ET-805 ATS Domain Module Ramkumar.Rajendran@iitb.ac.in
From Last Class ATS and ITS, Architecture ● Course Contents and Assessment ● Muddy Points - Components in ITS architecture - We will discuss each components in detail in our course - Difference between ATS and ITS 2
Activity - Peer Instruction Describe the working of MindSpark PI question: Would you classify this as an ITS or an ATS? ITS ● ATS ● Discussion and then vote again. 3
Activity - PI Describe the working of UrbanSim PI question: Would you classify this as an ITS or an ATS? ITS ● ATS ● Discussion and then vote again. 4
Activity - TPS Think: Pick one concept in a topic in a domain of your choice. For example, algebra for class 6. Develop your domain model for that topic. (3 minutes) Pair with your Neighbour: Identify one important characteristic that is present in your neighbours model which is missing in your own. Together, identify 1-2 difficulties in developing a domain model. (5 minutes) Share to the class (7 minutes) 5
Class Response Characteristics of Domain Module - Multiple concepts in a topic, hierarchical order - Assessment - Prior knowledge - How to solve the problem - experts view Challenges -
Domain Module Needs to have explicit model of domain knowledge, including ● identification of areas where students might have difficulty and what action is to be taken For example: Algebra for class 6 ● Teach value of Multiple. One Pencil cost 3 Rs. then five ○ pencil will cost Introduce variables ○ Expression with Variables. Ram have 5 pencils more than ○ Raju (5+X) Nkambou, R. (2010). Modeling the domain: An introduction to the expert module. In Advances in intelligent tutoring systems (pp. 7 15-32). Springer, Berlin, Heidelberg.
Domain Module Should provide mechanism to validate learner’s interaction in ● the system wrt model For example: Class 6 Algebra ● Questions at each level to validate students performance ○ If the system can detect confusion then new content to ○ topic Nkambou, R. (2010). Modeling the domain: An introduction to the expert module. In Advances in intelligent tutoring systems (pp. 8 15-32). Springer, Berlin, Heidelberg.
Domain Module Should be written in structured format that can be compared ● with student’s actions. For example ● Hierarchical Semantic Network ○ Concept Maps ○ Nkambou, R. (2010). Modeling the domain: An introduction to the expert module. In Advances in intelligent tutoring systems (pp. 9 15-32). Springer, Berlin, Heidelberg.
Black Box Models SOPHIE I - a reactive learning environment ● ○ Picture Credit from: Brown, J. S., & Burton, R. R. (1974, January). SOPHIE: a pragmatic use of artificial intelligence in CAI. In Proceedings of the 1974 annual ACM conference-Volume 2 (pp. 571-579). ACM. 10
SOPHIE
Interaction Scenario in SOPHIE
Activity - Peer Instruction In SOPHIE, the mechanism to validate learner’s interaction can be found in line _____ Discuss with your neighbour. Answer again. 13
Activity - Each one say one Why do you think SOPHIE is called a Black Box Model? Share to the class Class Response - We don’t know what is inside the box? - No detailed rules to solve the problem - 14
Black Box Models - Students can ask system to measure and ask question to troubleshoot the circuit and create their hypothesis - An expert system validates students responses - Tutor suggests or poses questions by solving the problem as experts do - It is student’s responsibility to understand the response and create his/her own hypothesis
GUIDON with MYCIN as Domain Expert - Guidon is a tutor program to teach diagnostic problem-solving tasks - MYCIN is the database or domain module that is used in GUIDON - Knowledge base developed by physicians and AI experts - MYCIN consists of 450 rules for diagnosis of bacteremia, meningitis, cystitis infections - MYCIN explanations are based on its rules and goals - Why => “How is this information useful” => "In what rule does this goal appear, and what goal does the rule conclude about?" Clancey, W. J. (1983). GUIDON. Journal of computer-based instruction
MYCIN - Sample Rules Questions: “Is the patient a compromised host (rule 343) ---~ is the patient immunosuppressed? (rule 395) ---> is the patient receiving steroids? (rule 543) is e.coli causing the meningitis? “
MYCIN - Sample Rules
MYCIN Database “how a request for data is related to a goal, how one goal leads to another, and how a goal is achieved”
MYCIN Rule Example
GUIDON and MYCIN features - Exhaustive backward search - GUIDON - Tutor Module compares student’s question with possible MYCIN questions for the same problem
Activity - TPS Think: What is happening in MYCIN and what is the domain model. Write down what is the expert knowledge in MYCIN. (3 minutes) Pair with your Neighbour and discuss why MYCIN might be called Glass box model? Write down your reasons. (5 minutes) Share to the class 22
Class Response - Rules and goals are clear or visible - glass - Although the mycin developed on expert knowledge it is confined to its rules -
Glass Box Models Unlike Black Box models, rules and goals are visible to the learner. - Constructed by an human expert. - Each reasoning step can be inspected by student. However, justification of rule is only done by an expert, so it is still a box; learner cannot go outside the rules created by the expert.
Activity - TPS Think: What are drawbacks that persist in both Black box and Glass box models? Each one write down one point. (3 minutes) Share to the class 25
Class Response - Rules are fixed - Not scalable - Both are confined to set of restriction - both are box - Expensive and time consuming - No feedback based on learners’ interaction/response
Problems with Black/Glass Box Model Justification is by Expert ● Problem that are new to domain are not addressed ● Time consuming ● Hours of consultation with experts ○ Learner’s hypothesis is not supported ●
In today class... Domain Module from historical Perspective ● SOPHIE - Black box Domain Model ● MYCIN - Glass box Domain model ● 28
Next class ... Rule-based Domain Models ● Constraint-based Domain Models ●
Last Activity - Muddy Points List down - two important and - two least clear (muddy) points from today’s class - https://tinyurl.com/et8 05mp 30
Domain Languages Production rules - E.g., MYCIN ● Semantic Networks ● Nodes and arc ○ Inheritance and other ○ relations between categories can be represented Not ideal for data ○
Conceptual Graphs - Based on Semantic networks but can be represented as first-order predicate logic
Activity Pair create domain model using Semantic of conceptual graphs for the topic you at the beginning of the today’s class (7 minutes) Share to the class 33
Recommend
More recommend
