You are on page 1of 14

KNOWLEDGE

REPRESENTATIONS: REASONING,
ISSUES & ACQUISITION
https://www.linkedin.com/in/neil27/ - NeilHarwani@NirmaUni.ac.in
MBA FT – ML & AI course – February 2021
Simplified videos

Propositional Calculus

Predicate Calculus

Agenda Rules – Subset of predicate logic

Resolution

Symbolic Reasoning under uncertainty

Basic Knowledge Representation Issues


◦ https://www.youtube.com/watch?v=V-O-RFSRe-E -
Knowledge Representation in AI | Semantic Networks |
Artificial Intelligence Tutorial | Edureka
◦ https://www.youtube.com/watch?v=9iN3O_oL2ac -
Knowledge Representation and Reasoning in Artificial
Simplified Intelligence | Logic, Semantic Net, Frames etc (Hindi –
GATE Smashers)
videos ◦ https://www.youtube.com/watch?v=6490tKrGEic -
Propositional Logic in Artificial Intelligence in Hindi |
Knowledge Representation| All Imp Points – GATE
Smashers
◦ https://www.youtube.com/watch?v=UroprmQHTLc -
Predicate Logic 1: Video – MIT OpenCourseWare
Propositional Calculus –
The world consists of
statements
◦ Propositional Calculus – elementary atomic sentences – only 2 values –
True/False
◦ Statements have – Syntax, Semantics, Properties & Inferences
◦ Every Propositional calculus and its truth symbol is a sentence/statement
◦ Valid sentences are WFF ( Well Formatted Formulas)
◦ Propositional Calculus can be denotes in Syntax using he pre- define
symbols
◦ Simple and compound formulas can be used to represent sentences
◦ Parentheses have highest preference followed by other connections ( inset)
◦ Semantics ( meanings) of statements - Truth or False value of a sentence
◦ Designing of a truth table - Hypothesis (P) and Conclusion ( Q)
◦ Expressions are equivalent if they have identical truth tables
◦ Properties of statements – Satisfiable, Contradiction, Validity, Equivalence,
Logical consequence
◦ Valid statements is satisfiable and invalid is contradictory
◦ Equivalence Laws – Idempotency, Associativity, Commutativity, Ditributivity,
De-Morgans Laws, Conditional elimination, Biconditional Elimination
◦ Inference Rules – Means for logical proofs and deductions
◦ Symbols of predicate calculus –
◦ Truth Symbols – Truth /False
◦ Constant Symbols
◦ Variable Symbols
◦ Function Symbols

Predicate ◦ Semantics of predicate calculus – Friends (


Gem, John)

Calculus – ◦ Terms - constant or variable functional


expressions
◦ Atomic Sentences – Constructed using
The world consists of terms and predicate symbols
objects, functions & ◦ Complex Sentences – logical connectives
relationships to construct complex sentences
◦ Statements have Interpretations
◦ Quantifiers – Universal, Existential,
Uniqueness,
Rule Based Knowledge system benefits
◦ Majority of expert systems employ rule
basis
◦ Less Expensive
◦ Do not require costly hardware and
Rule Based minimal expenditure in training
Knowledge ◦ Widespread availability – Knowledge
engineers focus on KB ( critical for
Representation developing Expert system)
(In Expert ◦ Easy to learn how to learn rule
development – faster learning curve
Systems) ◦ Rule bases can be relatively modified
◦ Validation and Verification is a relatively
simple process
Rules –
Subset of predicate
logic
Types of Rules
◦ Production rules – IF- Then types
◦ Attribute – Value Pair properties – Name, Type, prompt, legal values, specified values,
◦ Clause Properties – Premise and conclusion clauses
◦ Rule Properties – Nomenclature, Premise, Intermediate and final conclusions, Notes and
formal references us, associated, Priority, status, grouping and symbols
Predicate calculus uses inference rules – Infer a new set of correct expressions from true
assertions
Knowledge Representation using Rules - IF – THEN ( LHS- RHS)
◦ Forward and Backward Chaining
◦ Forward chaining – data driven - Start with sentences and draw conclusions
◦ Backward Chaining – Goal driven - Start with what we want to prove and find
implication sentences
◦ Algorithms manage forward & Backward chaining in rules engines
◦ Combination of forward and Backward reasoning is an efficient mechanism of solving
problems

