1 / 8

Project 3 Neural Networks CS 539

Project 3 Neural Networks CS 539. Skyler Whorton September 28, 2012. Objectives. Faces dataset Direction classification Which specific directions are easier/harder to identify? Name classification Who gets mistaken for whom? Spambase dataset

wood
Télécharger la présentation

Project 3 Neural Networks CS 539

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Project 3Neural NetworksCS 539 SkylerWhortonSeptember 28, 2012

  2. Objectives • Faces dataset • Direction classification • Which specific directions are easier/harder to identify? • Name classification • Who gets mistaken for whom? • Spambase dataset • Are there relationships in the data on than entropy measures can’t detect?

  3. Faces - Preprocessing • 30x32 pixel images = 960 predicting attributes • Convert ‘binary’ PGMB to ASCII-readable PGMA • Transform irregular text file to comma-separated pixel data • Two variations: each instance labeled with Name or Direction

  4. Neural Net Topology • Network topology • MultilayerPerceptron: suggested defaults of: • ‘a’ = (instaces + classes) / 2 • ‘o’ = classes. Put into two hidden layers: ‘o, o’ • Nprtool: searched for least-error single hidden layer topology:

  5. Faces - Direction • 960 predicting attributes • 4 classes • 624 instances • Nprtool ‘Optimal’ network topology: 15 nodes

  6. Faces - Name • 960 predicting attributes • 20 classes • 624 instances • Nprtool: 10 nodes

  7. Faces – Name, cropped • 538 predicting attributes (Trimmed 4 pixels from each edge) • 20 classes • 624 instances • Nprtool: 10 nodes

  8. Confusion • Direction • ‘straight’ sometimes seems like ‘up’ • ‘left’ and ‘right’ seldom confused • Name • The models are too good! • Only misclassify ~2-4 instances a b c d <-- classified as 136 18 0 2 | a = straight 8 144 4 0 | b = up 3 3 151 0 | c = left 2 2 1 150 | d = right

More Related