辅导案例-CS5010
CS5010 Practical 2, 2019
CS5010 Artificial Intelligence Principles Practical 2

Game Playing: Learning and Self-Play


Aim
There is a long history in Artificial Intelligence of getting AI systems to play games. This includes deterministic
games like Chess, Go, and Checkers as well as games with randomness and hidden information like Poker,
Bridge, and Backgammon. You are to investigate the topics of Learning and Self-Play within Game Playing in
AI, and then write a report on these topics. While we teach the topic of Learning in the module, we have not
taught Game-Playing or Self-Play, so part of the practical is to research those topics.
“Self-Play” is taken here to refer to artificial game systems that have clones of the system play each other in
order to acquire skill at playing the game. That is, instead of acquiring skill just through human expert tuning of
the system, and/or by analyzing games played by the system against people, the system plays itself repeatedly.
This practical counts for 55% of the pactical marks for the module and 22% of the overall marks of the module.

Specification
Throughout the report, use at least two games to illustrate your answer. These should include at least one classic
deterministic game (e.g. Chess, Go, Checkers), and at least one classic game with random elements (e.g.
Backgammon, Poker, Bridge).
Your report should include the following four things:
1. Background: Provide a brief historical account of game-playing in AI.
2. Answers to the following Learning Issue Questions: To what extent has machine learning been important in
developing high-quality AI programs for game playing? Have key breakthroughs depended on machine
learning? If so, have they depended on new advances in machine learning or on using existing techniques in
new ways?
3. Answers to the following Self-Play Issue Questions: To what extent can self-play be applied to a variety of
games? Superficially at least, self-play appears to allow any AI game player to acquire great and continually
increasing expertise even if initially it has no expertise at playing well beyond playing according to the game
rules. Is this actually true? If so, why? If not, why not? Should analysis and conclusions vary for different games
and AI mechanisms? If so, why? If not, why not? How has Deep Learning been successfully applied to self-
play?
4. Answer to the History Question: To what extent have the answers to the above Learning and Self-Play
Issues changed through the history of AI?

Research and Analysis
You should draw upon not only what you have been taught in class, but also external references you have
researched and read for this report. You will also need to investigate game playing AI in terms of general
principles, though you will not need to demonstrate a detailed understanding of the particular AI techniques
involved.
You should decide on the most appropriate way to cover the topic. While factual information from the literature
is important, we are also expecting you to engage with it and analyse it appropriately. For example, if a source
says that X is the reason for the success of an AI program, has X also been important for other similar games?
What are the AI principles that lead you to your conclusion?
You might wish to provide a literature review and analysis of relevant material, or to cover less material but
provide some practical material (e.g. downloading an AI game to play against itself and investigate the results).
CS5010 Practical 2, 2019

Deliverables
The deliverable for this project comprises a report written in the typesetting style of your choice. The advisory
word limit for the main text (including the Abstract) is 3,600 not including references. Your report should have
the following structure:

1. Title.
2. Abstract (maximum 200 words).
3. Introduction.
4. Content sections (and possibly subsections).
5. Conclusion/Summary
6. References/Bibliography

All figures and tables should have legends/captions. Any quotes from sources should be clearly formatted as
quotes and fully referenced (using APA referencing style). While books and research papers are preferable,
authoritative websites are also allowed in this case. Valid URLs for these sites should be included.

Hand in via MMS, by the deadline,
• A single document in PDF format containing your requirements and diagrams.
• If you have supplementary materials such as analysis of games you have played, then include these in a single
zip file including the PDF report

At time of writing the deadline is planned for 9.00pm on Friday of Week 9, 15th November, but the deadline as
stated in MMS is definitive.
Marking
See the standard mark descriptors in the School Student Handbook:
http://info.cs.st-andrews.ac.uk/student-handbook/learning-teaching/feedback.html#Mark_Descriptors
See also
https://info.cs.st-andrews.ac.uk/student-handbook/learning-teaching/assessment.html
for what is meant by "advisory word limit".

Lateness
The standard penalty for late submission applies (Scheme B: 1 mark per 8 hour period, or part thereof):
https://info.cs.st-andrews.ac.uk/student-handbook/learning-teaching/assessment.html#lateness-penalties

Good Academic Practice
The University policy on Good Academic Practice applies:
https://www.st-andrews.ac.uk/students/rules/academicpractice/

Some Hints
1. Avoid long quotes or paraphrasing without thought!

2. Try to view the report as a logical argument building up to a conclusion which is insightful and convincing.

3. Ask yourself after each paragraph what part of the specification is being addressed as directly as possible.

4. Back up claims with either logical argument or references to published work that backs up the claim.

Starter References
Rusell, S. and Norvig, P. (2010), Artificial Intelligence: A Modern Approach, 3rd Edition, Pearson. See chapter
on “Adversarial Search” for game playing.

Ian Gent, October 2019

51作业君 51作业君

扫码添加客服微信

添加客服微信: IT_51zuoyejun