AI – Week 21  Machine Learning: Macro Learning Lee McCluskey, room 2/09 Email  [email_address] http://scom.hud.ac.uk/scomt...
Term 2: Summary <ul><li>13 - Introduction to planning and learning </li></ul><ul><li>14 - Introduction to formulations/jar...
Types of Learning <ul><li>Learning by ROTE (remember facts) </li></ul><ul><li>- this is purely storing and remembering fac...
Types of Learning <ul><li>Learning by OBSERVATION (self-taught) </li></ul><ul><li>this is similar to the category above bu...
Recap: Knowledge Acquisition <ul><li>Knowledge  Acquisition   is the process of encoding knowledge in a way that intellige...
KNOWLEDGE/SKILL REFINEMENT <ul><li>Changing/integrating old knowledge to create better (operational) knowledge (Inputs no ...
KNOWLEDGE/SKILL REFINEMENT:  MACRO ACQUISITION AND USE FOR AI PLANNING <ul><li>Roughly: a planner solves a problem and ind...
Macro acquisition: algorithm <ul><li>Starting with goal G, we  regress  through the states backwards: </li></ul><ul><li>As...
Macro Use <ul><li>The triple (G, WP(T,G), T) can be stored and retrieved: </li></ul><ul><li>Given a planning  problem (S, ...
Macro (+Learning) Utility <ul><li>There are other kinds of macro creation: for a solution of size N, N -1 macros can be cr...
Conclusion <ul><li>There are various kinds of Learning manifest in nature and in AI </li></ul><ul><li>Two important roles ...
Upcoming SlideShare
Loading in …5
×

AI – Week 21 Machine Learning: Macro Learning

616 views

Published on

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
616
On SlideShare
0
From Embeds
0
Number of Embeds
5
Actions
Shares
0
Downloads
3
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

