Add to Watchlist

Duck Duck Moose [DEMO #5]

0 views

Citation of segment
Embed Code
Purchasing a DVD Cite video

Automated Media Analysis

Beta
Recognized Entities
Speech transcript
or what was the World War World War in and the W and W W W W Mrs. channel shamelessly photometer group of parts and brains and if you're interested in sort of this cross between neuro science today I highly recommend you check it out so I've been interested in the AI for a while and been not teaching myself for the the deep learning stuff and so I'm just curious how I could sort of have of these utterances are glued together and sort of a contrived example that I came up with was a a duct that use game remembers duct accuses a kid when you walk around the circle you patterns upon the head duct duct duct and eventually had to do this and it's like a it's an anomaly right so what I wanted to do was nuclear cannot force and note really handle image data on which is much larger there's not a hierarchical representation by deploying actually is really great at getting of the underlying features and the underlying hierarchy of the image itself and so what I decided to try to see if I can use the here coming if the if I could use the the deep learning networks T as a feature detectors so that I could use to actually generate the str automatically for the images so I used a set of images from indigent and if you
have a image net uh this is what everybody's crushing right now is that some the algorithms lately and what they do is they basically have labeled data they have a picture of abducting this case and
then they have that someone's labeled it was done by X in this dataset there's only a thousand categories so they had a like the selective and interestingly enough there's not a doubt is not 1 of the categories in this dataset they got all kinds of weird other strange you know exotic birds that they don't have duck and for them and adopt so I've got a lot of when you get these results back ownership of quick
you get the results back on their data effectively
probabilities so even tho it's not done have done what it gives suggestions that a similar to that break is like a lot of the other birds agreements will and and so you can see that's really dark ones so this is some of the categories a come back right so for this 1 of its 1st guess what was missing from the for this 1 it's 1st guess was that 73 per cent chance that that was what that was right then 2nd best guesses 21 % it was an albatross on and group of Member trouble and get up in the likelihood of only 10 per cent chance of developing so as well as I could take like the top 10 results the kind of crappy like American the great where really assume it's a bird apparently have some sort of word so it's only there is 1 per cent but it's still influences general idea that these things are birds right even though there's not adopt representation representation just by kind of subsampling here I want to go further than this action remove the classifier actually get some those underlying features but can ran out of time but and so did you know that the sort of dumb idea of the game is that you playing doctor cues
as you click this you can see this at the top 10 or as that you know some of them like use were actually so popular so there's the as try the format in which the linking has recognized so in this case it was used the categories for this image right that it said these are the top 3 guesses for that image right now and only the thousands you know in in the dataset categories and so some of these have cancer needs have 1 of the most have something and out so what you know what amounts of underrated move forward with it but it's the only dig into those underlying features actually get this much more much more or less porous patients so that was my projects and it's up and get out so
check it out so afraid cause
you're what were you so you want to go forward aside from getting less sparse representations what you think it would be at application for the past events through which you will this seemed like a normal networks deep never did learning networks are not really great they have some temporal components they're ready now with the the internet but that's a long term and a very you know compute heavy but and so I'm thinking maybe we can sort of get the best of both worlds would you that's really great is that sort of encoding so all doing like know who like trying to figure out how to encode my data we're all dinner with a dude boring stuff is really get at least certain types of data like images of actually generating that special presentation for so many like how how I can actually take that data and then conclude this other stuff so where we started that process but you know where it will not go to meeting particularly keyboarding library yet using torture but then just having because I've taking yarns uh down the current state giving a course at NYU is going all videos of online so charge that they give Frank variable few well about parts of brain circuits possibility of years users and so on and the in and
Nuclear space
Forcing (mathematics)
Set (mathematics)
Disk read-and-write head
Mereology
Local Group
Medical imaging
String (computer science)
Hierarchy
Computer network
Representation (politics)
Circle
Pattern language
Quicksort
Game theory
Medical imaging
Category of being
Algorithm
Weight
Resultant
Category of being
Word
Group action
1 (number)
Representation (politics)
Control flow
Quicksort
Game theory
Likelihood function
Resultant
Local Group
Medical imaging
Category of being
Causality
Personal digital assistant
File format
Projective plane
Right angle
Staff (military)
Digital electronics
Process (computing)
Presentation of a group
State of matter
Connectivity (graph theory)
Mereology
Variable (mathematics)
Cartesian coordinate system
Event horizon
Medical imaging
Internetworking
Term (mathematics)
Computer network
Videoconferencing
Representation (politics)
Quicksort
Data type
Library (computing)

Metadata

Formal Metadata

Title Duck Duck Moose [DEMO #5]
Title of Series 2015 Spring NuPIC Hackathon
Number of Parts 19
Author Carey, Frank
License CC Attribution 3.0 Unported:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor.
DOI 10.5446/18052
Publisher Numenta Platform for Intelligent Computing (NuPIC)
Release Date 2015
Language English

Content Metadata

Subject Area Information technology
Abstract Frank Carey of the NYC Bots and Brains Meetup works with ImageNet and Torch to create SDRs representing features as input for NuPIC.

Recommendations

Loading...
Feedback
AV-Portal 3.5.0 (cb7a58240982536f976b3fae0db2d7d34ae7e46b)

Timings

  383 ms - page object