Trained neural nets "see" the world in a different way than humans; this disparity has been the topic of a few recent papers. Here we further investigate these differences and along the way come up with two potentially useful methods for visualizing what supervised neural nets are really looking for.
Both methods use optimization to find images that cause high activations, but with different priors in image space. One directly regularizes to explore parts of image space that contain less high-frequency information, and another employs a genetic algorithm with a strong direct bias towards images exhibiting compositionality. This latter method produces fooling images of a completely different character than have previously been found, almost looking like abstract art!
Congratulations to Anh on a great paper!
I think this group could expand, but we'll need some more moderators.
"One might imagine that a stack of enough layers at different levels will provide a full model and/or implementation of the human mind. Even if the layers overlap, one just needs full coverage—and small gaps presumably will lend themselves to obvious filler layers.
For instance, you might say one layer is the Surfaces and Essences analogy engine, and another layer deals with consciousness, another with vision processing, another with body motion control, and so on.
But it’s not that easy (I know, I know…that’s pretty much the mantra of skeptical cognitive science).
I think a slice of abstraction space is probably more like a manifold or some other arbitrary n-dimensional space. And yes, this is an analogy.
These manifolds could be thought of—yay, another analogy!–as 3D blobs, which on this page will be represented as a 2D pixmap (see the lava lamp image)."
Judging Chatbots Without Opening Them
Yet another comment relating to the (mythical) "Turing Test"
In other words, people want simple explanations, one-ring-to-rule-them all theories...at most just a couple of layers. Scientists, I think, often fall into camps which want to explain as much as possible within their domain, and often within a very narrow fad in that domain.
This essay is my attempt to explicate a way of thinking about psychological / cog arch theories in concept space.
or if you prefer a white background:
- GnuBIO2014 - present
- Basis Technology Corp.software engineer, 2012 - 2014
- iRobot Corp.software engineer, 2005 - 2012
- MIT Lincoln Laboratoryintern, 2003 - 2003
- Charles River Analyticsintern, 2004 - 2004
- DEKAintern, 2002 - 2002
My formal education in AI includes undergrad Artificial Intelligence (Northeastern University), Society of Mind / Emotion Machine taught by Marvin Minsky (MIT), Commonsense Reasoning for Interactive Applications taught by Henry Lieberman (MIT), and Cognitive Architectures advised/taught by Whitman Richards (MIT). I attended Ray Solomonoff's IAP lectures at MIT in 2005. I've also had interesting discussions with Aaron Sloman, Rod Brooks, and the late Push Singh (mostly 2005-2006).
I have built and/or worked on several mobile robots (ground and underwater) ranging from autonomous to teleoperated. My day job is programming and interaction design.
Member of AAAI, IEEE, IEEE Computational Intelligence Society, IEEE Robotics & Automation Society, ACM, SIGAI.
I am also an actor and screenwriter.
- Northeastern UniversityBS
- Massachusetts Institute of TechnologyGraduate special student (not a degree program)Cognitive Architectures Research (9.364) Commonsense Reasoning for Interactive Applications (MAS.969) Society of Mind / The Emotion Machine (6.868)
RunKeeper Comes Out Hot for the Second Year in a Row, Sees 637% Boost in...
BostInno | Boston Startup, Tech, Education and News BostInno | Boston Startup, Tech, Education and News. Home; Tech; City News; Edu; Sports;
Cygwin bash script: open windows explorer anywhere
Here is my take on this job. It is a small script to open Windows Explorer from any path I give it, or the current directory if I don't spec
The Time I Planted a Pipe Bomb Under Someone's Truck
I almost crash into a truck, at first. I can turn on a dime and accelerate like a tiny nitrous-oxide-fueled bat out of hell—even though I'm
Bomb Patrol Afghanistan: Exclusive Clip From the New Best Thing on TV
In the first 60 seconds of Bomb Patrol: Afghanistan, you will be hooked. You'll then spend the rest of episode one with an elevated heart ra
Computerised car drives itself around Berlin - The Local
A high-tech concept car has been driving itself around Berlin and nearby autobahns since the summer, gaining accident-free experience in tra
iOnRoad – Augmented Reality Driving & Real-Time Collision Detection
iOnRoad - Augmented Reality Driving & Real-Time Collision Detection
The Power of Creativity: How Game Design Changes the Way We Think
Game designers, who must capture and retain players' attention and interest quickly, need to understand human psychology and culture
Apple's mind-bogglingly greedy and evil license agreement | ZDNet
Over the years, I have read hundreds of license agreements, looking for little gotchas and clear descriptions of rights. But I have never, e
The Information Diet: A Case for Conscious Consumption
Why "information overload" is the wrong lens on the issue, or what sugar and fat have to do with Hollywood. "You are a mashup of what you
CMU Researchers Create a Multitouch Surface Everywhere
In a joint effort between Microsoft and Carnegie Mellon Human Computer Interaction Institute the creation of a new interface has been born.
Artificial Intelligence Is Coming to the iPhone, And It's Going to Chang...
Your phone is about to become a whole lot smarter, thanks to Apple and its new artificial intelligence Assistant.