BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A talk in two parts: (1) AI Neuroscience: How much do deep neural 
 networks understand about the images they classify? (2) Robots that can ad
 apt like animals.  - Prof. Jeff Clune (U Wyoming)
DTSTART:20161107T110000Z
DTEND:20161107T120000Z
UID:TALK69040@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:A talk in two parts: (1) AI Neuroscience: How much do deep neu
 ral networks understand about the images they classify? (2) Robots that ca
 n adapt like animals. \n\nThe first part of the talk describes our sustain
 ed effort to study how much deep neural networks know about the images the
 y classify. Our team initially showed that deep neural networks are “eas
 ily fooled\,” meaning they will declare with near certainty that complet
 ely unrecognizable images are everyday objects\, such as guitars and starf
 ish. These results suggested that deep neural networks (DNNs) do not truly
  understand the objects they classify\, but instead latch onto a few discr
 iminative features per class. However\, our subsequent results reveal that
  DNNs actually have a surprisingly deep understanding of objects. These ne
 w techniques can also be applied to hidden units in the network\, enabling
  us to study the features that each neuron has learned within a network. O
 ur techniques also generate high-resolution\, realistic images\, and can t
 hus be thought of as generative models. I will present a new\, improved\, 
 unpublished\, generative model that we believe may represent the state of 
 the art in terms of generating a diverse collection of high-quality\, high
 -resolution images. The second part of the talk describes our Nature paper
  on learning algorithms that enable robots\, after being damaged\, to adap
 t in 1-2 minutes and soldier on with their mission. \n\nAI Neuroscience:\n
 * Nguyen A\, Yosinski J\, Clune J (2015) "Deep neural networks are easily 
 fooled: High confidence predictions for unrecognizable images.":http://www
 .evolvingai.org/fooling CVPR ("video summary":https://www.youtube.com/watc
 h?v=M2IebCN9Ht4) \n* Nguyen A\, Dosovitskiy A\, Yosinski J\, Brox T\, Clun
 e J (2016) "Synthesizing the preferred inputs for neurons in neural networ
 ks via deep generator networks":http://www.evolvingai.org/synthesizing. NI
 PS\n* Nguyen A\, Yosinski J\, Clune J (2016) "Multifaceted Feature Visuali
 zation: Uncovering the different types of features learned by each neuron 
 in deep neural networks.":http://www.evolvingai.org/mfv ICML Visualization
  for deep learning workshop.\n* Li Y\, Yosinski J\, Clune J\, Lipson H\, H
 opcroft J (2016) "Convergent Learning: Do different neural networks learn 
 the same representations?":http://yosinski.com/convergent ICLR ("video of 
 talk":http://videolectures.net/iclr2016_yosinski_convergent_learning/)\n* 
 Yosinski J\, Clune J\, Nguyen A\, Fuchs T\, Lipson H (2015) "Understanding
  neural networks through deep visualization.":http://yosinski.com/deepvis 
 ICML Deep Learning workshop ("video summary":https://youtu.be/AgkfIQ4IGaM)
 \n* Yosinski J\, Clune J\, Bengio Y\, Lipson H (2014) "How transferable ar
 e features in deep neural networks?":http://yosinski.com/transfer ("video 
 of talk":https://www.youtube.com/watch?v=IuyJRPxtJHU)\n\n"Robots that can 
 adapt like animals":http://www.evolvingai.org/files/2015_Cully_Nature.pdf 
 (2015) Nature ("video summary":https://youtu.be/T-c17RKh3uE?list=PLc7kzd2N
 KtSfLbnwxNgPJJRY2tAY_Fkk3)\n\nMore at http://www.evolvingai.org\n
LOCATION:Dyson Meeting Room on the Ground Floor
END:VEVENT
END:VCALENDAR