Our former PhD student Shane Legg is co-founder of deepmind (with Demis Hassabis and Mustafa Suleyman), just acquired by Google for ~$500m [1]. In my research group at the Swiss AI Lab IDSIA, Shane was a student of my senior researcher Marcus Hutter, a pioneer of the theory of Universal Artificial Intelligence [2], who is now professor in Canberra. Deepmind also employs my former PhD students Dr. Daan Wierstra (#4), Dr. Tom Schaul, Dr. Alex Graves. They were funded by SNF grants (Alex later also by my postdoc grants at TU Munich), and have been working on deep recurrent artificial neural networks (RNN) [3]. Alex focused on supervised learning with LSTM RNN [3,6,7], Daan and Tom on evolving RNN [4] and Reinforcement Learning RNN [5]. Search for their papers in the reference links [3-7]; compare published work of deepmind [8]; compare [9].
[1] https://www.theinformation.com/Google-beat-Facebook-For-DeepMind-Creates-Ethics-Board
[2] http://www.idsia.ch/~juergen/unilearn.html
[3] http://www.idsia.ch/~juergen/rnn.html
[4] http://www.idsia.ch/~juergen/evolution.html
[5] http://www.idsia.ch/~juergen/rl.html
[6] http://www.idsia.ch/~juergen/deeplearning.html
[7] http://www.idsia.ch/~juergen/handwriting.html
[8] http://arxiv.org/abs/1312.5602
[9] http://www.idsia.ch/~juergen/compressednetworksearch.html
Deeplearn it!
www.deeplearning.it
#machinelearning
#artificialintelligence
#computervision
#deeplearning
[1] https://www.theinformation.com/Google-beat-Facebook-For-DeepMind-Creates-Ethics-Board
[2] http://www.idsia.ch/~juergen/unilearn.html
[3] http://www.idsia.ch/~juergen/rnn.html
[4] http://www.idsia.ch/~juergen/evolution.html
[5] http://www.idsia.ch/~juergen/rl.html
[6] http://www.idsia.ch/~juergen/deeplearning.html
[7] http://www.idsia.ch/~juergen/handwriting.html
[8] http://arxiv.org/abs/1312.5602
[9] http://www.idsia.ch/~juergen/compressednetworksearch.html
Deeplearn it!
www.deeplearning.it
#machinelearning
#artificialintelligence
#computervision
#deeplearning
View 9 previous comments
Just to add my comment on Financial Times today article "DeepMind buy heralds rise of the machines":
Deep Learning is about to revolutionize almost all tasks performed by humans, from learning to optimisation, code generation, even what we call "creativity". These machines are completely different beasts from traditional AI. They can extract meaning and semantics from tons of unstructured data.
However, we are too much obsessed by the terminator hollywood myth. The superiority of these algorithms will not be materialized in killing machines but in a more subtle, ubiquous and pervasive ways, like recommendation systems, credit scoring, user profiling and predictive analytics.Jan 28, 2014
Kinda weird they say it's a "completely different beasts". Neural nets are here from the 1970s.Jan 28, 2014
I know. Even before that. The perceptron was born in the 60s. However, Deep Networks are much more than a perceptron.Jan 28, 2014
+Armando Vieira artificial neural nets from the 80s are very similar to deep nets, just more layers. :)
Even ConvNets which are winning in images are here from the 90s. nothing new...Jan 28, 2014
For those interested, some years ago I asked Shane Legg and Jürgen Schmidhuber about risks associated with artificial intelligence:
http://lesswrong.com/r/discussion/lw/691/shane_legg_on_risks_from_ai/
http://lesswrong.com/r/discussion/lw/682/j%C3%BCrgen_schmidhuber_on_risks_from_ai
The interview with Legg has now been linked to in the NYTimes (http://bits.blogs.nytimes.com/2014/01/28/google-buys-a-i-company-for-search-not-robots/) and recode (http://recode.net/2014/01/27/more-on-deepmind-ai-startup-to-work-directly-with-googles-search-team/).
For more see here: http://wiki.lesswrong.com/wiki/Interview_series_on_risks_from_AI (including one Q&A with +Abram Demski ).Jan 28, 2014
Are these two image search patents by DeepMind engineers together with it's #AI capabilities to enhance GoogleNow the real reason for the acquisition? http://robotenomics.com/2014/01/29/googles-robot-and-artificial-intelligence-acquisitions-are-anything-but-scary/Jan 29, 2014