Simplest rule based production systems as in diagram


◦ Resolution is a complete resolution procedure – it gives a deduction
◦ Statements need to be converted to FOL ( First Order Logic) for
computer to resolve
◦ Parent cluses are compared to yield the inference
◦ Strategies of resolutions
◦ Unit Preference – inferences that produce short sentences
◦ Set of support - cuts search space dramatically. Goal directed. Easier
for Humans to understand

Resolution ◦ Algos for propositional resolution –


◦ Resolution Inference rule
◦ Generalized Resolution – Implication
◦ Generalized Resolution – disjunctions
◦ Conversion to Normal format
◦ Conflict Resolution – assigning some preferences of rules to be assigned
within the matching process
◦ Preference based on rules
◦ Dealing with equality
◦ Non-Monotonic Reasoning
◦ First order predicate logic is called monotonic logic
◦ Conclusions from this logic are valid deductions
◦ Adding new axioms increases the knowledge

Symbolic ◦ First order KB systems grow monotonically till facts invalidate the old
knowledge
◦ Knowledge does not remain valid in a forever changing environments

Reasoning ◦ Truth Maintenance Systems


◦ Companions to inference systems

under
◦ Main job is to maintain consistency and not to perform any inference
◦ User can forget about consistency and focus on problem solving
◦ Maintain complete record for reasons and justification of beliefs

uncertainty ◦ Dependency directed backtracking


◦ Two types of justification records - support Lists ( SL), Conceptual
Dependencies ( CPs)
◦ Sources of uncertainty
◦ Uncertainty w.r.t validity of KB rule
◦ Uncertainty w.r.t validity of user response
◦ Probability & Baye’sTheorem

Symbolic ◦ Bayes theorem of conditional probability is used to model


behaviour of a system

Reasoning
◦ In a complex world Bayes theorem has limitations as it spins
too many joint probabilities
◦ Too many probabilities to be provides

under ◦ Very large space required to store all probabilities


◦ Too much time needed to compute probabilities

uncertainty ◦ Despite these probabilities Bayesian probability is used


◦ Uncertainty is still tackled through Baye’s rule when we have
good probability estimate
◦ Uncertainty Through Fuzzy Sets
◦ Tackling when source of uncertainty is consequence of user
inputs
◦ Knowledge can be declarative or procedural
◦ Knowledge in Expert system consists of
◦ A priori Knowledge – facts and rules are already known
◦ Inferred knowledge – facts and rules are derived upon consultation
with expert system
Basic ◦ Considerations in knowledge representation ( KB of expert system)
◦ Format compatible with computer
Knowledge ◦ Close correspondence with actual facts and rules
◦ Representation easily addresses, retrieved, modified and updated
Representation ◦ OAV triplets – each OAV concerned with specific entity or object

Issues ◦ Sematic networks –


◦ Network with multiple OAV triplets
◦ Several attributes for each object
◦ Highly visual and intuitive in nature
◦ Design of Semantic Networks
◦ Manipulation of Semantic Networks
Frames – data structures to represent mental models
◦ Frame Based Representation Languages – LISP
Basic ◦ Implementation of Frame Structures – property or
association lists
Knowledge Conceptual Graphs
Representation ◦ Relations between concepts

Issues ◦ Types, Individuals and Names


◦ Generalization and Specialization
◦ Propositional Nodes
◦ Conceptual graphs & Logic
◦ Domain expert – guidelines for knowledge
◦ Domain Selection – appropriate and distinct
advantage over alternate methods
◦ Selection of knowledge engineers
◦ Expert Selection
◦ Initial Meetings
Knowledge ◦ Follow-on meetings

Acquisition ◦ Knowledge engineer may be used as a domain


expert
◦ Inference Engines
◦ Essential part of expert system
◦ Major factor in defining the efficacy of the
model
THANK YOU
https://www.linkedin.com/in/neil27/ - NeilHarwani@NirmaUni.ac.in

You might also like