### CSE 471/598 Homework, Projects, and Exams

Notes: (1) Homework's and projects should be submitted before 4:00pm on the due date.
To make it more convenient to you, the deadline is further extended to Tues. instead of Mon.;
(2) Late penalty applies at 2 (absolute) points a day. Tip: don't be late as you're busy with
other courses too.
(3) You'll have a choice of doing 4 sets (7.5 points each) or 5 sets (6 points each) of homework.
As we agreed in the classroom, the last set is optional if you choose to do 4 sets
of  homework. You can make your decision on the last due date.
(4) You can submit your homework to the instructor in class or to TA: Mr. Dasari, GWC 309.

HW1: (Problem Solving) Ex 1.7, 2.4, 2.10, 3.3, 3.4

HW2: (Knowledge and Reasoning) Ex  6.7, 6.12, 7.2 (a, b, c, d), 7.11 (a, b, c, d) , 9.4, 9.5 (a, b)

HW3: (Learning) Ex 18.3, 18.3 extended (create a decision tree about the move-forward for the wumpus world, and propose how you can make a learning agent for this world - it's related to the second project), 18.5, 18.7
Deadline: Mar 24 (Fri.), 2000 during the class.

HW4: (Acting Logically) Ex 11.2 (a) and (b), 11.4 (Hint: without POP, can you do that? If you can't, why can't? You are actually discovering the anomaly!), 13.2 (The Wumpus World only)

HW5: (Uncertain Knowledge and Reasoning) Ex. 14.1 (Hint: The first principles are the definition of conditional probability), 14.6, 15.1 (a), (b), and (c)

P1: (Agent and its environment) 2.5 and 3.17 (4.14b is removed)
Deadline: Mar 20, 2000. Submitted to TA: Mr. Dasari
You need to submit your code with a short report that includes the following

• A brief summary what P1 is
• What're your approach and design
• How to run it and what is the expected result
• Any problem or discussion
P2: Create an intelligent agent via Machine Learning (Decision Tree Induction in the Wumpus world :-)
Deadline: May 1, 2000. Submitted to the instructor or TA: Mr. Dasari
Project Description:
1. Create an environment for the famous Wumpus world (refer to Pages 154-155, a grid of 4X4)
2. Build a random agent (RA) for the task
3. Design a learning scheme so that data from running RA can be collected for learning
4. Learn a more intelligent agent (IA) than RA so that it can react appropriately according to the percepts
5. How many trials (attempts by RA in order of 10, 100, 1000, 10000)  do you need to learn IA?
6. Provide a sample of your data (attribute-values) and list the rules learned from the data
7. Appropriately reacting is better than randomly reacting. However, can we learn even more? What kind of knowledge can your learning system can maximally learn?
8. How does your learning algorithm achieve that? What are the rules (knowledge)?
9. Analysis and discussion

There will be a quiz of 2 problems on  Wednesday, March 22 to help you  review the materials we've covered so far (March 8). The quiz is open book and about 10-15 minutes.

Two sessions of the late mid-term will  be held on Friday April 14, 2000 and Wednesday April 19, 2000 during class hours in the classroom. The first session is about Parts I, II, and III, and the second session is about Parts IV, V, and VI on the topics we have covered until April 7. It's close book.