CHA2555 Assignment 2: 1 Aims
To gain experience in representing and solving problems using AI methods based
on probabilistic reasoning and learning in uncertain environments, and to criti-
cally re ect on this experience.
2 Overall Speci cation
This assignment is divided into two parts. In the rst part, students will model
a physical system involving a motor using a Bayesian network and answer a
number of queries. In the second part, students will create a learning agent for
the Robocode framework and critically evaluate it.
Deliverables will be the code of the agent for the second part, and a short
descriptive and re ective essay for both parts. The essay for the rst part should
include the Bayesian network and a description thereof and the answers to the
questions.
The essay for the rst part should be between 1200 and 1400 words, the
essay for the second part should be between 800 and 1000 words. They should
be submitted by means of a single PDF document via turnitin on unilearn. The
created les should be submitted as a zip le via unilearn. The submission is
electronic only, no paper copies should be handed in.
Each assignment submission will receive a mark between 0 and 100 (each
part amounting to a maximum of 50 marks), which will make up 50% of the
overall module mark.
3 Detailed Speci cation
3.1 Part 1: Bayesian Network
3.1.1 Domain
Consider the following setting: the availability a critical computer system is
monitored by two independent mechanisms. One is sending ping packets over
the network and raises an alarm if there is no return packet within a a second.
The second measures the temperature of the computer system and if it is below
a threshold it will raise an alarm (as the system is then probably not running).
The computer system itself is certi ed by the manufacturer to be up and
running 99.9% of the time. The ping mechanism can be faulty as well, the
likelihood for this is 1%.
If the ping mechanism works correctly, it is reliable in showing uptime: if
the critical computer system is up and the ping mechanism is not faulty, then in
9999 out of 10000 cases the ping mechanism will report the computer system to
be up. It is less precise when the computer system is down, in that case even if
the ping mechanism is not faulty, the ping mechanism will report the computer
system to be down only in 700 out of 1000 cases.
If the ping mechanism is faulty and the computer system is up, then the
ping mechanism will report the system to be up only in 1 out of 10000 cases. If
the ping mechanism is faulty and the computer system is down, then the ping
mechanism will however report the computer system to be unavailable in 800
out of 1000 cases.
The reliability of the thermometer (the second mechanism) is in uenced by
the computer system, as it tends to be more faulty with high temparature (so
when the system is available). In particular, when the temparature is above
the threshold, the thermometer is faulty in 30% of all cases, while it is faulty
only in 3% of the cases when the temparature is below the threshold. When
the thermometer works correctly, it will identify high temparature correctly in
95 out of 100 cases, and it will identify low temparature correctly in 999 out of
1000 cases. If the thermometer is faulty, it will still identify high temparature
correctly in 93% of all cases, but low temparature will be misidenti ed in 98%
of all cases.
Create a Bayesian network that models this situation. It should be thor-
oughly described and justi ed in the essay.
3.1.2 Questions to Answer
Answer the following questions:
How likely is it that the ping mechanism raises an alarm when the system
is available?
Both mechanisms signal unavailability of the computer system. How likely
is it that the computer system is really unavailable?
How likely is it that computer system unavailability goes undetected (i.e.
neither the ping mechanism nor the thermometer indicate unavailability
when the system is actually unavailable)?
The computer system is unavailable, but neither the ping mechanism nor
the thermometer indicate unavailability. How likely is it that the ping
mechanism is faulty? How likely is it that the thermometer is faulty?
The answers should be given in the essay.
3.1.3 Essay
Using 1200-1400 words, describe the Bayesian network and answer the ques-
tions. You should also describe how you came up with the network and the
answers, referring to existing networks (if any) that in uenced the creation of
your network, and any tools that you used. Describe your experience with the
creation of the network and answering the questions. Finally, re ect on what
you achieved on this part of the assignment. (For instance, did everything work
as expected? What problems did you encounter? What did you learn?)
3.2 Part 2: Learning
3.2.1 A Learning Agent for Robocode
Use the Robocode framework in order to implement a learning agent. Your
agent should use reinforcement learning. Other than this requirement, you are
free to choose the tools and techniques, but you need to describe and justify
your choices in the essay. It is preferable to use Java, but agents written in
other languages are aceptable as well.
Obviously, you should come up with your own implementation, using an
existing agent is not acceptable. You may re-use code if you have the right to
do so and if you clearly specify this in the code itself and the essay.
Evaluate your agent by comparing it to three agents that come with the
Robocode framework.
3.2.2 Essay
Describe using 800-1000 words how you produced the agent, referring also to
existing agents (if any) that you have based yours on, to any tools you have used
in producing it. Describe the learning technique you implemented and justify
its use. Describyour experience with Robocode. Finally, re ect on what you
achieved on this part of the assignment.
4 Assessment Criteria
Each of the two parts of this assignment contribute by 50% to the total assign-
ment mark. The assignment mark itself makes up 50% of the module mark.
Part 1 will be marked along the following criteria:
Correctness, documentation, and justi cation of the Bayesian network
40%
Correctness, justi cation, and style. of the answers 30%
Essay content 20%
Essay style. 10%
Part 2 will be marked along the following criteria:
Correctness and style. of the agent code 30%
Documentation and description of agent 20%
Justi cation of tools and techniques 20%
Essay content 20%
Essay style. 10%
5 Academic Integrity
You are reminded of academic integrity. Any breach of academic integrity (such
as plagiarism) can entail sanctions such as the failure of the entire module.
6 Addressed Learning and Ability Outcomes
This assignment assesses learning outcome 1.2, 1.3, 1.4, 1.5 and ability outcomes
2.1 and 2.2 of the module speci cation.
1. Knowledge and Understanding Outcomes
1.2 describe some of central techniques in speci c AI areas eg in auto-
mated planning and/or machine learning
1.3 understand the role of speci c AI research tools and programming
languages in AI modelling and experimentation
1.4 discuss some of the major application areas of arti cial intelligence;
such as in Computer Games or Robot Control
1.5 describe some of the main sub-symbolic approaches used to imple-
ment intelligent systems, such as neural networks or Bayesian net-
works.
2. Ability Outcomes
2.1 Construct and reason with knowledge representations within a range
of AI formalisms, such as rules, action schema, classical logics, Baysian
nets etc
2.2 Con gure, apply and critically evaluate AI methods, and appropriate
tools and techniques, for implementing intelligent systems in appli-
cation areas such as computer games or robot control