Skip to main content

Home/ Advanced Concepts Team/ Group items tagged ai

Rss Feed Group items tagged

dharmeshtailor

Opening the Black Box of Deep Neural Networks via Information Theory - 1 views

Marcus Maertens

MIT, Mass Gen Aim Deep Learning at Sleep Research | NVIDIA Blog - 2 views

  •  
    Neural Networks to analyse sleeplessness.
Marcus Maertens

Using AI to count craters on the moon at U of T's Centre for Planetary Sciences - 2 views

  •  
    Works for mercury as well.
Marcus Maertens

What do blockchain, artificial intelligence and quantum computing mean for smart contra... - 1 views

  •  
    How to cram in all buzzword technologies in one title. I will give it all tags we have.
Marcus Maertens

Ubisoft's AI in Far Cry 5 and Watch Dogs could change gaming | WIRED UK - 0 views

  •  
    Commit Assist Tool allows predicting bugs in large code bases typically found in AAA-games.
Marcus Maertens

AI racks up insane high scores after finding bug in ancient video game * The Register - 2 views

  •  
    Evolutionary Strategies are able to explore broader areas of the search space than reinforcement learning techniques. Thus, they are able to encounter strange bugs resulting in large rewards.
  •  
    It will be the new hype in a few years when DL is settled....
dharmeshtailor

Comeback for Genetic Algorithms...Deep Neuroevolution! - 5 views

  •  
    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).
LeopoldS

Alibaba's AI Outguns Humans in Reading Test - Bloomberg - 4 views

  •  
    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/
jcunha

When AI is made by AI, results are impressive - 6 views

  •  
    This has been around for over a year. The current trend in deep learning is "deeper is better". But a consequence of this is that for a given network depth, we can only feasibly evaluate a tiny fraction of the "search space" of NN architectures. The current approach to choosing a network architecture is to iteratively add more layers/units and keeping the architecture which gives an increase in the accuracy on some held-out data set i.e. we have the following information: {NN, accuracy}. Clearly, this process can be automated by using the accuracy as a 'signal' to a learning algorithm. The novelty in this work is they use reinforcement learning with a recurrent neural network controller which is trained by a policy gradient - a gradient-based method. Previously, evolutionary algorithms would typically be used.

    In summary, yes, the results are impressive - BUT this was only possible because they had access to Google's resources. An evolutionary approach would probably end up with the same architecture - it would just take longer. This is part of a broader research area in deep learning called 'meta-learning' which seeks to automate all aspects of neural network training.
  •  
    Btw that techxplore article was cringing to read - if interested read this article instead: https://research.googleblog.com/2017/05/using-machine-learning-to-explore.html
jaihobah

Google's AI Wizard Unveils a New Twist on Neural Networks - 2 views

  •  
    "Hinton's new approach, known as capsule networks, is a twist on neural networks intended to make machines better able to understand the world through images or video. In one of the papers posted last week, Hinton's capsule networks matched the accuracy of the best previous techniques on a standard test of how well software can learn to recognize handwritten digits."

    Links to papers:
    https://arxiv.org/abs/1710.09829
    https://openreview.net/forum?id=HJWLfGWRb&noteId=HJWLfGWRb
  •  
    impressive!
  •  
    seems a very impressive guy :"Hinton formed his intuition that vision systems need such an inbuilt sense of geometry in 1979, when he was trying to figure out how humans use mental imagery. He first laid out a preliminary design for capsule networks in 2011. The fuller picture released last week was long anticipated by researchers in the field. "Everyone has been waiting for it and looking for the next great leap from Geoff," says Kyunghyun Cho, a professor"
jaihobah

New Theory Cracks Open the Black Box of Deep Learning - 0 views

  •  
    A new idea called the "information bottleneck" is helping to explain the puzzling success of today's artificial-intelligence algorithms - and might also explain how human brains learn.
jaihobah

A Brain Built From Atomic Switches Can Learn - 0 views

  •  
    A tiny self-organized mesh full of artificial synapses recalls its experiences and can solve simple problems. Its inventors hope it points the way to devices that match the brain's energy-efficient computing prowess.
Wiktor Piotrowski

Harnessing evolutionary creativity: evolving soft-bodied animats in simulated physical ... - 0 views

  •  
    Papers in the video description
jaihobah

Emergence of Locomotion Behaviours in Rich Environments - 1 views

shared by jaihobah on 11 Jul 17 - No Cached
jcunha liked it
  •  
    Some work by DeepMind on applying reinforcement learning to teach a computer to navigate complex environments. Come for the science - stay for the video: https://goo.gl/8rTx2F
Dario Izzo

How the Space Pope is helping to find real exoplanets by playing Eve: Online | Ars Tech... - 0 views

  •  
    serious gaming came back!
Paul N

Google's AI has learned how to draw by looking at your doodles - 0 views

  •  
    "To create Sketch-RNN, Google Brain researchers David Ha and Douglas Eck collected more than five million user-drawn sketches from the Google tool Quick, Draw! Each time a user drew something on the app, it recorded not only the final image, but also the order and direction of every pen stroke used to make it. The resulting data gives a more complete picture (ho, ho, ho) of how we really draw."

    It's funny because this David Ha used to be a quant banker ha ha
Paul N

A look at deep learning for science - 1 views

  •  
    Scientific use cases show promise, but challenges remain for complex data analytics.
Alexander Wittig

Google AI experiment: fast drawing for everyone - 0 views

  •  
    AutoDraw is a new kind of drawing tool. It pairs machine learning with drawings from talented artists to help everyone create anything visual, fast. There's nothing to download. Nothing to pay for. And it works anywhere: smartphone, tablet, laptop, desktop, etc.

    AutoDraw's suggestion tool uses the same technology used in QuickDraw, to guess what you're trying to draw. Right now, it can guess hundreds of drawings and we look forward to adding more over time. If you are interested in creating drawings for others to use with AutoDraw, contact us here.

    We hope AutoDraw will help make drawing and creating a little more accessible and fun for everyone.
jcunha

New rules for robots backed by European Parliament committee - 1 views

  •  
    The European Parliament's Legal Affairs Committee voted in favour of a resolution calling for new laws addressing robotics and artificial intelligence (AI) to be set out to sit alongside a new voluntary ethical conduct code that would apply to developers and designers.
1 - 20 of 317 Next › Last »
Showing 20 items per page