Skip to main content

Home/ Advanced Concepts Team/ Group items tagged ai

Rss Feed Group items tagged

jcunha

DeepMind's AI team explores navigation powers with 3-D maze - 0 views

  •  
    After the Go, real-time RPG as Hendrik alluded?
Guido de Croon

New theory allows drones to see distances with one eye - 2 views

  •  
    Inspired by the work that was done at the ACT, I continued working on optical flow landing at TU Delft. Today Bio & Bio published my article on a new theory that allows drones to see distances with a single camera. It shows that drones approaching an object with an insect-inspired vision strategy become unstable at a specific distance from the object. Turning this weakness into a strength, drones can actually use the timely detection of that instability to estimate distance. The new theory will enable further miniaturization of autonomous drones and provides a new hypothesis on flying insect behavior.

    Video: https://www.youtube.com/watch?v=Fm7SMJp8EA4&feature=youtu.be
    Article: http://iopscience.iop.org/article/10.1088/1748-3190/11/1/016004
jcunha

Computer model matches humans at predicting how objects move - 0 views

  •  
    We humans take for granted our remarkable ability to predict things that happen around us. Here, a deep learning model trained from real-world videos and a 3D graphics engine was able to infer physical properties of objects against humans.
Ingmar Getzner

The First Person to Hack the iPhone Built a Self-Driving Car. In His Garage - 4 views

  •  
    Read this this morning in the train, what a story! Awesome guy, I wish him all the luck kicking against the established companies... Seems he has a bet with Elon Musk to outperform the current autonomous driving algoritms using his AI techniques. He is actually driving a lot with his car via Uber, to gain material to train his NN on :)
Marcus Maertens

Big Hero 6's Programmable Nanobots Are on the Horizon - 2 views

  •  
    This collaborating swarm of drones acts as 3D pixels (voxels) to create giant, flying interactive displays.
  •  
    I have never understood the flying part of these things. Isn't it really impracticle to have all those tiny quadrocopters zooming around. My money is on holography or still a google glass type of device, if only considering the energy requirements for doing anything kinetically.
Ingmar Getzner

Controversial Quantum Machine Bought by NASA and Google Shows Promise - 4 views

  •  
    I am having less and less faith in the Dwave machine, but nonetheless, maybe we should have a look at our future encryption techniques...
  •  
    why less and less ... ?
Paul N

Computers Learn How to Paint Whatever You Tell Them To - 3 views

  •  
    Most self-respecting artists wouldn't agree to paint a portrait of a toilet in the middle of a field. Fortunately, advancements in artificial intelligence have given computers the ability to imagine just about any scenario, no matter how bizarre, and illustrate it. Take a look at this image.
  •  
    Those are some creepy faces among them.. This is also just completely random isn't it?
  •  
    Well, biased to the data it was trained on. Computing a net is pretty deterministic.
    But not everything is perfectly correlated yet.
    Still nice progress.
Nina Nadine Ridder

Roboticists learn to teach robots from babies - 2 views

  •  
    Babies learn about the world by exploring how their bodies move in space, grabbing toys, pushing things off tables and by watching and imitating what adults are doing. But when roboticists want to teach a robot how to do a task, they typically either write code or physically move a robot's arm or body to show it how to perform an action.
Daniel Hennes

Google Just Open Sourced the Artificial Intelligence Engine at the Heart of Its Online ... - 2 views

  •  
    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
Ma Ru

Intelligent Machines - BBC News - 3 views

  •  
    AI week on BBC - interesting opportunity to get a glimpse into how the field is viewed by laymen...
Alexander Wittig

Neuronal Networks: Computers paint like van Gogh - 1 views

  •  
    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
jcunha

Introducing A Brain-inspired Computer [IBM TrueNorth] - 0 views

  •  
    Built in Silicon technology (Samsung's 28 nm process), its power is measured as one million neurons and 256 million synapses. It contains 5.4 million transistor being the largest IBM chip in these terms. All this said, it consumes less than 100 mW!!

    "These systems can efficiently process high-dimensional, noisy sensory data in real time, while consuming orders of magnitude less power than conventional computer architectures." IBM is working with initLabs to integrate the DVS retinal camera with these chips = real time image neuro-like image processing.

    In what seems to be a very successful project hugely funded by DARPA, "Our sights are now set high on the ambitious goal of integrating 4,096 chips in a single rack with 4 billion neurons and 1 trillion synapses while consuming ~4kW of power."
Nina Nadine Ridder

Robots collaborate to deliver meds, supplies, and even drinks - 2 views

  •  
    At the recent Robotics Science and Systems (RSS) conference, a CSAIL team presented a new system of three robots that can work together to deliver items quickly, accurately and, perhaps most importantly, in unpredictable environments. The team says its models could extend to a variety of other applications, including hospitals, disaster situations, and even restaurants and bars.
Juxi Leitner

Robots to the Rescue!: JPL's RoboSimian and Surrogate Robots are here to Help - 3 views

  •  
    Robots to the Rescue!: JPL's RoboSimian and Surrogate Robots are here to Help
  •  
    Also many other interesting videos of the Karman Lectures
Alexander Wittig

Picture This: NVIDIA GPUs Sort Through Tens of Millions of Flickr Photos - 2 views

  •  
    Strange and exotic cityscapes. Desolate wilderness areas. Dogs that look like wookies. Flickr, one of the world's largest photo sharing services, sees it all. And, now, Flickr's image recognition technology can categorize more than 11 billion photos like these. And it does it automatically. It's called "Magic View."

