Group items matching

in title, tags, annotations or url

The latest from Schmidt / Lipson / Eureqa. A significant improvement over their previous work is that now "The algorithm selects between multiple candidate models by designing experiments to make their predictions disagree."

nice paper ... 

pretty impressive stuff!

LSTMs: that was also the first thing in the paper that caught my attention! :) I hadn't seen them in the wild in years... My oversight most likely. The paper seems to be getting ~100 citations a year. Someone's using them.

There are a few papers on them. Though you have to be lucky to get them to work. The backprop is horrendous.

maybe we can use some neural networks to do animal "speech" recognition? :P

nice overview of the 3 main space related projects attempting to bring internet access to the world, namely SpaceX+Google micro-satellites network, OneWeb+Virgin Galactic OneWeb and Project Loon by Google. not a lot of technical details though...

Big data analytic systems are reputed to be capable of finding a needle in a universe of haystacks without having to know what a needle looks like. The very best ways to sort large databases of unstructured text is to use a technique called Latent Dirichlet allocation (LDA). Unfortunately, LDA is also inaccurate enough at some tasks that the results of any topic model created with it are essentially meaningless, according to Luis Amaral, a physicist whose specialty is the mathematical analysis of complex systems and networks in the real world and one of the senior researchers on the multidisciplinary team from Northwestern University that wrote the paper. Even for an easy case, big data analysis is proving to be far more complicated than many of the companies selling analysis software want people to believe.

Most of those companies are using outdated algorithms like this LDA and just apply them like retards on those huge datasets. Of course they're going to come out with bad solutions. No amount of data can make up for bad algorithms.

TensorFlow is an open source software library for numerical computation using data flow graphs. Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) communicated between them. The flexible architecture allows you to deploy computation to one or more CPUs or GPUs in a desktop, server, or mobile device with a single API. TensorFlow was originally developed by researchers and engineers working on the Google Brain Team within Google's Machine Intelligence research organization for the purposes of conducting machine learning and deep neural networks research, but the system is general enough to be applicable in a wide variety of other domains as well.

And the interface even looks a bit less retarded than theano

This is the biggest breakthrough in game AI (and one of the biggest in AI in general) since Deep Blue beat Kasparov in chess: For the first time, a human professional player was defeated in the game of Go. The approach was a combination of tree search and deep neural networks. Very proud of a former colleague on the team at Google Deepmind!

Funny enough, facebook also had a very similar paper around the same time.

In this paper published on Nature, AI researchers used deep Q-network with very good adaptability and obtained performances comparable to those of a human games tester.

Bastards! And that was to be my next idea. Still no recurrency as I see it so far, so this is just some fancy way to do a markov model. Not sure if this is that particular paper or an earlier version but here it is for those interested: http://www.cs.toronto.edu/~vmnih/docs/dqn.pdf

When Australian mobile provider Telstra offered its mobile customers unlimited data for two separate days this year as compensation for network outages, some customers took it as a challenge to download as much as they possibly could in one day. On Sunday, 27-year-old Sydney resident John Szaszvari outdid himself and everyone else by ploughing through almost a whole terabyte of data.

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

List anonymous wikipedia edits from interesting organizations

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

Genetic algorithms are a competitive alternative for training deep neural networks for reinforcement learning. For paper see: https://arxiv.org/pdf/1712.06567.pdf

Interesting pointers in this one! I would like to explore neuroevolution as well, although it seems extremely resource-demanding?

Not necessarily, I think it can be made to be much faster hybridizing it with backprop and Taylor maps. Its one ideas in the closet we still have not explored (Differential Intelligence: accelerating neuroevolution).

The race for the AI chips intensifies.

Deconvnets for astro images .... resolving galxies :)

Faced with a navigational challenge, neural networks spontaneously evolved units resembling the grid cells that help living animals find their way.

credible?

Blockchains, now with more buzzwords!

Time travel! Finally. Can we also use it for antigravity?