110 likes | 245 Views
Active Sonar Target Identification Using Evolutionary Neural Logic Networks. Athanasios Tsakonas Dept. of Financial and Management Engineering, University of the Aegean , Greece Georgios Dounias Dept. of Financial and Management Engineering, University of the Aegean , Greece
E N D
Active Sonar Target Identification Using Evolutionary Neural Logic Networks Athanasios Tsakonas Dept. of Financial and Management Engineering, University of the Aegean, Greece Georgios DouniasDept. of Financial and Management Engineering, University of the Aegean, Greece Nikitas Nikitakos Dept. of Shipping, Trade and Finance, University of the Aegean, Greece Presenting Author:Emmanouil Vasilakis, University of the Aegean, Greece
Contents • Neural-Symbolic Systems and Integration • Neural Logic Networks • Expressing NLNs into PROLOG rules • Constructing NLNs from data • Past approaches • The Evolutionary NLNs • Applications • Summary NeSy-2006, Riva Del Garda
Neural – Symbolic Integration Unification systems Hybrid Systems Neuronal Modeling/Neuroscience Connectionist Logic Systems Hybrid Systems by Translation Hybrid Systems by Function Neural-Symbolic Systems and Integration Source: D’Avila Garcez (2002) NeSy-2006, Riva Del Garda
Neural Logic Networks • Finite directed graph • Consisted by a set of input nodes and an output node • The possible value for a node can be one of three ordered pair activation values (1,0) for true, (0,1) for false and (0,0) for don't know NeSy-2006, Riva Del Garda
Expressing NLNs into PROLOG rules • We may create rules into the programming language PROLOG directly by every neural logic network. NeSy-2006, Riva Del Garda
Constructing NLNs from data : Past approaches • Tan et al. 1996 • Teh 1995 NeSy-2006, Riva Del Garda
Constructing NLNs from data : Past approaches • Chia and Tan 2001 NeSy-2006, Riva Del Garda
Constructing NLNs from data : the Evolutionary NLNs NeSy-2006, Riva Del Garda
Active Sonar Identification • 86.27 % in test set (CNLN (P1 (P1 (P1 (S1 (S1 (In T10) (Rule 0 0) E) (Rule 0 0) (S2 E (Rule 0 0) E)) (P1 (S1 (In T4) (Rule 0 0) (P2 E (Rule 10 3) (S2 E (Rule 10 3) E))) (In T11))) (P1 (P1 (In T3) (S1 (In T48) (Rule 12 8) E)) (P1 (P1 (P1 (S1 (S1 (In T10) (Rule 0 0) E) (Rule 0 0) (S2 E (Rule 0 0) E)) (P1 (S1 (S1 (In T4) (Link 50 0 (Rule 0 0)) E) (Rule 10 3) E) (P1 (S1 (In T4) (Rule 12 8) (P2 E (Rule 10 3) (S2 E (Rule 0 0) E))) (In T11)))) (P1 (P1 (In T58) (S1 (In T24) (Rule 12 8) (P2 E (Link 133 0 (Rule 0 0)) E))) (P1 (P1 (S1 (In T52) (Rule 0 0) E) (P1 (S1 (In T4) (Link 133 0 (Rule 10 3)) (P2 E (Rule 0 0) (S2 E (Rule 0 0) E))) (P1 (S1 (In T28) (Rule 0 0) (P2 E (Rule 10 3) (S2 E (Rule 0 0) E))) (In T11)))) (P1 (P1 (In T58) (S1 (In T24) (Rule 12 8) (P2 E (Link 133 0 (Rule 0 0)) E))) (P1 (P1 (S1 (In T31) (Rule 10 3) E) (In T49)) (In T42)))))) (In T4)))) (In T50)) (Rule 2 8)) NeSy-2006, Riva Del Garda
Summary • Neural-Symbolic Integration • Neural Logic Network • We proposed an evolutionary technique that uses: • Cellular encoding • Genetic programming • Grammar-based search guidance • Results – Application in Active Sonar Identification NeSy-2006, Riva Del Garda