    Magical deep learning! Buzzword attack!
  • ...4 more comments...
  •  
    and here comes my standard question: how can we use this for space? fast detection of natural disasters onboard?
  •  
    Even on ground. You could for example teach it what nuclear reactors or missiles or other weapons you don't want look like on satellite pictures and automatically scan the world for them (basically replacing intelligence analysts).
  •  
    In fact, I think this could make a nice ACT project: counting seals from satellite imagery is an actual (and quite recent) thing:
    http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0092613
    In this publication they did it manually from a GeoEye 1 b/w image, which sounds quite tedious. Maybe one can train one of those image recognition algorithms to do it automatically.
    Or maybe it's a bit easier to count larger things, like elephants (also a thing).
  •  
    In HiPEAC (High Performance, embedded architecture and computation) conference I attended in the beginning of this year there was a big trend of CUDA GPU vs FPGA for hardware accelerated image processing. Most of it orbitting around discussing who was faster and cheaper with people from NVIDIA in one side and people from Xilinx and Intel in the other. I remember of talking with an IBM scientist working on hardware accelerated data processing working together with the Radio telescope institute in Netherlands about the solution where they working on (GPU CUDA).

    I gathered that NVIDIA GPU suits best in applications that somehow do not rely in hardware, having the advantage of being programmed in a 'easy' way accessible to a scientist. FPGA's are highly reliable components with the advantage of being available in radhard versions, but requiring specific knowledge of physical circuit design and tailored 'harsh' programming languages. I don't know what is the level of rad hardness in NVIDIA's GPUs...

    Therefore FPGAs are indeed the standard choice for image processing in space missions (a talk with the microelectronics department guys could expand on this), whereas GPUs are currently used in some ground based (radio astronomy or other types of telescopes).

    I think that on for a specific purpose as the one you mentioned, this FPGA vs GPU should be assessed first before going further.
  •  
    You're forgetting power usage. GPUs need 1000 hamster wheels worth of power while FPGAs can run on a potato. Since space applications are highly power limited, putting any kind of GPU monster in orbit or on a rover is failed idea from the start. Also in FPGAs if a gate burns out from radiation you can just reprogram around it.

    Looking for seals offline in high res images is indeed definitely a GPU task.... for now.
  •  
    The discussion of how to make FPGA hardware acceleration solutions easier to use for the 'layman' is starting btw http://reconfigurablecomputing4themasses.net/.
alekenolte

Research Blog: Inceptionism: Going Deeper into Neural Networks - 0 views

  •  
    Deep neural networks "dreaming" psychedelic images
  • ...3 more comments...
  •  
    Although that's not technically correct. The networks don't actually generate the images, rather the features that get triggered in the network already get amplified through some heuristic. Still fun tho`
  •  
    Now in real time:
    http://www.twitch.tv/317070
  •  
    Yes, true for the later images, but for the first images they start with random noise and a 'natural image' prior, no?
    But I guess calling it "hallucinating" might have been more accurate ;)
  •  
    Funny how representation errors in NNs suddenly become art. God.... neo-post-modernism.
krzysieknowak

Scientists Are Turning Their Backs on Algorithms Inspired By Nature - 4 views

  •  
    "Over the past couple of decades, the research literature has filled up with endless new nature-based metaphors for algorithms. You can find algorithms based on the behaviour of cuckoos, bees, bats, cats, wolves, galaxy formation and black holes. (...) All researchers have been doing is wasting time on developing new approaches that are probably little better than existing ones. And the language of each metaphor then invades the literature, distracting people from using the already sufficiently expressive terminology of mathematics and, above all, working together to find the best way forward."

    The golden era of fireworks-like algorithm is about to end
  •  
    Lies, lies, all lies. They will never go away. Papers need to be published.
jcunha

Training and operation of an integrated neural network based on memristors - 0 views

  •  
    Almost in time for the workshop last week!
    This new Nature paper (e-mail me for full paper) claims training and usage of neural network implemented with metal-oxide memristors, without selector CMOS. They used it to implement a delta-rule algorithm for classification of 3x3 pixel black and white letters. Very impressive work!!!!
  •  
    For those not that much into the topic, see the Nature's News and View section www.nature.com/nature/journal/v521/n7550/full/521037a.html?WT.ec_id=NATURE-20150507 where they feature this article.
Dario Izzo

Study maps extroversion types in the brain's anatomy - 7 views

  •  
    Anna will rule the world!!!!
  •  
    So... start preparing to be required to attach your brain scan along with your job application...
Juxi Leitner

Game-playing software holds lessons for neuroscience : Nature News & Comment - 4 views

  •  
    DeepMind actually got a comp-sci paper into nature...
1 - 20 of 284 Next › Last »
Showing 20 items per page