Hi can you please remove me from your mailing list for the ai class?
Corey James Koberna
Work: [address removed]
SCAD: [address removed]
On Dec 5, 2011, at 11:58 PM, Jason Felice <[address removed]> wrote:
> So I've often thought of different ways to approach my kanban board problem. One way I've frequently thought is to train a classifier to identify card corners. I know that some people do things like this to recognize features in an image.
> One thing that has always bothered me is that the data set is _really_ skewed. I might have 100 examples of good corners, and one example of a bad corner for each pixel in each image aside from those 100. The other thing that worried me about this approach is what Ng touched on in the anomaly video. There's some super-huge number of things considered anomalous, and you won't have examples of all of them, let alone sufficient examples to train.
> My problem is the opposite (only because I want negative classification for the super-huge set).
> Q1. Am I missing something?
> One approach could be to train an RBM or something similar only on the positive images, then stimulate the input to output, then stimulate output to input, and reject the image if the error between the original and recreated is large.
> Q2. Would this work?
> Please Note: If you hit "REPLY", your message will be sent to everyone on this mailing list ([address removed])
> This message was sent by Jason Felice ([address removed]) from Cleveland AI + ML support group.
> To learn more about Jason Felice, visit his/her member profile: http://www.meetup...
> To unsubscribe or to update your mailing list settings, click here: http://www.meetup...
> Meetup, PO Box 4668 #37895 New York, New York[masked] | [address removed]