INTESYS / Intelligent Systems De La Salle University College of Computer Studies Course Syllabus Prerequisites : DISCTRU Type of Course : Basic course Term : Term 1 AY 2012-13 Faculty : Dr. Raymund Sison and Ms. Rhia Trogo (Contact details on the last page) Course Description: Artificial Intelligence (AI) is the study of how machines can exhibit at least one of the following aspects associated with intelligent behavior: (a) problem solving, or the performance of non-trivial, goal-directed cognitive tasks even in the face of inadequate (e.g., incomplete, incorrect, inconsistent, or vague) data; (b) reasoning, or the drawing of logical inferences and conclusions from possibly inadequate evidence; (c) learning, or the improvement of performance through experience; and (d) natural language processing. This 3.0-unit course introduces the CS-ST major fundamental concepts, principles, and techniques in searchbased problem solving, reasoning, and machine learning, and in the representation of the knowledge needed perform these tasks. Learning Outcomes: At the end of the course, the student is expected be able : LO1. Collaboratively design and evaluate algorithms and representations for effective path finding and tactical/strategic decision making in a game environment; LO2. Collaboratively model the knowledge of, engineer, and evaluate an expert system protype that addresses a need of a particular user group, community, or organization; and LO3. Collaboratively perform comparative analyses of algorithms for problem space search and machine learning using real-world datasets. The learning outcomes (LOs) are aligned with the Expected Lasallian Graduate Attributes (ELGAs) as follows: ELGA LO Critical and Creative Thinker LO1, LO2, LO3 Effective Communicar LO1, LO2, LO3 Reflective Lifelong Learner LO1, LO2, LO3 Service-Driven Citizen LO2, LO3 Major Course Outputs: As evidence of attaining the above learning outcomes, the student is required do and submit the following during the indicated dates of the term. The specs of these outputs can be downloaded from http://ccs1.dlsu.edu.ph/intesys/. Learning Outcome Required Output Due Date LO1 MCO1: An AI Cap n Game Bot and Evaluation of its Performance Week 5 LO2 MCO2: Protype and Evaluation of a Small Expert System Week 10 LO3 MCO3: Comparative Analysis of Machine Learning Algorithms Week 13 1 of 7
Each student must do each project with a different group. Moreover, although the major course outputs are all be done collaboratively in groups, every student must write code in MCO1, write rules for and run the group s expert system in MCO2, and use Weka generate and evaluate classifiers in MCO3. If, during an MCO demo, a student is unable debug the group s code for MCO1 or MCO2, or run Weka on a slightly modified dataset in MCO3, he/she will get a 0.0 in that MCO. Rubrics: MCO1: An AI Cap n Game Bot and Evaluation of its Performance (Maximum points: 15) Criterion Exemplary Satisfacry Developing Beginning Effectiveness of Bot 4 points The bot is among the p 10% of the batch. The bot is among the p 25% of the batch. The bot is among the p 50% of the batch. The bot lost early in the urnament. Analysis of Bot s Intelligence Quality of Recommendation (for improving bot s performance) 7 points make the bot described correctly, thoroughly, and using carefully thought out figures and graphs. 4 points There are several recommendations, all of which are reasonable and insightful; the thorough explanations reveal how insightful the explanations are. 5-6 points make the bot described correctly and adequately; figures or graphs are adequate. There are several recommendations, all of which are reasonable and adequately explained; some are insightful. make the bot described, but there are either some errors or the description is inadequate. There are a few recommendations, but the explanations are not adequate; some are insightful. 1- make the bot described inadequately and incorrectly. There are only a few recommendations, none of which is adequately explained and insightful. MCO2: Protype and Evaluation of a Small Expert System (Maximum points: 20) Criterion Exemplary Satisfacry Developing Beginning Originality 5 points The ES clearly addresses an important need of a local or international group, community, or organization that involves or caters the poor. The ES clearly addresses a real need of a local group, community, or organization. The ES somewhat addresses a real need of a local group, community, or organization. The ES does not address a real need of any local group, community, or organization. Quality of ES Recommendations 9-10 points There is a considerable number of rules and levels of rules, all of which are correct (faithful reality) and represented correctly; the diagnoses/ recommendations of the system are clearly demonstrated be the same as what an expert in the domain would make. 6-8 points There is a sufficient number of rules and levels of rules, all of which are correct (faithful reality) and represented correctly; the ES diagnoses/ recommendations are accurate more than 50% of the time. 3-5 points There is an insufficient number of rules and levels of rules, though all are correct (faithful reality) and represented correctly. 1- There is an insufficient number of rules and levels of rules, some of which are incorrect (faithful reality) or represented incorrectly. 2 of 7
