August 16, 2009
Adapting the MDP
Taking a baby step forward after Whimsy and Smarty's conversation the other day, here are a few more thoughts.
What, precisely, is a Markov Decision Process?
- The definition includes a probability distribution, P, showing the belief that a particular action will cause a transition from state to state .
- The definition of an MDP includes a reward function that lets the designer influence the choices made. The reward, R, is defined over the same space s the probability, i.e. a reward is assigned for a given action that maps a previous state to subseuent state .
- I believe that MDPs usually operate over discrete time.
- You usually pre-define a set of actions.
- You usually pre-define the set of states.
The last two points pose a problem fro my situation, but I think there must be some way to extend the MDP to operate over unbounded action and state spaces, probably by allowing some other restriction instead, somewhere else. Maybe you just need a more flexible algorithm during the decision making step.
But to me, the most interesting part of the MDP is P and R, i.e. the first 2 points. Looking at P: It's like it assumes we have pre-defined the states and just don't know which one we're in. But this is not true for my situation. I am using an *incomplete* state, really. And my set of state transitions is fully known (if you choose to look at a teaching strategy as a series of steps, or, their own little state transitions in a way.)
So where is the uncertainty? (Where can we use probability?) How about, the uncertainty is "that my state space is in X", where X is some proposed enumeration. X can be my projection/anticipation of the learner's set of building materials, kind of. I talked bout this more in this older post.
Other lingering thoughts:
- I want to talk about X some more, to explore this concept of the flexible set, that is defined in terms of preceptors rather than of subsets of some fully enumerated superset.
- 'Will have to re-read my older thoughts on the utility function.
- preceptors to specialize on subsets of the state
Post a comment
Index to Steph's NotesFeb. 24th 2007 - Weee! This new part of my website is not an entry, but rather a permanent fixture whose purpose is to "Look Down on All Those Notes With Some Grand Vision of Organization". Wish me luck. LOL
- Representing meta-data (fuel) & the different kinds of "hooks" that intelligent systems can use (how fuel is injected into the motor of the engine)
- Motivation: Semantic net / Rationalizable to a machine
- Semantic network
- Genetic graph
- Prerequisite AND/OR graph
- Constraint Satisfaction Problems
- Bayesian networks / causal graphs
- Technology & Philosophy: RDF, modus ponens,
- Predicates, Logic & situation calculus
- What kinds of data? - What kinds of meta-data would an AIEd system possibly need, and how is it represented?
- task domain knowledge
- "is-prerequisite-to"-type knowledge
- interactions with learning objects & other learners - (location, composition is-a/part-of, sequencing by restricting navigation, personalization, ontologies for LO context)
- lesson plans, curriculum plans, practicing sessions (What is stored, what is generated on the fly? What is remembered?)
- How to organize it - When is it stored in a database? Meta-data? Agent memory banks? Protocols? Repositories? XML files? Home-servers? WSDL services? Frameworks? Portable banks? P2P access?
- Database of object-agent interactions
- Concept of "Home" on a P2P network -- maybe the bulk of a learning object's usage data is on its home server and can be queried using WSDL or something ? Similar homes for each student's usage history, etc. Baggage problem.
- Links to the ontologies
- referring to a concept/relationship - ex. AgentOwl?
- Generation of this data
- Rationalization: For use by other AIEd systems
- What is generated - discuss items under part I.C.
- When it's generated - describe procedural model, which parts of the engine generate what (isa-part-of data, XML feeds, web services, meta data bout groups and collaboration, protocols, examples Friend of A Friend FOAF project)
- Technical notes of HOW it's generated: JENA, issues of implementation demo, my Hermione & Ron agent examples, lol
- Usage of this generated data - see part IV. A.
- Given the engine, who uses it?
