Announcements

Week 10

• The research paper summaries (assignment 4) have been graded.
• On Wednesday, Students will give a 10-minute presentation on their (in-progess) segmentater for the CROHME competition handwritten math data. Students should prepare a 7 minute presentation, leaving 3 minutes for questions.
• Project 2 (math symbol segmentation) is due Friday at noon.
• The final exam will be held in our regular classroom (GOL-3550) on Friday Nov. 18th from 10:15am-12:15pm.

Week 9

• Assignment 4 is due this Wednesday at the start of class. This will be the last assignment for the quarter, and your assignment grade will be determined by the highest 3 of your four assignment grades (provided that all four assignments have been submitted).
• Extension: The presentations for Project 2 will be given in-class next Wednesday, November 9th, but the final code and write-up for Project 2 will be due Friday, Nov. 11th at noon.

Week 8

• Project 1's deadline has been extended to Wednesday of Week 8.
• The solution key for Assignment 3 has been posted.
• Casey and Lecolinet's character segmentation survey, and a survey on recognition and retrieval of math notation are available through MyCourses.

Week 7

• Readings on Clustering from the Duda, Hart and Stork text have been posted on MyCourses.

Week 6

• Assignment 3's deadline has been extended to Tuesday at noon (12pm).
• The key for assignment 2 has been posted on MyCourses.

Week 5

• There will be a make-up lecture this Friday from 10am-noon in room 3445.
• The PRTools MATLAB toolkit is available here here; the "nuts and bolts" data is available here. Readings from the related textbook have been placed on MyCourses.

Week 4

• The solution key for Assignment 1 has been posted on MyCourses.
• Assignment 2 is due Wednesday, before the start of class.
• Slides from Wednesday's lecture have been posted on MyCourses. The material presented in the slides is covered in Bishop, in Chapters 4.1.7 (perceptrons) and 5.1-5.3 (more general material on neural networks). Slide 30 provides an example of computing weight updates in the backpropogation algorithm, when using the sum of squared errors as the error function to minimize.
• An illustration of using neural nets for recognizing handwritten digits has been provided by Yann LeCun: (LeNet 5). Note that the networks used are feedforward, but a type of network known as a convolutional neural net, which places constraints on the weight space (so that image 'masks' are learned) and on the number of nodes in each layer.

Week 3

• Assignment 2 has been posted on MyCourses.

Week 2

• There will be no classes in Week 3, and there will be a make-up lecture in Week 5 (Friday Oct. 7, 10am-12pm, Room 3445).
• Assignment 1 has been corrected and re-posted on MyCourses; it now includes a description of an algorithm for computing a sample from a probability density function (e.g. our 1D and 2D Gaussians). The deadline is now 5pm on Thursday (Sept. 15).
• For assignment 1, you will need to generate test samples (one for 1D classification, and another for 2D classification. In each case, sample 100 points from both classes' gaussian likelihood function).
• MATLAB documentation may be found here. There are also a number of online tutorials available, from Mathworks; another user-friendly introduction is available from Clarkson University (Dept. Mathematics).

Week 1

• The course schedule and syllabus are now available.
• Course notes are available through MyCourses (link is at top-right)
• The classroom for lecture has changed. Classes will now be held in ICL 3 (GOL-3550)
• Assignment 1 has been posted on MyCourses. It is due next Wednesday before the start of class.