1 / 14

University of Patras and CTI, Patras

2 nd Panhellenic Conference on Artificial Intelligence, 11-12 April, 2002, Thessaloniki. Multi-inference with Multi-neurules. University of Patras and CTI, Patras. I. Hatzilygeroudis J. Prentzas. Introductory observations.

cyrah
Download Presentation

University of Patras and CTI, Patras

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. 2nd Panhellenic Conference on Artificial Intelligence, 11-12 April, 2002, Thessaloniki Multi-inference with Multi-neurules University of Patras and CTI, Patras I. Hatzilygeroudis J. Prentzas

  2. Introductory observations • Connectionist expert systems is an effort to integrate symbolic representation and neural networks (e.g. MACIE, Gallant 1988 & 1993, Ghalwash 1998) • A disadvantage is the lack of naturalness in representation, inference and explanation • Another disadvantage is the absolute reliance on empirical data • Reason: they give pre-eminence to the connectionist framework both in representation and inference

  3. Neurules • Introduced by Hatzilygeroudis and Prentzas, 2000 & 2001 • Give pre-eminence to the symbolic framework in representation • Give pre-eminence to neurocomputing in the inference process

  4. f(x) 1 x -1 Ci : {1 , 0 , 0.5} {true , false , unknown} D : {1 , -1} {success , failure} Neurules: the model Neurule D Adaline Unit (sf0) (sf1) (sfn) (sf2) . . . C1 C2 Cn

  5. Neurules: the syntax (sf0) ifC1 (sf1),C2 (sf2),…, Cn(sfn) thenD1,D2 • Ci: conditions (‘fever is high’) • Di: conclusions (‘disease isinflammation’) • sf0: bias factor • sfi: significance factors variable predicate value

  6. Neurules: disadvantages • Create Neurule bases with multiple representations of the same piece of knowledge (due to the non-separability problem) • The associated inference mechanism is rather connectionism oriented, thus reducing naturalness

  7. Multi-neurule m m m m (CFn) (CF0 ) (CF1 ) (CF2) Multi-adaline Unit Multi-neurules: the model D . . . C1 C2 Cn i= 1, n (sf-tuples)

  8. Multi-neurules: the syntax (sf11, sf12, …, sf1m) (sf01, sf02, …, sf0m) ifC1 , C2, … Cn thenD (sf21, sf22, …, sf2m) (sfn1, sfn2, …, sfnm) … size: m RFm RF1 RF2 RFi = (sf1i, sf2i, …, sfni) i = 1, m (sf-sets)

  9. An example NR5: (-2.2) if Treatment is Biramibio (-2.6), Treatment is Placibin (1.8) then Treatment is Posiboost NR6: (-2.2) if Treatment is Placibin (-1.8), Treatment is Biramibio (1.0) then Treatment is Posiboost NR5: (-2.2, -2.2) if Treatment is Placibin (-1.8, 1.8), Treatment is Biramibio (1.0, -2.6) then Treatment is Posiboost

  10. kn-sum rem-sum Neurules evaluation (1) • Known sum(weighted sum of evaluated conditions E) • Remaining sum(largest possible weighted sum of unevaluated conditions U) • Firing potential fp = • A multi-neurule of size m has m different kn-sums, rem-sums and fps, one for each RFi.

  11. 1 if kn-sum  0  fired D in WM -1 if kn-sum < 0  blocked D in WM Neurules evaluation (2) • Evaluated conditions: their value (true or false) has been known. • If |kn-sum| > rem-sum(fp >1) or rem-sum=0 for a neurule: D =

  12. Inference processes • Connectionism-oriented process • The choice of the next neurule to be considered is based on fp. • Firing or blocking of a neurule results in updates of the fps of the affected neurules. • Symbolism-oriented process • It is based on a backward chaining strategy and textual order.

  13. KB Connectionism oriented process Symbolism oriented process ASKED EVALS ASKED EVALS ANIMALS (20 inferences) 162 (8.1), 364 (18.2) 142 (7.1) 251 (12.5) LENSES (24 inferences) 79 (3.3) 602 (25.1) 85 (3.5) 258 (10.8) ZOO (101inferences) 1052 (10.4) 8906 (88.2) 1013 (10) 1963 (19.4) MEDICAL (134inferences) 670 (5) 25031 (186.8) 670 (5) 11828 (88.3) Experimental results

  14. Conclusions • Multi-neurules result in more concise representation, at a possible expense of some naturalness • The symbolism-oriented inference process, apart from being more natural, is also more efficient than the connectionism-oriented one

More Related