Current Topic: 7.2.1. Knowledge representation, reasoning and knowledge engineering concepts
You have a privilege to create a quiz (QnA) related to this subject and obtain creativity score...
Knowledge representation, reasoning and knowledge engineering concepts
For a long time, Artificial Intelligence (AI) lived on the bottom of the lake of opportunities. Recent years turned the lake into the ocean and the underwater current brought AI back to the surface. Nothing else is growing so quickly with the demand for new skills and talents. [1]
Google, Tesla, Apple, Facebook and Microsoft compete for knowledge engineers capable to develop intelligent systems, such as self-driven cars, drones, and health and business analytics.
Can we define “knowledge” and distinguish it from data and information?
We will try, although any definition in this area depends on the viewer’s perspectives.
We can consider row data, such as numbers, information as organized data, such as a database, and knowledge as well organized information, which allows acting upon, making decisions, etc.
How information is represented?
There are two major parts to electronically stored information: structured data and unstructured data.
Structured data are very formal. They are defined in terms so precise that even computers can easily understand them.
Databases, Business and Data Models, Services and XML Files are structured and understood by computers.
Unstructured data are documents and communications artifacts, like taped messages and video clips that make sense to people. The knowledge captured in unstructured data is generally unintelligible to computer systems. We’d like to change this. We’d like to use this knowledge to make computers a bit smarter.
Rules and Semantic technologies help us to better organize unstructured information and create a conceptual model (ontology) of the knowledge that is represented there.
The majority of the information used daily in the routine of a business has never been captured.
It is so-called Tribal Knowledge. My conservative estimate of the ratio between structured, unstructured and “tribal” knowledge is 10%, 20% and 70%. [2]
Knowledge representation is based today on Semantic Web Standards, the Resource Description Framework (RDF) and Web Ontology Language (OWL). Both are XML family languages.
Related fields and concepts
Integrated Software and Knowledge Engineering is a relatively new area, which applies Knowledge Engineering concepts to create intelligent software applications [1, 2, and 3] and provide conversational semantic decision support [2, 4].
Much more ambitious is Cognitive Computing, which involves Machine Learning in the attempt to simulate human thinking processes.
Machine Learning started from pattern recognition and became a growing field of computer science.
In this section we will focus on the very first step in this direction: moving from Java code to business rules. We are not going to replace our programs with the rules yet. But following the Delegation Design Pattern, we recognize code conditions, which are most often subjects to business changes.
Then we delegate these blocks of code or in other words separate them from other code pieces.
Transforming often changeable code into rules allows business owners to better understand and more actively participate in the development.
A Rules Engine is a container to hold the rules, make them conveniently visible and manageable.
JBoss BRMS is a good example of such Rules Engine.
Was it clear so far?
The mechanism of walking over the rules to make a conclusion based on rules conditions is called Reasoning.
Unfortunately, Knowledge Engineering is rarely a subject in current schools and even Critical Thinking is not always present in the curriculum.
Let me introduce several basic concepts related to knowledge, logic and thinking processes.
Inference is similar to Reasoning as a word often describing a thinking process for computers.
Inference Engine is a computer component, often in a Rules Engine, responsible for inferring process.
While Reasoning is considered as a formal logic based on rules and facts, Inference is defined more loosely with the possibility of unproven assumptions.
Two main directions, Deduction and Induction are applicable to the thinking process.
Deduction is specialization, when we take a general idea and apply to a specific case.
For example, an accountant takes tax law and regulations and applies to a specific tax return of a specific business to conclude (or deduct) that the company violated the law.
Induction is generalization, when we take several specific cases to generalize into a common idea. For example, a programmer considers several tasks related to geometric figures and decided to create a generic parent class, Shape.
Reasoning Systems are responsible for reasoning by computers. There are two major ways of reasoning:
- Forward Chaining, when facts (run-time data) presented with the rules (conditions) are driving the conclusion.
- Backward Chaining has exactly opposite direction, which starts with a goal or a conclusion business wants to achieve, then is looking for steps, or sub-goals to achieve this parent-goal.
And, of course, there is Hybrid Reasoning System (HRS), which includes both capabilities.
JBoss Drools uses Rete enhanced algorithm, which originally started as a Forward Chaining, but currently represents a Hybrid Reasoning System.
Reasoning process usually takes multiple steps. In the process of describing the steps a useful mechanism is describing the steps as a sequence of states. This mechanism is called State Machine.
Each State in State Machine can have:
- A set of possible input events
- A set of resulting actions or output events
- A set of new states that may result from the input
This sequence of states is usually represented with a State Machine Diagram, similar to a simple example below.
->Inputs -> State 1 Name -> output / inputs to State 2 -> State 2 Name -> output -> Resulting State
2016-07-13_10:59 by Christian Nasr
I would be concerned to give IV insulin push if I do not know what the serum potassium is.
2016-07-14_23:36 by Irl Hirsch
It is done frequently, but as noted can be dangerous, particularly in cardiac patients. I tend to discourage its use-it just takes one law suit to change one's mind, and many years ago I was an "expert witness" for someone given IV insulin prior to surgery and had a subsequent cardiac arrest. The patient was most likely hypokalemic and on digoxin (this was many years ago) but reading or seeing that one time is all it takes to change one's practice.
2016-07-16_12:43 by Ketan Dhatariya
In the UK we have a national guideline - free to download at http://www.diabetologists-abcd.org.uk/JBDS/Surgical_guideline_2015_summary_FINAL_amended_Mar_2016.pdf (all of our UK inpatient guidelines are free to download from http://www.diabetologists-abcd.org.uk/jbds/jbds.htm)
The surgical guideline says the following about using short acting insulin to correct pre and post operative hyperglycaemia. I am aware of some local audit data to show that this approach has led to far fewer procedures being cancelled / postponed.
Pre-operative hyperglycaemia: (blood glucose greater than 12mmol/L (216mg/dl) with blood ketones less than 3mmol/L or urine ketones less than +++)
Type 1 diabetes: give subcutaneous rapid acting analogue insulin (i.e. Novorapid®, Humalog® or Apidra®). Assume that 1 unit will drop the blood glucose by 3mmol/L. Recheck blood glucose 1 hour later to ensure it is falling. If surgery cannot be delayed commence a VRIII.
Type 2 diabetes: give 0.1 units/kg of subcutaneous rapid acting analogue insulin, and recheck blood glucose 1 hour later to ensure it is falling. If surgery cannot be delayed or the response is inadequate, commence a VRIII (variable rate intravenous insulin infusion).
Post-operative hyperglycaemia: (blood glucose greater than 12mmol/L with blood ketones less than 3mmol/L or urine ketones less than +++)
Type 1 diabetes: give subcutaneous rapid acting analogue insulin. Assume that 1 unit will drop blood glucose by 3mmol/L BUT wherever possible take advice from the patient about the amount of insulin
Type 2 diabetes: give 0.1 units/kg of subcutaneous rapid acting analogue insulin, and recheck blood glucose 1 hour normally required to correct a high blood glucose. Recheck the blood glucose 1 hour later later to ensure it is falling. Repeat the subcutaneous insulin after 2 hours if the blood glucose is still above12mmol/L. In this situation the insulin dose selected should take into account the response to the initial dose – consider doubling the dose if the response is inadequate. Repeat the blood glucose after another hour. If it is not falling consider introducing VRIII to ensure it is falling. Repeat the subcutaneous insulin dose after 2 hours if the blood glucose is still above 12mmol/L. In this situation the insulin dose selected should take into account the response to the initial dose – consider increasing the dose if the response is inadequate. Recheck the blood glucose after 1 hour. If it is not falling consider introducing VRIII.
