Group items tagged

microsoft just bought github ...

Cool repository with python source code for GNC applications of AI technology.

A neuronal network trained to paint the scene of a given photograph in the style of Kandinsky, van Gogh, or Munch. Their results look quite impressive. Unfortunately the article is in German, but the English paper (with plenty of pictures) is here: http://arxiv.org/pdf/1508.06576v2.pdf Malen wie Kandinsky, wie van Gogh, wie Munch nur auf Basis einer Fotovorlage? Natürlich gibt es begabte Kunstfälscher, die das können. Jetzt aber gelingt es auch Computern, und zwar auf höchst eindrucksvolle Weise. Drei Forscher von der Universität Tübingen haben es geschafft, einem sogenannten künstlichen neuronalen Netzwerk das Malen beizubringen.

Impressive stuff indeed. Paper came out one week ago. Multiple independent implementations have popped out since then: * https://github.com/Lasagne/Recipes/blob/master/examples/styletransfer/Art%20Style%20Transfer.ipynb * https://github.com/jcjohnson/neural-style * https://github.com/kaishengtai/neuralart

A sonification experiment taking data from ATLAS and translating it into music. The outcome was played at Montreux jazz fest, listen to the results in soundcloud https://soundcloud.com/sonification-quantizer. The way it works is "A tiny subset of collision data from the ATLAS Detector (in CERN, Switzerland) is being generated and streamed in real-time into a sonification engine built atop Python, Pure Data, Ableton, and IceCast." Code's in github https://github.com/cherston/Quantizer_public

Anybody tried this out already?

nobody? where is Francesco hiding?

Now this looks quite convenient! :o

Echoing this, in 2009 Google researchers Alon Halevy, Peter Norvig, and Fernando Pereira penned an article under the title The Unreasonable Effectiveness of Data. In it, they describe the surprising insight that given enough data, often the choice of mathematical model stops being as important - that particularly for their task of automated language translation, "simple models and a lot of data trump more elaborate models based on less data." If we make the leap and assume that this insight can be at least partially extended to fields beyond natural language processing, what we can expect is a situation in which domain knowledge is increasingly trumped by "mere" data-mining skills. I would argue that this prediction has already begun to pan-out: in a wide array of academic fields, the ability to effectively process data is superseding other more classical modes of research.

The catalog, a master list organized into 15 categories, is intended for industry, academia, other government agencies, and general public. The catalog covers technology topics ranging from project management systems, design tools, data handling, image processing, solutions for life support functions, aeronautics, structural analysis, and robotic and autonomous systems. NASA said the codes represent NASA's best solutions to an array of complex mission requirements. McMillan reported that "Within a few weeks of publishing the list, NASA says, it will also offer a searchable database of projects, and then, by next year, it will host the actual software code in its own online repository, a kind of GitHub for astronauts."

What is the best (wisest) size for a group of individuals? Couzin and Kao put together a series of mathematical models that included correlation and several cues. In one model, for example, a group of animals had to choose between two options-think of two places to find food. But the cues for each choice were not equally reliable, nor were they equally correlated. The scientists found that in these models, a group was more likely to choose the superior option than an individual. Common experience will make us expect that the bigger the group got, the wiser it would become. But they found something very different. Small groups did better than individuals. But bigger groups did not do better than small groups. In fact, they did worse. A group of 5 to 20 individuals made better decisions than an infinitely large crowd. The problem with big groups is this: a faction of the group will follow correlated cues-in other words, the cues that look the same to many individuals. If a correlated cue is misleading, it may cause the whole faction to cast the wrong vote. Couzin and Kao found that this faction can drown out the diversity of information coming from the uncorrelated cue. And this problem only gets worse as the group gets bigger.

Couzin research was the starting point that co-inspired PaGMO from the very beginning. We invited him (and he came) at a formation flying conference for a plenary here in ESTEC. You can see PaGMO as a collective problem solving simulation. In that respect, we learned already that the size of the group and its internal structure (topology) counts and cannot be too large or too random. One of the project the ACT is running (and currently seeking for new ideas/actors) is briefly described here (http://esa.github.io/pygmo/examples/example2.html) and attempts answering the question :"How is collective decision making influenced by the information flow through the group?" by looking at complex simulations of large 'archipelagos'.

A nice GPU-based framework that is basically an emulator of the brain of the fruit fly. If you need a fruit fly brain - here it comes!

Interesting ideas - though some metaphors are a bit far fetched. Personally, I think it could be interesting if every scientific article would also have a how-to or tutorial section that gives a recipe of how to apply the newly gained knowledge. Of course, that might be tough to do... :-)

The API of the world is already there (a bit), it is the previous knowledge developed by others. Open Source projects such as the wheel or the brick, allow everyday amazing new APPs to be build such as buildings and cars .... There still is merit, though, in learning from software developments techniques in the everyday world projects. This is indeed the motivation for the ACT to do work in open source (SOCIS, GSoC) and push its members to use stuff like wiki, svn, github, jenkins, and alike. This way we are performing and fostering (http://www.oxforddictionaries.com/definition/english/foster) research into working methods in the hope we will be able to export some of its benefit to the larger ESA.

here we go again...

slashdot: "Russian computer scientist Vladimir Romanov has conceded that his previously published solution to the '3 SAT' problem of boolean algebra does not work."

And the winner is... C++ :) Love the comment on Slashdot: "C++ Templates will turn the most pious programmer into a curse-slinging, chain-smoking alcoholic."

Nice one... However note the sample could be biased, because I'd expect some interaction between "using github" and "being a curse-slinging, chain-smoking alcoholic" ;-)

Fair enough :)

See title...

Just read some of the samples of text generated with their neural networks, insane.

It's really lucky that it was OpenAI who made that development and Elon Musk is so worried about AI. This way at least they try to assess the whole spectrum of abilities and applications of this model before releasing the full research to the public.

They released a smaller model, I got it running on Sandy. It's fairly straight forward: https://github.com/openai/gpt-2

any papers or insights on methods available?

Couldn't find a paper for Alibaba's results but Microsoft Research's performance on this dataset was very close. The paper is here: https://www.microsoft.com/en-us/research/wp-content/uploads/2017/05/r-net.pdf Btw the 'reading test' is a publicly available dataset called 'Stanford Question Answering Dataset (SQuAD)'. Their website shows a leaderboard: https://rajpurkar.github.io/SQuAD-explorer/

This has attracted a lot of attention in the DL community lately. Article: https://lilianweng.github.io/lil-log/2017/09/28/anatomize-deep-learning-with-information-theory.html Paper: https://arxiv.org/pdf/1703.00810.pdf

"A list of the many different animals, plants, microbes, natural phenomena and supernatural activities that can be spotted in the wild lands of the metaphor-based computation literature"

One of the best use cases for MS Excel so far.

"The Neuromorphic Computing Platform allows neuroscientists and engineers to perform experiments with configurable neuromorphic computing systems. The platform provides two complementary, large-scale neuromorphic systems built in custom hardware at locations in Heidelberg, Germany (the "BrainScaleS" system, also known as the "physical model" or PM system) and Manchester, United Kingdom (the "SpiNNaker" system, also known as the "many core" or MC system)."