Browsing by Author "Liao, Shih-Hui"

Now showing 1 - 9 of 9

A Hybrid of Cuckoo Search and Simplex Method for Fuzzy Neural Network Training
(2015)
Chang, Jyh-Yeong
;
Liao, Shih-Hui
;
Wu, Shang-Lin
;
Lin, Chin-Teng
;
電控工程研究所
In this paper, a new hybrid algorithm mixing the simplex method of Nelder and Mead (NM) and the cuckoo search (CS), abbreviated as NM-CS, is proposed for the training of the Fuzzy Neural Networks (FNNs). In standard CS, cuckoo birds engage the obligate brood parasitism by laying their own eggs to other host birds. If a host bird discovers the alien eggs, they will either throw these eggs away or abandon its nest and build a new nest elsewhere. In the proposed hybrid algorithm, instead of using the probability to discover an alien egg for the CS, we use the concept of a simplex which is used in the NM algorithm to abandon and generate the new nests. Our proposed method puts more emphasis on exploration of the search space and enhances the ability to avoid local optimum. Some simulation problems will be provided to compare the performances of the proposed method and other methods in training an FNN. In these simulations, it is observed that the proposed method outperforms other methods.
Adaptive Least Trimmed Squares Fuzzy Neural Network
(2012)
Chang, Jyh-Yeong
;
Liao, Shih-Hui
;
Lin, Chin-Teng
;
電控工程研究所
;
Institute of Electrical and Control Engineering
In this paper, we propose the adaptive least trimmed squares fuzzy neural network (ALTS-FNN), which applies the scale estimate to the least trimmed squares fuzzy neural network (LTS-FNN). The emphasis of this paper is particular on the robustness against the outliers and the choice of the trimming constant can be determined adaptively. Some numerical examples will be provided to compare the robustness against outliers for usual FNN and the ALTS-FNN. Simulation results show that the ALTS-FNN in the paper have good performance for outlier detection.
Dynamic group-based differential evolution using a self-adaptive strategy for global optimization problems
(2013-07-01)
Han, Ming-Feng
;
Liao, Shih-Hui
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
;
電控工程研究所
This paper describes a dynamic group-based differential evolution (GDE) algorithm for global optimization problems. The GDE algorithm provides a generalized evolution process based on two mutation operations to enhance search capability. Initially, all individuals in the population are grouped into a superior group and an inferior group based on their fitness values. The two groups perform different mutation operations. The local mutation model is applied to individuals with better fitness values, i.e., in the superior group, to search for better solutions near the current best position. The global mutation model is applied to the inferior group, which is composed of individuals with lower fitness values, to search for potential solutions. Subsequently, the GDE algorithm employs crossover and selection operations to produce offspring for the next generation. In this paper, an adaptive tuning strategy based on the well-known 1/5th rule is used to dynamically reassign the group size. It is thus helpful to trade off between the exploration ability and the exploitation ability. To validate the performance of the GDE algorithm, 13 numerical benchmark functions are tested. The simulation results indicate that the approach is effective and efficient.
Neuro-Fuzzy System Design Using Differential Evolution with Local Information
(2011)
Lin, Chin-Teng
;
Han, Ming-Feng
;
Lin, Yang-Yin
;
Liao, Shih-Hui
;
Chang, Jyh-Yeong
This paper proposes a differential evolution with local information for TSK-type neuro-fuzzy system optimization. The differential evolution with local information consider neighborhood between each individual to keep the diversity of population. An adaptive parameter tuning based on 1/5th rule is used to trade off between local search and global search. For structure learning algorithm, the on-line clustering algorithm is used for rule generation. The structure learning algorithm generates a new rule which compares the firing strength. Initially, there is no rule in neuro-fuzzy system model. The rules are automatically generated by fuzzy measure. For parameter learning, the parameters are optimized by differential evolution algorithm. Finally, the proposed neuro-fuzzy system with novel differential evolution model is applied in chaotic sequence prediction problem. Results of this paper demonstrate the effectiveness of the proposed model.
Preliminary Study on Additive Radial Basis Function Networks
(2010)
Liao, Shih-Hui
;
Lin, Chin-Teng
;
Chang, Jyh-Yeong
;
電控工程研究所
;
Institute of Electrical and Control Engineering
In this paper, a new class of learning models, namely the additive radial basis function networks (ARBFNs) for general nonlinear regression problems are proposed. This class of learning machines combines the radial basis function networks (RBFNs) commonly used in general machine learning problems and the additive models (AMs) frequently encountered in semi parametric regression problems. In statistical regression theory, AM is a good compromise between the linear parametric model and the non parametric model. Simulation results show that for the given learning problem, ARBFNs usually need fewer hidden nodes than those of RBFNs for the same level of accuracy.
