MY LOG CONTINUED
                                                                                            BACK TO HOME PAGE
                                                                                          
                                                                                            
BACK TO PROJECT PAGE

August 3rd
Today I worked on my presentation.  I have collected some articles on how facial recognition is used for security. I think the students will be interested in learning about facial recognition and neural networks in general because of the vast amount of uses. I want to show them how the self-organizing map recognizes the face and then stores it in the database if that is what it is going to be used for. 

August 4th
Today is Sunday. This is my last week here. I am trying to finish all of my work for the project while trying to pack all of my junk (that means all of the new stuff I have accumulated while I have been here).

August 5th
I worked on the face recognition today. Then I turned my attention to a project Professor Valova wanted me to toy around with. Professor Valova wants me to try the self-organizing map with the proteins but this time instead of having a 2-Dimensional grid, using a 1-Dimensional grid. The rationale behind this is that since proteins are essentially 3-Dimensional but are made of long chains of amino acids it only makes sense that their shape should be mimicked one-dimensionally. This will allow the map to spiral, we hope. The encoding will remain as is; the training will be increased to 5000 epochs because since the map is tighter it can organize more efficiently. The map is 250 nodes.

August 6th
Today I put the finishing touches on my facial recognition presentation. I also tried to see if increasing the training epochs allows the map to spiral. If the map can assume a spiral shape then it can be used to predict the shape of proteins. The problem I encountered with this program is that the map never takes on a complete spiral shape; it looks too linear. It does however; capture the training data more accurately than the 2-dimensional map.

August 7th
Today I attended the Freshmen Summer Institute. The first speaker gave presentation on image processing that was quite interesting. The presentation gave students a little bit of background knowledge on images and characteristics of images. For instance, the number of pixels in an image, color, image isolation. I spoke about facial recognition. I showed the students how the self-organizing map is used in conjunction with a convolution neural network. This allows a face to be represented in eigenspace, and then the eigenfaces, which are eigenvectors of the facial space, are presented to the network. Each face has its own specific eigenvalue, this allows the face to be recognized by the network. Also only certain parts of the face are needed to compute the eigenvalue. As people recognize faces, partly by memorization but more so by focusing on certain characteristics of someoneís face like the shape of their face, size of their nose, etc, the network will compute the eigenvalues based on those characteristics.

August 8th
Well I am almost finished with the project. Today I went over some last minute things with Professor Valova about how I should try to encode the amino acids in the protein chain so the shape conforms to a spiral. Once again I was about 95% accurate. The spiral is still too linear yet it maintains its ability to fully capture every data point. Professor Valova and I agreed that I should not give up on this goal after the project finishes. I would like to try to see what I can accomplish with 1-Dimensional Self Organizing Maps.

August 9th
Today is my last day. Today is a fun-filled day of packing and making sure I have completed all of the tasks initially set at the beginning of the project. This entire project has helped me gain a perspective of neural networks. I am amazed at all of the applications it can be used for. Neural networking is a wonderful tool for pattern recognition and I believe as technology advances neural networks will be building laying the groundwork for artificial intelligence and all other aspects of computer science.


                                                         
BACK TO LOG ENTRIES
1