Evolving artificial neural networks using an improved PSO and DPSO

@article{citeulike:2221492, abstract = {This paper presents an improved particle swarm optimization (PSO) and discrete PSO (DPSO) with an enhancement operation by using a self-adaptive evolution strategies (ES). This improved PSO/DPSO is proposed for joint optimization of three-layer feedforward artificial neural network (ANN) structure and parameters (weights and bias), which is named ESPNet. The experimental results on two real-world problems show that ESPNet can produce compact ANNs with good generalization ability.}, author = {Yu, Jianbo and Wang, Shijin and Xi, Lifeng }, citeulike-article-id = {2221492}, comment = {benchmark: credit-card-assessment, diabete To evaluate the performance of ESPNet, they have applied it to two benchmark problems: the Australian credit card assessment problem and the diabetes problem. Both data sets were obtained from the UCI machine learning benchmark repository. discrete PSO, DPSO In ESPNet, the 'control dimensions' are evolved by DPSO, while the 'connection dimensions' are evolved by PSO. evolution-strategy ES is used in ESPNet when DPSO cannot improve the best global position to enhance DPSO. MPANN, EENCL ESPNet is compared to the results of EENCL and MPANN. variable-size-structure ESPNet evolves simultaneously the structure and parameters of ANNs. * (In comparison to GA) the applications of PSO for evolving ANNs are relatively sparse. * The problem of designing a near optimal ANN structure by using PSO for an application remains unsolved.}, doi = {http://dx.doi.org/10.1016/j.neucom.2007.10.013}, journal = {Neurocomputing}, keywords = {08, 4s, benchmark, credit-card-assessment, diabete, direct-encoding, dpso, eencl, espnet, evolution-strategy, feedforward, gaussian-mutation, genotype-encoding, mpann, neural-network, nn, particle-swarm-optimization, pso, structure-evolution, variable-structure}, posted-at = {2008-01-12 02:58:34}, priority = {0}, title = {Evolving artificial neural networks using an improved PSO and DPSO}, url = {http://dx.doi.org/10.1016/j.neucom.2007.10.013}, volume = {In Press, Corrected Proof}, year = {2008} }

TY - JOUR ID - citeulike:2221492 L3 - citeulike-article-id:2221492 TI - Evolving artificial neural networks using an improved PSO and DPSO JF - Neurocomputing VL - In Press, Corrected Proof N2 - This paper presents an improved particle swarm optimization (PSO) and discrete PSO (DPSO) with an enhancement operation by using a self-adaptive evolution strategies (ES). This improved PSO/DPSO is proposed for joint optimization of three-layer feedforward artificial neural network (ANN) structure and parameters (weights and bias), which is named ESPNet. The experimental results on two real-world problems show that ESPNet can produce compact ANNs with good generalization ability. KW - 08 KW - 4s KW - benchmark KW - credit-card-assessment KW - diabete KW - direct-encoding KW - dpso KW - eencl KW - espnet KW - evolution-strategy KW - feedforward KW - gaussian-mutation KW - genotype-encoding KW - mpann KW - neural-network KW - nn KW - particle-swarm-optimization KW - pso KW - structure-evolution KW - variable-structure N1 - benchmark: credit-card-assessment, diabete To evaluate the performance of ESPNet, they have applied it to two benchmark problems: the Australian credit card assessment problem and the diabetes problem. Both data sets were obtained from the UCI machine learning benchmark repository. discrete PSO, DPSO In ESPNet, the 'control dimensions' are evolved by DPSO, while the 'connection dimensions' are evolved by PSO. evolution-strategy ES is used in ESPNet when DPSO cannot improve the best global position to enhance DPSO. MPANN, EENCL ESPNet is compared to the results of EENCL and MPANN. variable-size-structure ESPNet evolves simultaneously the structure and parameters of ANNs. * (In comparison to GA) the applications of PSO for evolving ANNs are relatively sparse. * The problem of designing a near optimal ANN structure by using PSO for an application remains unsolved. AU - Yu, Jianbo AU - Wang, Shijin AU - Xi, Lifeng PY - 2008/// UR - http://dx.doi.org/10.1016/j.neucom.2007.10.013 ER -