Simplified Interval Type-2 Fuzzy Neural Networks
(2014-05-01)
Lin, Yang-Yin
;
Liao, Shih-Hui
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
;
電控工程研究所
This paper describes a self-evolving interval type-2 fuzzy neural network (FNN) for various applications. As type-1 fuzzy systems cannot effectively handle uncertainties in information within the knowledge base, we propose a simple interval type-2 FNN, which uses interval type-2 fuzzy sets in the premise and the Takagi-Sugeno-Kang (TSK) type in the consequent of the fuzzy rule. The TSK-type consequent of fuzzy rule is a linear combination of exogenous input variables. Given an initially empty the rule-base, all rules are generated with on-line type-2 fuzzy clustering. Instead of the time-consuming K-M iterative procedure, the design factors q(l) and q(r) are learned to adaptively adjust the upper and lower positions on the left and right limit outputs, using the parameter update rule based on a gradient descent algorithm. Simulation results demonstrate that our approach yields fewer test errors and less computational complexity than other type-2 FNNs.
Study on Adaptive Least Trimmed Squares Fuzzy Neural Network
(2013-09-01)
Liao, Shih-Hui
;
Han, Ming-Feng
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
;
電控工程研究所
In the largest samplings of data, outliers are observations that are well separated from the major samples. To deal with outlier problems, a least trimmed squares (LTS) estimator is developed for robust linear regression problems. It is meaningful to generalize the LTS estimator to fuzzy neural network (FNN) for robust nonlinear regression problems. In addition, the determination of the trimming constant is important when using the LTS estimator. In this paper, we propose the use of an adaptive least trimmed squares fuzzy neural network (ALTS-FNN), which applies a scale estimate to a LTS-FNN. This paper particularly emphasizes the robustness of the proposed network against outliers and an automatic determination of the trimming percentage. Simulation problems are provided to compare the performance of the proposed ALTS-FNN, with an LTS-FNN and typical FNN. Simulation results show that the proposed ALTS-FNN is highly robust against outliers.
Study on Least Trimmed Absolute Deviations Artificial Neural Network
(2013)
Liao, Shih-Hui
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
;
電控工程研究所
;
Institute of Electrical and Control Engineering
In this paper, the least trimmed sum of absolute deviations (LTA) estimator, frequently used in robust linear parametric regression problems, will be generalized to nonparametric least trimmed sum of absolute deviations-artificial neural network (LTA-ANN) for nonlinear regression problems. In linear parametric regression problems, the LTA estimator usually have good robustness against outliers and can theoretically tolerate up to 50% of outlying data. Moreover, a nonderivative hybrid method mixing the simplex method of Nelder and Mead (NM) and particle swarm optimization algorithm (PSO), abbreviated as SNM-PSO, will be provided in this study for the training of the parameters of LTA-ANN. Some numerical examples will be provided to compare the robustness against outliers for usual artificial neural network (ANN) and the proposed LTA-ANN. Simulation results show that the LTA-ANN proposed in this paper have good robustness against outliers.
Training neural networks via simplified hybrid algorithm mixing Nelder-Mead and particle swarm optimization methods
(2015-03-01)
Liao, Shih-Hui
;
Hsieh, Jer-Guang
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
;
電控工程研究所
In this paper, a new and simplified hybrid algorithm mixing the simplex method of Nelder and Mead (NM) and particle swarm optimization algorithm (PSO), abbreviated as SNM-PSO, is proposed for the training of the parameters of the Artificial Neural Network (ANN). Our method differs from other hybrid PSO methods in that, n+1 particles, where n is the dimension of the search space, are randomly selected (without sorting), at each iteration of the proposed algorithm for use as the initial vertices of the NM algorithm, and each such particle is replaced by the corresponding final vertex after executing the NM algorithm. All the particles are then updated using the standard PSO algorithm. Our proposed method is simpler than other similar hybrid PSO methods and places more emphasis on the exploration of the search space. Some simulation problems will be provided to compare the performances of the proposed method with PSO and other similar hybrid PSO methods in training an ANN. These simulations show that the proposed method outperforms the other compared methods.