NIPS*96 Rule Extraction W'shop

Robert Andrews robert at fit.qut.edu.au
Wed Aug 21 01:42:18 EDT 1996






=============================================================
                   FIRST CALL FOR PAPERS

               NIPS*96 POST-CONFERENCE WORKSHOP 
        --------------------------------------------
        RULE-EXTRACTION FROM TRAINED NEURAL NETWORKS
        --------------------------------------------
               Snowmass (Aspen), Colorado, USA
                   Fri December 6th, 1996 

             Robert Andrews & Joachim Diederich
               Neurocomputing Research Centre
            Queensland University of Technology
            Brisbane 4001 Queensland, Australia
                  Fax:   +61 7 864-1801
               E-mail: robert at fit.qut.edu.au
               E-mail: joachim at fit.qut.edu.au



Rule extraction  can be  defined as the  process  of deriving  a symbolic 
description of  a trained  Artificial  Neural Network (ANN).  Ideally the 
rule extraction process results in a symbolic  description  which closely 
mimics the behaviour of the network in a concise and comprehensible form. 

The   merits  of  including  rule  extraction techniques as an adjunct to 
conventional  Artificial  Neural Network techniques include:
  a)	the provision of a 'User Explanation' capability;
  b)	improvement  of  the  generalisation  capabilities of ANN
	solutions by allowing  identification of regions of input
	space not adequately represented;
  c)	data exploration and the  induction of scientific theories
	by the discovery and  explicitation of  previously unknown
	dependencies and relationships in data sets;
  d)	knowledge acquistion for symbolic AI systems by overcoming
	the knowledge engineering bottleneck;
  e)	the potential to contribute  to the  understanding  of how
	symbolic  and   connectionist  approaches  to  AI  can  be 
	profitably integrated.

An ancillary problem to that of rule extraction from trained ANNs is that 
of using the  ANN for  the `refinement' of existing rules within symbolic 
knowledge bases. The goal  in rule  refinement is to use a combination of 
ANN learning and rule extraction techniques to produce a `better'  (ie  a 
`refined')  set  of symbolic rules which can  then be applied back in the 
original problem domain. In the rule refinement process, the initial rule 
base (ie what may be termed `prior knowledge') is inserted into an ANN by 
programming  some  of  the  weights.  The rule  refinement  process  then 
proceeds  in the same way as normal rule extraction  viz  (1)  train  the 
network on the  available  data set(s); and (2) extract (in this case the 
`refined') rules - with the proviso  that  the  rule  refinement  process 
may involve a  number of iterations of the training  phase  rather than a 
single pass.

The objective of this workshop is  to  provide a  discussion platform for 
researchers  and  practitioners  interested   in  all  aspects  of   rule 
extraction from  trained  artificial neural networks.  The  workshop will 
examine  current  techniques for providing an  explanation  component for 
ANNs   including   rule   extraction,   extraction of  fuzzy rules,  rule 
initialisation and  rule refinement.  Other topics for discussion include 
computational  complexity of  rule  extraction  algorithms,  criteria for 
assessing rule quality, and issues relating to generalisation differences 
between  the  ANN and  the extracted  rule set.  The workshop  will  also 
discuss  ways in   which  ANNs and  rule  extraction  techniques  may  be 
profitably employed in commercial, industrial, and scientific application 
areas.

The  one day  workshop  will  be a  mixture of  position papers and panel 
discussions. Papers presented in the mini-conference sessions will be  of 
20 minutes duration with ample time for questions/discussions afterwards. 

DISCUSSION POINTS FOR WORKSHOP PARTICIPANTS

1.  Decompositional vs. learning  approaches  to   rule-extraction   from
ANNs  -  What are the advantages and disadvantages   w.r.t.   performance,
solution  time,  computational  complexity,  problem   domain   etc.   Are
decompositional approaches always dependent on a certain ANN architecture?

2. Rule-extraction from trained neural networks v symbolic induction. What
are the relative strength and weaknesses?

3. What are the most important criteria for rule quality?

4. What  are  the  most  suitable  representation languages for  extracted
rules? How does the extraction problem vary across different languages?

5. What is the  relationship  between  rule-initialisation (insertion)  and
rule-extraction?  For   instance, are  these  equivalent  or  complementary
processes?  How  important  is rule-refinement by neural networks?

6.  Rule-extraction from trained neural networks and computational learning
theory.Is generating a minimal rule-set which mimics an ANN a hard problem?

7. Does rule-initialisation  result  in  improved generalisation and  faster
learning?

8. To  what  extent  are existing  extraction  algorithms  limited in  their
applicability?  How  can  these  limitations  be  addressed?

9.  Are there any  interesting  rule-extraction  success stories?  That  is,
problem  domains in  which the  application  of rule-extraction  methods has
resulted in an interesting or significant advance.



SUBMISSION OF WORKSHOP EXTENDED ABSTRACTS/PAPERS

Authors are invited to submit 3 copies of either an extended abstract  or
full paper relating to one of the topic areas listed above. Papers should
be written in English in single column format and should be limited to no
more than eight, (8) sides  of A4 paper including figures and references.
NIPS style files are available at
  http://www.cs.cmu.edu/afs/cs/project/cnbc/nips/formatting/nips.sty
  http://www.cs.cmu.edu/afs/cs/project/cnbc/nips/formatting/nips.tex
  http://www.cs.cmu.edu/afs/cs/project/cnbc/nips/formatting/nips.ps

Please include the following information in an accompanying cover letter: 
Full title of paper, presenting author's name, address, and telephone and
fax numbers, authors e-mail address.

Submission Deadline is October 7th,1996  with  notification to authors by
31st October,1996.


For further information,  inquiries,  and paper  submissions 
please contact:

	Robert Andrews
	Queensland University of Technology
        GPO Box 2434 Brisbane Q. 4001. Australia.
        phone  +61 7 864-1656
        fax    +61 7 864-1969
        email  robert at fit.qut.edu.au	


More information about the NIPS*96 workshop  series is available from:

   WWW:  http://www.fit.qut.edu.au/~robert/nips96.html



More information about the Connectionists mailing list