Non-Symbolic Artifical Intelligence 2004
Inman Harvey COGS 5C12 ext 8431 inmanh@cogs.sussex.ac.uk

Lectures

Lectures are Tue 09:00/Thu 12:00/Fri 12:00 in ARUN-401, in weeks 1 to 5 of Summer term.

My lecture notes will be placed up on this website, usually the day they are delivered. First link to web pages, second link to Powerpoint file.

Lec 1 html or ppt
Lec 2 html or ppt
Lec 3 html or ppt
Lec 4 html or ppt
Lec 5 html or ppt
Lec 6 html or ppt
Lec 7 html or ppt
Lec 8 html or ppt
Lec 9 html or ppt
Lec 10 html or ppt
Lec 11 html or ppt
Lec12 html or ppt or handouts (.prn)
Lec 13 html or ppt or handouts (.prn)
Lec 14 html or ppt or handouts (.prn)
Lec 15: now incorporated into Lec 16
Lec 16 html or ppt or handouts (.prn)

Link to last year's 2003 NSAI Lectures

Use this link.

Seminars

Seminars will run from Week 2 to Week 6. You will be in 4 separate groups -- up-to-date versions are listed here:-

Group A Mon 14:00 in PEV1-2A12
Group B Mon 15:00 in PEV1-2A12
Group C Tue 14:00 in PEV1-2A12
Group D Tue 15:00 in PEV1-2A12

Subjects for each seminar are as follows:-

Seminar week 2
Seminar week 3
Seminar week 4
Seminar week 5
Seminar Week 6

Coursework

50% of your assessment comes from a Programming exercise, with a short (max 2000 words) report. This must be completed and handed in to Informatics School office by 4pm on Thurs May 27 (week 6).

(Remember: penalties 10% up to 24 hrs late, 100% penalty after that!)

Assessed Programming Project
Your program should implement a 3 layer Artificial Neural Net (ANN), with 4 Input nodes, 4 Hidden nodes and 2 Output nodes. The Hidden nodes each receive weighted inputs from all of the previous Input layer, plus a bias; the Output nodes likewise from the Hidden layer. Sigmoid transfer functions [ 1/(1+e^(-x)) ] should be used at nodes where appropriate. Using this ANN code, two separate training methods should then be implemented, for training the weights and biases on any set of Input/Output training examples:-

Firstly, a separate training part of the program should be written such that this ANN can have its weights and biases trained by back-propagation (or a variant).

Secondly, you should write an alternative Genetic Algorithm training method for finding suitable values for all the weights and biases of the same network. Appropriate methods for encoding all the weights and biases on the genotype should be used, and a suitable fitness function designed.

You should then use independently each training method, backprop and GA, on a version of the 4-bit parity problem. Here the 4 inputs can be any combination of 0s and 1s, and the desired target output of the first Output node is (as close as possible to) 0.0 when there is an even number of input 1s (i.e. 0, 2 or 4 1s) and 1.0 otherwise; the desired target for the second Output node is the opposite (1.0 for even, 0.0 for odd).

Each training method, backprop and GA, should be optimised as far as possible, and then a comparison drawn between performance with the two methods. Is this problem more appropriate for one method than the other?

Exam

50% of your assessment comes from an unseen exam, scheduled for Monday 21st June 2004, at 09:30. This is one and a half hours, and you should answer 2 out of the given 3 questions.
For reference, copies of past exam papers can be found [here]. Syllabus for 2003 was a bit different, for 2002, 2001 was similar to this year.
Warning: you will have to type in "Non-Symbolic Artificial Intelligence" in full as the name of the course, the database doesn't seem to like abbreviations.