paper here http://mi.eng.cam.ac.uk/~qp202/

Well, impressive indeed... have to try it out...