Rule Based Systems - Science topic

In computer science, a rule-based system is used to store and manipulate knowledge to interpret information in a useful way. It is often used in artificial intelligence applications and research. Normally, the term rule-based system is applied to systems involving human-crafted or curated rule sets.

If your question is when should a rule (or practice) that seems to work (i.e. produce the desired result in terms of human behaviour) in country A be adopted in country B, read Nancy Cartwright’s and Jeremy Hardie’s short book on ‘Evidence-based Policy’. Even the first few pages will illuminate the issue.

I suggest you follow the functions in this attached link : https://au.mathworks.com/help/fuzzy/referencelist.html?type=function&category=simulation&s_tid=CRUX_topnav also https://au.mathworks.com/help/fuzzy/building-systems-with-fuzzy-logic-toolbox-software.html

https://au.mathworks.com/matlabcentral/answers/74045-in-the-rule-viewer-in-the-fuzzy-toolbox-how-to-to-insert-many-inputs-at-one-time-and-get-the-output

Thank you Salah Mortada for your explanation. However, I would like to draw a figure that compares different methods that used in the experiment result and clearly show the overfitting, under-fitting methods. If you have an example of such an experimental result will be grateful.

The information gain is the amount of information gained about a random variable or signal from observing another random variable.

Entropy is the average rate at which information is produced by a stochastic source of data, Or, it is a measure of the uncertainty associated with a random variable.

An example at:

https://medium.com/coinmonks/what-is-entropy-and-why-information-gain-is-matter-4e85d46d2f01

By using weights that represents the relative importance of each rule

Sorry, your question need to be simplified more to be understood...

Dear Demilade . . .

Adapting/adjusting the controller parameters for such cases (where plant parameters were not accurately known) is called adaptive control. so as it can be infer, you should have an acceptable approximation of your dynamic system to tune the adaptive control parameters.

recently i have read a paper about this subject which i want to suggest it to you to read it. its about how to design the law for your MRAC.

it might be helpful.

best regards

As far as I know, it is not necessary to have all the inputs and outputs in a rule. For example, in your system a rule can have only 2 inputs which are along with at least one output. This can be applied in the Matlab fuzzy logic toolbox as well.

Hope it could help you.

Thanks Dr. Azedine Boulmakoul for sharing your research paper. We have seen that and in fact willing to write you to understand the inferencing used in that article. Lot of things you have not disclosed in that paper.

Hello Nabila,

First of all, the data set has too much less data points. Hardly any good classifier may work well on this. So, my suggestion is to increase more observations for better classification accuracy.

Thanks,

Sobhan

In my opinion GA and NN are similar in term of CPU time. Both take a lot of resources :-)

But currently it is not my subject of interest so maybe new NN algorithms are much better. If you interested, see my old papers:

See the following link https://googleweblight.com/i?u=https://www.linkit.nl/knowledge-base/228/Natural_Language_Processing_and_Machine_Learning_the_core_of_the_modern_smart_chatbot&hl=en-IN&tg=133&tk=12774569711187705643

Minimum Spanning Trees (MSTs) may be of interest to address overlapped clustering problems:

MSTs may be used in a fuzzy context. Refer to:

Gaussian Mixing Models (GMMs) can then be used for deriving a measure of component overlap. Refer to :

Finally, an overview of existing methods is available from:

While most academics would suggest Prolog I would say that it depends on the IT infrastructure that you are going to deploy such system and the requirement for coupling with existing systems.

I would also recommend that if you are going to do program such a system for large scale production you should optimize as much as possible.A simlple suggestion is to use hashes and use fast string matching to reduce lookup time(some Prolog implementations use these but you must check to see if the selected Prolog engine does so).

A good article about interpretability indices:

Cannone, R., Alonso, J. M., & Magdalena, L. (2011). An empirical study on interpretability indexes through multi-objective evolutionary algorithms. In Fuzzy Logic and Applications (pp. 131-138). Springer Berlin Heidelberg.

Many years ago, Tony Arrott showed me an optimization program he had written using a values-based, as opposed to a rules-based algorithm.  It solved the Travelling Salesman problem (with almost perfect optimization) many times faster and was also able to solve the Inventory problem (turning production facilities on an off optimally in response to fluctuations in demand) in real time on an old Mac. 

