Please enable JavaScript.
Coggle requires JavaScript to display documents.
Rule-based systems - Coggle Diagram
Rule-based systems
How do we learn
Documented
Extract manually
Machine Learning
Information Gain Formula IRS-MR-Day3-slide78
IRS-RS-Day4
See Knowledge discovery flow chart
Undocumented
Interviews
Knowledge models
IRS-MR-Day 1-Slide 14 to Slide 30
Slide 30 is Design process for knowledge models
How do we reason (IRS-MR-Day1)
Task types
Analytic tasks: Task with template for solution
Synthetic task: Task without template for solution
Search and Optimisation problems (see Search techniques flow chart)
Reasoning
Deductive Reasoning types (Universal > Individual)
Fuzzy Logic, Rule Engine,
Inductive Reasoning (Individual > Universal)
Reinforcement Learning, Decision Tree, Neural Networks, Association Rule and Bayesian Net
Analogical Reasoning (Known > Inferred)
Abductive (Observations > Likelihood)
Mapping how we learn and reason into the machine
The Machine
Conflict Set: When a rule match a data
Conflict Resolution: When multiple rules match a data
Rule Engine: Maps to our brain, which decides which rule should be fired first
Rule Firing Priority
IRS-MR-Day 3-Slide 9
Knowledge Base: map to our reasoning, holds all the rules
Working Memory: Maps to our past knowledge, holds all the data
Chaining
Forward Chaining
Backward Chaining: not Hypothesis ^ Knowledge Base = {}
Not cat and KB pointed to the fact is is a cat is empty set, means cat hypothesis is true
If working memory is huge, backward chaining will require less computation
When we are not sure
How confident we are: Certainty Factors
-1 False, 0 Not sure and +1 True
Mapping it to the machine
Multiple rules, add up all their net certainty factors.
For each rule, their net certainty factor = CF(Fact) * CF(Rule)
If Rule consists of Rule 1 and Rule 2 then use min CF of the two rules to multiply by CF(Fact)
If Rule consists of Rule 1 or Rule 2, take the max CF of the two rules to multiply by CF(Fact)
IF multiple rules go to the same conclusion, use IRS-MR-Day3-Slide 49
How much it belongs: Fuzzy Logic
IRS-RS-Day3-slide 61
If there are two factors for the fuzzy logic, AND takes min, OR takes max
Examples of such systems
IRS-MR-Day 4
Adding our rule based system to other systems
Hybrid Machine flow chart
Cognitive systems: Systems that can exhibit human like intelligence like learning/reasoning/memory