- Students / Learners / "Me"
- instructional planning, student model, pre-requisites, tutoring, coaching, collaboration,constructivism
- Teachers / Educators / "Me"
- putting together lessons
- be able to browse through task domain knowledge in an objective / encyclopaedia format, then be able to pick-and-choose what you need for your students
- compose examples, design explanations, pull together diagrams, learning objects, etc. Haystack Relo?
- Administration / Governement / Structure / Crowd Control
- as restrictions/obstacles/sand pit to the robot in agent environment
- can't just have a swarm of students and teachers out there -- need structure of courses, curriculum, objectives, requirements (at least, we do in this day and age!) - Report cards, evaluation, feedback
- government, marks, certificates, requirements, funding, curriclum, attendance, delinquent, non-attending, motivation
- school''s images, goals, strengths, payroll, HR, security, accounts, permissions, privacy
- registration, failed courses
- User Environment -- How does this engine work? What does the user see on the screen?
- Introduction - Given a background in educational psychology, how does the system present itself -- what does the user see, and were does this data come from? Links to thoughts from part I.)
- Task Domain Browsing - Suppose you're you're just idly browsing through the "raw" content. How would it look when it's not wrapped around a learning-context or lesson or tutorial or anything. 'Cross between browsing a raw task domain ontology and browsing a learning object repository.
- Cleaning up the data -- Visualizing the data for humans to pick through the task domain and work on it. Suppose the "Subject Expert" discovers an advancement in science and needs to update the "world's" domain knowledge. (I used the "Subject Expert" terminology from Ontologies to Support Learning Design Context - Thanks Chris) How would they make corrections to ontologies and learning objects, or at least point the users of "old" objects towards adopting the newer ones.
- "Modes" - Learning & Lessons / Checklist - Homework, Assignments, Courses being taken / Collaborative mode / Teaching mode / Calendar- email -adminisrative mode -- See also the different kinds of scenarios in the ActiveMath system
- Evolution of this engine
- target some key implementation hooks discussed in part I - design an experiment/demo
- scrape a page - (Note, scraping can only give objective data, not in-context dat)
- LO repository - related to browsing the task domain?
- a learners "To Do" list - where does it come from? Assignments, courses.
- sample group scenario
- sample teacher lesson planning
- sample data "left behind"
- sample use of that data
- Data mining (for what? lol )
- discovery / generation of ontologies - when do you need to hunt for them, and when do you have to have a solidly-known & predictable ontology?
- I/O - where it happens, which languages, protocols, which agents perform i/o and when, precepts, actuators
- Role Assignments
- My Environment Adapts to me
- Displaying feedback from the server on JSP pages (Software engineering considerations)
- Sketching out a design (Content planning vs. Delivery planning)
- agent negotiations / social structures / ummm... Web 2.0 ?
- garbage collection of meta data
- Artificial Intelligence & Evolution
- Memory Culling: Necessary part of intelligence? (artificial or human)
- Applications for the Genetic/Evolutionary algorithm
- open learning environments
- Agents, pets, grouping, Community modelling
- Protocols - finding groups, cyber dollars, state diagrams (?)
- "Community Studies" - graphs & communication hubs, types of communities (free-for-all, hierarchy of authority, etc.)
- implications of joining a community - what do you share, which parts of your student model are relevant
- Walls & sand traps -- deliberate restrictions as problem-solving for learning
- Communication channels - individual-to-individual, individual-to-community, chat channels, agent-only "administrative" communications, ex. requests for related learning objects in a particular community, etc.
- Educational/Pedagogical focus (this part probably shouldn't be its own section but rather incorporated into the whole picture, but it's separate for me right now because I'm still only just starting to learn about it.)
- Semantics - what there is to talk about in Education
- ex. Merril's First Principles of Instruction, linking educational terms to AI terms
- Pedagogical skills for tutors -- supporting human *and* artifical tutors
- Student modelling - what the machine needs to know about the student, pedagogically-speaking, about learning history/preferences
- Roles - Simulated students, Coaches, Tutors, Teachers,