AI – Week 21 Machine Learning: Macro Learning

  1. 1. AI – Week 21 Machine Learning: Macro Learning Lee McCluskey, room 2/09 Email [email_address] http://scom.hud.ac.uk/scomtlm/cha2555/
  2. 2. Term 2: Summary <ul><li>13 - Introduction to planning and learning </li></ul><ul><li>14 - Introduction to formulations/jargon of Planning </li></ul><ul><li>15 - Operator Schemas and state space search </li></ul><ul><li>16 - Planner Implementation: BreadthFS in Prolog </li></ul><ul><li>17 - Planner Implementation: BreadthFS, BestFS, Heuristics </li></ul><ul><li>18 - Graphplan Planning Alg </li></ul><ul><li>19 - Graphplan Planning Alg </li></ul><ul><li>20 – reading week </li></ul><ul><li>21 – Knowledge Acquisition + GIPO </li></ul><ul><li>22 - Learning example 1: Macro Learning  ****** we are here </li></ul><ul><li>23 - Learning example 2: GIPO's Opmaker </li></ul><ul><li>24 - Learning example 3: Information Extraction </li></ul>
  3. 3. Types of Learning <ul><li>Learning by ROTE (remember facts) </li></ul><ul><li>- this is purely storing and remembering facts without integrating or recognising the meaning of the facts </li></ul><ul><li>Learning by BEING TOLD (programmed) </li></ul><ul><li>- this is storing and remembering facts / procedures, but implies some kind of understanding / integration of what is being told, with previous knowledge. </li></ul><ul><li>Learning by EXAMPLE/ANALOGY (trained/taught) </li></ul><ul><li>this invovles a benevolent teacher who gives classified examples to the leaner. The learner performs some generalisation the examples to infer new knowledge. Previous knowledge maybe used to steer the generalisations. In analogy the learner performs the generalisation based on some previously learnt situation. </li></ul>
  4. 4. Types of Learning <ul><li>Learning by OBSERVATION (self-taught) </li></ul><ul><li>this is similar to the category above but without classification by teacher - the learner uses pre-learned information to help classify observation (eg conceptual clustering) </li></ul><ul><li>Learning by DISCOVERY </li></ul><ul><li>this is the highest level of learning covering invention etc and is composed of some of the other types below </li></ul><ul><li>TWO ASPECTS OF LEARNING: </li></ul><ul><li>KNOWLEDGE/SKILL ACQUISITION </li></ul><ul><li>Inputting NEW knowledge </li></ul><ul><li>KNOWLEDGE/SKILL REFINEMENT </li></ul><ul><li>Changing/integrating old knowledge to create better (operational) knowledge (Inputs no or little new knowledge) </li></ul>
  5. 5. Recap: Knowledge Acquisition <ul><li>Knowledge Acquisition is the process of encoding knowledge in a way that intelligent processes can use effectively. </li></ul><ul><li>? Do we always need an AI expert to encode knowledge? </li></ul><ul><li>? Can we get programs to learn – or acquire knowledge for themselves ? </li></ul><ul><li>Next week we will see how GIPO can be used to learn new planning operators </li></ul>
  6. 6. KNOWLEDGE/SKILL REFINEMENT <ul><li>Changing/integrating old knowledge to create better (operational) knowledge (Inputs no or little new knowledge) </li></ul><ul><li>Examples: </li></ul><ul><li>Learning heuristics (improve search) </li></ul><ul><li>Re-representing knowledge (improve search space) </li></ul><ul><li>Learning procedures (remove search altogether!) </li></ul><ul><li>Automatically removing bugs in representations </li></ul>
  7. 7. KNOWLEDGE/SKILL REFINEMENT: MACRO ACQUISITION AND USE FOR AI PLANNING <ul><li>Roughly: a planner solves a problem and induces one or more macros from the solution sequence by “compiling” the operator sequence into one macro. </li></ul><ul><li>Definition: </li></ul><ul><li>WP (weakest precondition) of operator O to achieve goal(s) G </li></ul><ul><li>WP(O,G) = Elements of G that O does not achieve UNION O’s preconditions </li></ul><ul><li>Learning task: </li></ul><ul><li>Find a solution T = (o(1),..,o(N)) to goal G from initial state s(0) </li></ul><ul><li>Form a Macro- Operator (macro) based on: </li></ul><ul><li>Pre-condition: WP (T, G) </li></ul><ul><li>Post-condition: G </li></ul>
  8. 8. Macro acquisition: algorithm <ul><li>Starting with goal G, we regress through the states backwards: </li></ul><ul><li>Assume we have the last operator o(N) applied to s(N-1) to form the final state s(N), with add list o(N).add and precondition o(N).pre </li></ul><ul><li>Then regressing G through o(N) gives </li></ul><ul><li>NewG = WP(o(N),T) = G – o(N).add UNION with o(N).pre </li></ul><ul><li>NewG can now be regressed further until the initial state is reached. The full regressed goal is the weakest precondition that T achieves the goal G. </li></ul>
  9. 9. Macro Use <ul><li>The triple (G, WP(T,G), T) can be stored and retrieved: </li></ul><ul><li>Given a planning problem (S, G1), </li></ul><ul><li>IF S => WP(T,G) and G => G1 then </li></ul><ul><li>achieve G1 by applying sequence T to S. </li></ul><ul><ul><li>The stored‘ macro (G, WP(T,G), T) can be further generalised by changing the constants to variables ranging through all objects of the same sort as the original constants. In OCL for example, as all objects of the same sort share the same behaviour, this generalisation has some justification </li></ul></ul><ul><ul><li>The macro could increase future performance as it may cut out the need to search for a solution to G. </li></ul></ul>
  10. 10. Macro (+Learning) Utility <ul><li>There are other kinds of macro creation: for a solution of size N, N -1 macros can be created as each of the regressed goals can form a macro. This may cause utility problems , however. </li></ul><ul><li>Too many or too general macros may </li></ul><ul><ul><li>Increase the search space </li></ul></ul><ul><ul><li>Increase the time of searching as the planner spends time looking for macros </li></ul></ul>
  11. 11. Conclusion <ul><li>There are various kinds of Learning manifest in nature and in AI </li></ul><ul><li>Two important roles for Learning are in Knowledge Acquisition and Knowledge Refinement </li></ul><ul><li>Macro acquisition is a form of KR where procedures are learned to make plan generation more efficient </li></ul><ul><li>Sometimes, Learned information can degrade performance </li></ul>

×