Friday, September 7, 2012

Backward Chaining


A couple of years ago I blogged on this topic but I see that it still is running rampant across the internet. So, I thinks to myself, thinks I, one more time into the fray.

Backward Chaining (BC) Definitions

Level One BC: Goal Oriented
First Condition Element (CE) contains a goal that, basically, changes the focus of all of that particular rule set.  Until that goal is satisfied then, or no more rules are true that have that Goal as its first CE, then no other Goal can be considered.  Goals are usually inserted in reverse order of order of desired firing because most rulebased systems use a LIFO stack such that the last goal on the stack will be the first one to be considered.

Level Two BC-OoD: Backward Chaining Object On Demand
  • When Needed: This rule is not evaluated until there is a need for the status of this object that is being evaluated in another rule.
  • When Changed: This rule would fire ONLY when the value of the data element changed.
Level Two was/is used extensively in ND Advisor (later FICO Blaze Advisor) as part of the Nexpert additions to Advisor.

Level Three FOBC: Full-Opportunistic Backward Chaining
Condition Elements have the following constraints
  • True
  • False
  • Not Known (NK) – the fact is not known and there has not been an attempt to evaluate
  • Unknown (UK) – the fact is not known.  There has been at least one attempt to evaluate the True or False condition but it has failed.
Level Four BC:  
Level Three BC can be extended such that there could be either a first failure for AND statements as well as a first success in an OR statement or there could be an exhaustive evaluation of that rule or all rules regardless of the nature a predetermined outcome.  This is normally only done when there is a method (function) in one of the condition elements of one or more rules.  This was done in the Neuron Data Nexpert tool and copied in the early versions of Advisor.  The reasoning being that even though the rule did not fire you still wanted the function/method to be evaluated and possibly a value stored somewhere.  Really bad programming that was, nevertheless, used by some major clients.  Really, really bad programming.

Fuzzy Logic (FL) in a rulebase.
The probability of this outcome being true is xx% if all of the CEs are True.

Extended Fuzzy Logic (XFL) – sometimes called the MYCIN Approach
If the CEs all evaluate to True, then there is a certain confidence interval (CI) that the rule is true as well as a certain Doubt Interval (DI) that the rule is still not true because, in the opinion of the expert(s) this rule cannot be 100% verified.  For example, if the CI == 60% True and 20% false then there remains a 20% Space of Doubt (SD) that cannot be accounted for by ordinary means.  The complete description of this is contained in the seminal book on MYCN that dealt with this problem.

In the Level Three FOBC approach, as one leaf of the CE is evaluated and a condition is NK then the elements are evaluated to determine whether it is true or false or UK.  As each leaf is traversed and is marked if a leaf is found False then the rule is no longer evaluated.  If the rule is found UK then the rule is no longer evaluated.  If the CE if found True, the the next CE if evaluated until all of the CEs are evaluated and the rule can be determined to be True or False.

Any questions, comments or corrections would be appreciated.


Tuesday, September 4, 2012

Three Things In Life


My father told me that there are three things to consider in life.  Only three things that you should remember.  No, four things.  Someday you will die so you have to consider: (1) When you will die,  (2) where you will die and (3) how you will die.  You can't control any of these.  Not normally.  Not unless you commit suicide.  Not the normal way out.  The fourth you can control.  What happens between the time you are born and the time that you die.  Or, between the time that you read this blog and the time that you die.

If we knew the time, place and the circumstances of our death, we probably would lead a completely different life.  Most would live a life of abandon until the final year when they realized the futility of what they had done.  Some would actually live a life worth having been lived.  Me?  I have no idea.  I'm just glad that I don't know.  I actually never thought that I would live to be over forty or fifty and really didn't plan to be this old.  It's nice, but I didn't plan on it.  G-d has been nice to me but I've been ready for many years now.  Only recently did I realize how totally unprepared that I was actually to die and move on so, I thank Him that he let me live this long.

And you?  Well, what are you doing to prepare for your final destination?  What kind of memory will you leave behind?  After all, most of us will leave some kind of legacy behind for some kind of brief period.  Some longer than others.  Abraham left quite a long legacy.  Beethoven left a long legacy.  Will I leave anything like George Washington or Thomas Jefferson?  Not bloody likely.  Mine will probably last 30 years - 40 at most.  Just about as long as my youngest son.  See, I don't have any grandchildren to continue my legacy so it won't last much longer.  Not my fault, it just is.

Again, how about you?  Will your writings last beyond your lifetime?  How about your work?  Your grandchildren?  Great-grandchildren?  On the other hand, does it really matter?  Does the fourth question really matter after all?  Probably not.  But I like to think that, like Steve Jobs, that I did make a small dent in the universe.  Just not so big a dent as he made.  I didn't help invent the iPhone, the MacIntosh, the iPad nor any of the other cool Apple products.

But, along with Mark Procter of Drools (and Greg Barton and Rolando Hernandez of Biz Rules) I did get October Rules Fest started back in 2008 in Dallas, Texas.  (Why is another story.)  We did it again in 2009.  Then in 2010 I turned it over to a friend of mine and it moved to California.  Then it morphed into IntelliFest 2012 and that became the greatest AI event in the world so far.  It features a reunion of the infamous CMU Gang of Three (so far).  So, if you don't do anything else, help make my small legacy the greatest event the world had seen this year.  (This is completely shameless!!)  Come out to San Diego this October and pay us a visit.  Hang out with Forgy, Miranker, Grossman, Riley, Berlioz-Matignon, Ceranno-Morales, Feldman, Saint Marie, Laun, Lenat, Raden, Vincent, Fry and all the rest of the AI, medical, and business greats.  Three Days Of Total AI Glory!!!  Number FIVE is ALIVE !!!