He explained that the strategy of "scoring" each solution according to an array of values (such as "Having the item in stock is good," and "Shutting down or restarting a factory is bad.") was a lot easier and faster than trying to obey rules about which action should follow which. 

I'm pretty sure this is how our brains work.  It helps to explain our societal obsession with single-valued logic (e.g., "No price is too great to save a single human life!"or "The only good X is a dead X!" ).  As stupid as it may be, it's a very easy way of reaching a decision.  The danger in overutilizing this method is obvious. 

(Caveat: again, I have no credentials to speak of such matters.  :-)

As Reisel González Pérez  has mentioned, Pyke is the packages which give you the ability of developing rule based systems.

Besides the given links above you can see some Pyke examples here : http://cyberdelia.github.io/pricing/ 

Hi Saba et al.,

I am very sure that you are aware of it, but hopefully I can point you in a few directions that you might not be familiar with (at the same time as promoting the technology to others who have not heard of it).

Certainly the largest following for knowledge representation "languages" is the Resource Description Framework (RDF), which is the model that powers both the Semantic Web and the Linked Data web. It's a W3C standard, which is a very well respected consortium. It has plenty of ontologies ready for use in different domains. It is also enriched by a large web of open data in the form of the linked open data (LOD) cloud. SPARQL is the most well used query language, that acts a little bit like SQL does to a relational database. The power of RDF really comes when you pair it with cross-domain hyperlinking, and that's where the Linked Open Data cloud comes in.

I'd very highly recommend everyone looking into it. Here are some links....

* Linked Open Data cloud: http://lod-cloud.net/

* The LOD2 project which promoted LOD to public and private organisations in Europe, and provided them with consultancy, tutorials and a bespoke stack of technologies: http://lod2.eu/

* SKOS (Simple Knowledge Organisation System), an RDF ontology for making frame-like data: http://www.w3.org/2004/02/skos/

* Rule Interchange Format (RIF): http://www.w3.org/2005/rules/wiki/Primer

* TRiG is very important for serialisation of Named Entities and RDF datasets: http://wifo5-03.informatik.uni-mannheim.de/bizer/trig/

* I've used TRiG quite a while ago with XLWrap which transforms Excel spreadsheets into RDF: http://xlwrap.sourceforge.net/

I hope that all helps, and is interesting!

Daniel

The largest and most comprehensive hierarchical vocabulary of medical terms is SNOMED Clinical Terms (SNOMED CT). It is more a knowledge taxonomy than an ontology, but it is currently the best knowledge base of general medical terms there is. 

References:

There are several other medical ontologies under development but they are significantly less expressive (i.e. smaller) than SNOMED. Some are academic and open/free, some are commercial. Additionally, a great deal of these projects is unfinished and the work on them stopped a while back.

My advice would be to first take a look at SNOMED, OGMS, GALEN and - by all means - GLIF.

Kind regards,

Marko

Dear Saba Ghashghaee madam first read the book on Artificial Intelligence by D W Patterson and another by Rich and Knight. you will understand the basics of frames and rule based system. then you will be able to implement them in any tool.

All the best.

Ok. Merging rules can combine either conditions or results.  Combining results would reduce the rule set, e.g., A*B->C and A*D->C becoming A*(B+D)->C, if your allow or-function as well as and conditions.  BUT, combining conditions creates a new rule and the original rules maybe less restrictive, e.g., A->C, B->C becomes A+B->C.  If you can't use or on the left, I'd have to think more about what's possible. . .

You may want to take a look at pruning algorithms, e.g. used in RIPPER. These algorithms reduce rule complexity (similar to penalization in regression analysis)

Hi Ivy,

Depending on what you want to do, there are two solutions:

(1) If you just want to syntactically process RuleML rule bases from within a Java program, then you can simply use a Java XML parser (I personally prefer DOM4J).

(2) If you want to use a RuleML-compliant rule engine, then I would take a look at the links posted by Nacéra - but note that not all of these rule engines can be integrated in custom applications.

Hope this helps,

- Stratos

@Ahmad : Crisp rules from neural networks can be extracted. There has been many papers on that topic in the last 20 years. Just write on Google Scholar "Rule Extraction" and you will obtain a long list. You can also have a look to my papers on that topic in my ResearchGate publications' page.