CSC7130

Advanced Artificial Intelligence Fall 2005

The main CSC7130 website

Irwin King

Department of Computer Science and Engineering

The Chinese University of Hong Kong

Shatin, New Territories

Notes

The pdf files are created in Acrobat 4.0. Please obtain the correct version of the Acrobat Reader from Adobe.

Lecture Notes

1. Introduction & Biological Models & Applications (2004/12/8) [ b&w pdf | color pdf ]
2. Correlation Matrix Memory (2004/12/8) [ b&w pdf | color pdf ]
3. Radial Basis Function Networks (2004/12/8) [ b&w pdf | color pdf ]
4. Principal Component Analysis (PCA) (2004/12/8) [ b&w pdf | color pdf ]
   • Data Projection Movie (2004/12/8) [ movie ]

      

Homework Assignments

2005 Fall Semester Homework Assignments

1. Homework assignment 1 Due on Nov. 3, 2005 (2005/10/10)

• Written Assignment: There are 6 problems--1.2.7, 1.2.8, 1.3, 1.6, 1.8, and 1.9.
• TA: Mr. Patrick Lau <tplau@cse.cuhk.edu.hk>

2005 Spring Semester Homework Assignments

1. Homework assignment 1 Due on Feb. 17, 2005 (2005/1/20)

• Written Assignment: There are 6 problems--1.2.4, 1.2.5, 1.3, 1.6, 1.8, and 1.9.
• TA: Mr. Patrick Lau <tplau@cse.cuhk.edu.hk>

2004 Homework Assignments

1. TBA

• TA: TBA

2003 Homework Assignments

1. Homework assignment 1 Due on October 9, 2003 (2003/9/18)

• TA: Mr. Kaizhu Huang <kzhuang@cse.cuhk.edu.hk>

2002 Homework Assignments

1. Homework assignment 1 Due on November 14, 2002 (2001/10/28)

2001 Homework Assignments

1. Homework assignment 1 Due on October 25, 2001 (2001/10/11)
   • Suggested homework solutions (2001/11/30)
   • Homework assignment marks (2001/11/30).

2000 Homework Assignments

1. Homework assignment 1 Due on June 5, 2000 (2000/5/19)
   • FAQ on the Homework Assignment
     1. The transfer function is applied after the summation. So in 2(c) you will need to sum up all the input times the respective weights before applying the sigmoid function. For 2(b), the transfer function for the McCulloch-type neuron is a special one, the output is either a '0' or a '1'.
     2. In question 2(c), the value a can take on a range of values. You may try to play around with it and see how the a can affect the sigmoid function. In this case, you may use any value to calculate the output as long as it is stated clearly in your solution.
     3. For question 5(c), you need not normalize the input patterns since it is somewhat normalized already (divide all values of a1 by -20 and you get a small values for the middle two values). Hence this is similar of having -20 * a1, with a1 being very close to [0 0 0 1], in which it is orthogonal to a2 and a3 so that the inner product of any two vectors will be close to 0. You need not divide other vectors by the same value since it is just a constant multiplication factor.
     4. The "Euclidena sense" means to find the distance between the two vectors (patterns). In this case, use the L2 distance measure. It is defined as the square root of the sqare of their differences, e.g., ((x-y)^2)^0.5.
     5. For question 6(b), since it is an autoassociative memory, the input and the output patterns are the same so you will not need to define the memorized patterns by yourself.

        When investigating how close to perfect the memory autoassociates, you should take a look at what you are receiving as the output and then compare it to the input. The difference is the error. For example, the input a1 will result in a1' and the error is then (a1 - a1'). If it is close to 0 then it is close to perfect.

   • Suggested homework solutions for neural networks assignment in pdf format. (2000/6/9)
   • Suggested homework solutions for neural networks assignment in postscript format. (2001/10/25)

Examination Schedule

• T B A

Course Description

Advanced Artificial Intelligence covering artificial neural networks, speech processing and recognition, genetic algorithms, and learning theory

Irwin King, HSH 908, +(852) 2609-8398, king@cse.cuhk.edu.hk