Analysis of ES 5 points The analysis clearly shows and explains, both quantitatively and qualitatively, all the major strengths and weaknesses of the ES. The analysis clearly shows and explains, mostly qualitatively, the major strengths and weaknesses of the ES. The analysis shows and explains, mostly qualitatively, some of the major strengths and weaknesses of the ES. The analysis does not explain clearly any of the major strengths or weaknesses of the ES. MCO3: Comparative Analysis of Machine Learning Algorithms (Maximum points: 15) Criterion Exemplary Satisfacry Developing Beginning Description of the Experiments Datasets Analysis of the Learning Curves study and the design of the experiments are described clearly and thoroughly; the experiments are designed correctly; the algorithms are described well using original illustrations. 4 points The dataset is sourced locally; the dataset and its described clearly and correctly. 7-8 points The learning curves are correct; expected as well as unexpected aspects of the learning curves are discussed clearly, reasonably, and thoroughly. Other Requirements and Assessments: study and the design of the experiments are described clearly and thoroughly; the experiments are designed correctly; the algorithms are described well. The dataset and its described clearly and correctly. 5-6 points The learning curves are correct; some expected and unexpected aspects of the learning curves are discussed reasonably. study is described clearly; the design of the experiments is discussed, but not thoroughly, though it appears be correct; the algorithms are not described so clearly. The dataset and its described, but not clearly enough or there are some minor errors in the description. The learning curves appear be correct, though they are not adequately discussed. 0 point study does not appear be clear the group; the design of the experiments is not discussed thoroughly and appears be incorrect; the algorithms are not described clearly. The dataset and its described neither clearly nor correctly. 1- At least one learning curve is incorrect. Aside from the three major course outputs above, this course has 3 other kinds of requirements: class participation, exams, and a final assignment. Class participation (recitation, snap quizzes, special oral reports) Students may accumulate up 20 points through class participation. Students who recite must e-mail the professor within 48 hours the number of points and questions answered. The subject of the e-mail must be: INTESYS <section> Recitation - <points> - <last name if not on e-mail address> <first name if not on e-mail address>. For example: INTESYS S17 Recitation - 1 - Sison Raymund In the body of the e-mail, specify your name (last name first), the question asked by the professor, and the answer that you gave which should, of course, be correct. The slides of any special oral reports must use the masters of the slides of this course, and must be revised address the comments of the professor and class and e-mailed the professor within 48 3 of 7
hours of the oral presentation. The subject of the e-mail must be: INTESYS <section> Special Report - <last name if not on e-mail address> <first name if not on e-mail address>. For example: INTESYS S17 Special Report - Sison Raymund Exams There will be two exams. The first exam will cover search and representation; the second, reasoning and machine learning. The weeks of these exams are specified in the learning plan below. The exact dates and times will be determined by the professor and students on the first week of classes. Final Assignment A project proposal for using AI address a real need of the country (or of a local group, community, or organization) will be written and presented. Grading System: To pass this course, one must accumulate at least 60 points through the course requirements discussed above. The maximum points that a student can obtain through each requirement are shown below. Assessment Task Maximum points Class Participation 20 Exam 1 10 Exam 2 10 Major Course Output 1 (AI Cap n) 15 Major Course Output 2 (Expert System) 20 Major Course Output 3 (Machine Learning) 15 Final Assignment (Local Application of AI) 10 TOTAL POINTS 100 Teaching Methods / Strategies: Discussion (D), exercises (E), lecture (L), recitation, reflection (R), quizzes, written reports, oral reports (OR), oral presentations, assignments, group projects and intermediate and final deliverables (MCO) Learning Plan: LO 1 Topics and Readings Course Requirements; AI Definition and Hisry (AIMA2E*Ch.1) Search: Blind (AIMA2E*Ch.3) Wk/ Date 1 5/24 5/30 2 5/31 6/6 Learning Activities 1. Self introductions 2. Explanation of course requirements and policies 3. L: AI is everywhere 4. R (e.g.: Machines/programs are becoming more and more intelligent.) Submission of 1x1 ID picture 1. D: Early AI (1950s - 1960s) 2. D: Boomtimes (1970s - mid 1980) 3. D: The AI Winter (mid late 1980) 4. D: AI s Reemergence (late 1980 1990s) 5. R (e.g.: AI day is less conspicuous but more ubiquius than it was prior the AI winter.) MCO1: Submission of groups for MCO1 1. D: Problem solving as search/problem space formulation 2. D: General tree search algorithm (ff AIMA2E*) 3. D: BFS & DFS 4. R (e.g.: Problem solving can be viewed as state space search.) [MCO1: AI Cap n Orientation ] 1. D: BFS and DFS variants 2. D: Comparative analysis (time and space complexities) 3. D: Best first search & introduction A* 4. R (e.g.: Heuristics, which are a type of knowledge, can vastly speed up state space search.) 4 of 7
LO 2 Topics and Readings Search: Heuristic (AIMA2E*Ch.4) Search: Game Trees (AIMA2E*Ch.6) Representation: Propositional Logic (AIMA2E*Ch.7) Representation: Predicate Logic (AIMA2E*Ch.8) Reasoning: PROLOG (PROgramming in LOGic) (Luger, 2005, Ch.15) (Blackburn et al, 2006) Reasoning: Rule-based Wk/ Date 3 6/7 6/13 4 6/14 6/20 5 6/21 6/27 6 6/28 7/4 7/5 7/11 Learning Activities 1. D: A* 2. D: Admissibility & dominance of heuristics 3. D&E: Designing heuristics 4. R (e.g.: Some heuristics are better than others.) [MCO1: AI Cap n Workshop ] 1. D: A video game environment 2. D: Application path finding in a video game environment 3. OR: Basic path finding in AI Cap n 4. R (e.g.: A* is the most important search technique remember.) 1. D: Hill climbing and the problem of local maxima 2. D: Simulated annealing 3. D: Genetic algorithms 4. R (e.g.: GAs are an effective albeit subsymbolic approach problem solving.) [MCO1: AI Cap n Tournament ] 1. D&E: Minimax 2. D&E: Minimax with alpha-beta pruning 3. A: Extending minimax n players 4. R (e.g.: The problem space of turn-based games take in consideration the fact that players oppose each others moves.) Submission of MCO1 MCO2: Submission of groups, tentative domain, diagnostic task, and experts 1. D: Propositional logic syntax and semantics 2. D: Modus ponens and forward and backward chaining 3. D: Resolution 4. R (e.g.: Propositional logic has limited expressive power.) 1. D: Predicate logic syntax and semantics 2. D: Unification 3. D: Forward and backward chaining 4. R (e.g.: Predicate logic is a powerful, general-purpose language for representing knowledge.) 5. MCO2: Professor s feedback on tentative domain, diagnostic task, and expert MCO2: Submission of domain, diagnostic task, 2 experts (one supply knowledge, another evaluate the system), partial knowledge base (AND/OR graph only) 1. D: Conjunctive Normal Form 2. D&E: Resolution 3. E: Resolution 4. R (e.g.: Resolution is an efficient way make logical inferences.) 1. D: Prolog facts, rules, procedures, and queries 2. D: Lists in Prolog (member/2) 3. D&E: length/2, sumlist/2 4. A: Read and implement 15.7.1 of Luger (2005) 5. R (e.g.: Prolog is a fascinating implementation of Horn clause logic.) 6. MCO2: Professor s feedback on domain, diagnostic task, experts, and partial knowledge base 7 Submission of pre-final knowledge base (AND/OR graph only) 1. D: A meta-intepreter for Prolog 2. D: Running append/3 through the meta-interpreter 3. E: Running reverse/2 through the meta-interpreter 4. A: Read and implement 15.7.2 of Luger (2005) 5. R (e.g.: It is easy in Prolog write programs that manipulate expressions written in Prolog.) 1. D: Sample consultation with MYCIN 2. D: Basic expert system architecture 3. D: Mini ES shell in Prolog: The knowledge base 4. R (e.g.: The most important component of an expert system is its knowledge base.) 5. MCO2: Professor s feedback on pre-final knowledge base Exam 1 8 1. D: Mini ES shell in Prolog: The inference engine 2. D: Mini ES shell in Prolog: The user interface 3. D: Sample consultation 5 of 7
LO 3 Topics and Readings Expert Systems (Surrey, 2001) (Luger, 2005, Ch.8) Reasoning: Uncertainty (Luger, 2005, 9.2) Machine Learning: Decision Trees (AIMA2E*Ch.18) Machine Learning: Neural Networks (AIMA2E*Ch.20) Philosophy of AI (AIMA2E*Ch.26) Wk/ Date 7/12 7/18 9 7/19 7/25 10 7/26 8/1 11 8/2 8/8 12 8/9 15 Integration 13 8/16 8/22 Learning Activities 4. R (e.g.: A backward-chaining inference engine is appropriate for diagnostic tasks.) MCO2: Presentation of expert system protype MCO2: Submission of evaluation results Additional Prolog exercises: 1. E: append/3 2. E: reverse/2 3. E: rember/3 4. E: ordered/1 5. R (e.g.: Recursion, rather than iteration, is more appropriate for declarative languages.) 