Date post: | 01-Jan-2016 |
Category: |
Documents |
Upload: | velma-cardenas |
View: | 17 times |
Download: | 0 times |
Proof System
HY-566
Proof layer
Next layer of SW is logic and proof layers. – allow the user to state any logical principles,– computer can to infer new knowledge by applying
these principles on the existing data.
This project implements a defeasible reasoning system which presents explanations to users for the answers to their queries.
Defeasible Logic
A defeasible theory is a triple (F, R, >)– F is a set of literals (called facts)– R a finite set of rules– > a superiority relation on R
Two kind of rules– Strict rules (A→P) can’t be defeated– Defeasible rules (A=>P) can be defeated
Defeasible Logic Metaprogram
Simulates the proof theory of defeasible logic contains clauses :– denite provability– defeasible provability– When rule is blocked/unblocked – When rule is defeated/undefeateds
Explanation Example
An e-shop tell to Bob’s that he owns 30$ Explanation (facts and rules):
– purchase(Bob,DVD) - fact
– price(DVD,30) - fact– delivered(DVD,Bob) - fact– purchase(Bob,DVD), price(DVD,30),
delivered(DVD,Bob) → owes(Bob,30) - rule
Extension of RULML
RuleML is an XML based language that supports rule representation for the Semantic Web.
A new XML schema, extension of RuleML, is proposed for explanation representation in defeasible logic
Ext. RULML - Atom, Fact
Atom: operator, var or const, optionaly NOT <Atom>
<Not>
<Op> rich </Op>
<Ind> Bob </Ind>
</Not>
</Atom>
A Fact is consisted by an Atom that comprise a certain knowledge
Ext. RULML - Rules
Strict - Defeasible Rules. – Head: is an Atom – Body: number of Atoms
<Defeasible_rule Label="r1"> <Head> <Atom> <Op> rich </Op> <Ind> Bob </Ind> </Atom> </Head>
<Body> <Atom> <Op> wins_lotto </Op> <Ind> Bob </Ind> </Atom> </Body></Defeasible_rule>
Ext. RULML - Explanations
Definitely Provable Explanations– Denote the Atom– Definite Proof
Definite Proof– Fact for that Atom– Strict Rule with Head the Atom and Body
(multiple) atoms that must be proved definitely.
Explanations Example
<Definitely_provable><Atom> <Op> rich </Op> <Ind> Bob </Ind></Atom><Definite_Proof> <Strict_rule Label="r1"> <Head>
<Atom> <Op> rich </Op> <Ind> Bob </Ind>
</Atom> </Head> <Body> <Atom> <Op> wins_lotto </Op>
<Ind> Bob </Ind> </Atom> </Body> </Strict_rule>
<Definitely_provable> <Definite_Proof> <Fact>
<Atom> <Op> wins_lotto
</Op> <Ind> Bob </Ind>
</Atom> </Fact> </Definite_Proof> </Definitely_provable></Definite_Proof>
</Definitely_provable>
Proof tree construction (1/3)
The foundation of the proof system – Prolog metaprogram implements rules of Defeasible Logic– The trace of the XSB implementation of prolog
XSB: logic programming engine used to run the metaprogram.
To communicate with the XSB the invocation of the XSB executable was used (Javas exec method)
– Send commands to the XSB interpreter – Receive the output that was produced as an effect.
Load the metaprogram and the defeasible theory
Proof tree construction (2/3)
Load the metaprogram and the defeasible theory
At the evaluation of a query XSB will print a message each time a predicate is:– Initially entered (Call)– Successfully returned from (Exit),– Failed back into (Redo), and– Completely failed out of (Fail).
Proof tree construction (3/3)
A tree whose nodes are traced predicates is constructed by the Java XSB invoker when trace is parsed.
Each node has information– A string for the predicates name– The predicates arguments– Whether it was found to be true (Exit) or false
(Fail)– Whether it was failed back into (Redo)– Boolean attribute tells if predicate is negated.
Why the tree needs pruning?
XSB trace contains data not needed for proof – A metaprogram to translate the DL into logic
programming is used. Additional clauses are needed which add information to trace
– Prolog shows successful and unsuccessful paths
The tree produced by the XSB trace is built according to the metaprogram structure but the final tree needs to be compliant with the XML schema
Pruning Rules
Heuristic rules are used in order to prune the proof tree
According to the truth value and the type of the root node we may want to maintain– only successful paths– only failed paths – combinations of them.
Pruning Motivation Example1
Suppose we have the following defeasibly theory translated in logic programming as:– fact(a).– fact(e).– defeasible(r1,b,a).– defeasible(r2,b,e).– defeasible(r3,~(b),d).
Example1: Defeasible provability of b
Pruned
Example1: Defeasible provability of b
We are interested in successful (True) paths The pruning algorithm removes
– the subtree with the false goal to prove that b is denitely provable
– the false predicate to find a strict supportive rule for b
– the metaprogram additional negation clause.
Pruning Motivation Example2
Suppose we have the following defeasibly theory translated in logic programming as:– fact(a).– defeasible(r1,b,a).– defeasible(r2,~(b),a).
Example2: Defeasible provability of b
We are interested in unsuccessful paths and the pruning algorithm keeps the initial proof tree.
Pruning resume
The proof tree after using the heuristic techniques is similar to an explanation derived by the use of pure DL
Drawback: heuristics are fully dependent on the metaprogram. – Changes at metaprogram => changes at pruning
implementation.
Pruning Example
The marked rule is pruned
Pruning Example 2
The marked rule is pruned
Agent Interface to the Proof System
The system makes use of two kinds of agents– 'Agent' which issues queries– 'Main Agent' which is responsible to answer the
queries. Both agents are based on JADE (Java Agent
DEvelopment Framework) – a software framework to develop agent-based
applications
Process to answer a query (1/3)
MainAgent
XSB
XML Writter
Invoker
Pruner
Agent
2) Predicate 1)Question3) Predicate
4) result trace
5) result tree
7) Pruned result6) result tree
8) Pruned result 9) XML proof
10) answer or proof
Process to answer a query (2/3)
1. An agent issues a query to the Main Agent.- predicate::(proof|answer)
2. Main Agent sends Predicate to the Invoker– Invoker is responsible to communicate with XSB
3. Invoker executes the Predicate.
4. XSB returns the full result trace.
5. Invoker returns result tree to Main Agent.
6. Main Agent sends result tree to the Pruner
Process to answer a query (3/3)
7. Pruner returns pruned result to Main Agent.8. Main Agent sends the pruned result to the
XML writer (only if proof requested)9. XML writer returns the XML Proof.10. Main Agent returns Answer or Proof
- ANSWER(true | false)- PROOF:(proof string)- ERROR:(error message)
Visual Agent
Command Line Agent
Reads in a random way the questions from a configuration file
Sends the question to the Main Agent with the order read
The format of the questions is of the form – predicate::(proof|answer)
The answers and proofs are stored at a files