1. D: Incremental learning using + and - examples 2. D: A data set and a corresponding decision tree 3. D: ID3 s basic algorithm 4. R (e.g.: Both positive as well as negative examples of a concept are necessary for a correct definition of the concept be learned.) 5. MCO2: Professor s feedback on evaluation results This meeting be called off in lieu of Exam 1 Submission of MCO2 MCO3: Submission of groups and dataset for MCO3 1. D: Information content and gain 2. E: Computing gain and selecting root for sample dataset 3. E: Completing the decision tree 4. R (e.g.: Heuristics for AI algorithms can come from various disciplines.) 1. D: Components of a neuron 2. D: Perceptron training algorithm 3. E: Using Excel for perceptron learning for sample dataset 4. R (e.g.: The delta rule for perceptron learning weights the difference of the expected and actual outputs.) 5. MCO3: Professor s feedback on dataset MCO3: Submission of decision tree 1. D: Comparing decision tree and perceptron learners 2. D: Linear inseparability problem of perceptrons (e.g., the exclusive-or or XOR function) 3. D: A multilayer network for XOR 4. D: Backprop learning algorithm 5. R (e.g.: The perceptron delta rule is expanded in the Backprop algorithm include the contribution of the nodes in the hidden layers.) MCO3: Submission of results of a single cross validation of both classifiers 1. D: The Chinese room 2. D: Consciousness, identity, and emotion 3. D: Is strong AI possible? 4. R (e.g.: Strong AI remains an elusive goal.) This meeting be called off in lieu of Exam 2 Exam 2 MCO3: Professor s feedback on results of single cross validation Submission of Final Assignment 1. Selection of best proposal per group of 8 2. Presentation of best proposals Submission of MCO3 1. R: Most important lesson learned in this course 2. D: Where do we go from here? (The Road Ahead) Notes: * AIMA2E stands for Artificial Intelligence: A Modern Approach, 2 nd Ed. These activities are held outside classroom hours. The AI Cap n seminars and competition will be handled by the Game Lab. This pic is a reading assignment for which there will be no lecture. 6 of 7
Text / Materials: Russell, S. & Norvig, P. (2010). Artificial Intelligence: A Modern Approach, 3rd Ed. New Jersey: Prentice-Hall. This is the AI textbook used by p Computer Science departments in the world. A low price 2 nd edition (with a green soft cover) is available in Philippine booksres. There are also a few copies at the DLSU library; two of these have been placed in the Circulation-Reserve section. References (including online resources): Blackburn, P., Bos, J. & Striegnitz, K. (2006). Learn Prolog Now! UK: College Publications. There is a free online version of this book at: http://www.learnprolognow.org/. Jones, M. T. (2008). Artificial Intelligence: A Systems Approach. Hingham, Massachusetts: Infinity Science. This is the newest addition the set of good AI textbooks. It is easier read but is less comprehensive than both (Russell & Norvig, 2010) and (Luger, 2009). It also contains C implementations of the basic AI algorithms. A copy is available at the DLSU library, and has been placed in the Circulation-Reserve section. Luger, G. (2009). Artificial Intelligence: Structures and Strategies for Complex Problem Solving, 6 th Ed. Harlow, Essex: Addison-Wesley. This is easier read than (Russell & Norvig, 2009). A copy of the third edition is available at the DLSU library, and has been placed in the Circulation-Reserve section. Also available in Philippine booksres. Surrey. (2001). Expert System Case Studies: MYCIN. University of Surrey. This archived site contains a sample consultation with MYCIN, which established the methodology of contemporary expert systems (Luger, 2005): http://www.computing.surrey.ac.uk/ai/profile/ mycin.html. General Report Formatting and Submission Guidelines: The General Report Formatting and Submission Guidelines are on a separate document that can be downloaded from the professor s website or some other location that the professor will specify. When submitting any written documents in this course, these guidelines must be adhered. Nonadherence a guideline will merit deductions. Note, o, that the penalty for plagiarism is failure in the course. Class Policies: Other policies specific the professor s classes are on a separate document that can be downloaded from the professor s website or some other location that the professor will specify. Faculty Contact Details: Dr. Raymund Sison E-mail raymund.sison@delasalle.ph (Ensure that the subject of your e-mail contains the course code (INTESYS) and your section.) URL Ms. Rhia Trogo E-mail rhia.trogo@delasalle.ph http://mysite.dlsu.edu.ph/faculty/sisonr/#teaching (Class participation, general report formatting guidelines, and classroom policies of Dr. Sison s classes will be posted here.) Course download site http://ccs1.dlsu.edu.ph/introai/ (The course slides and specs for the major course outputs are here.) /RCS 7 of 7