Papers by Rohitash Chandra
Cooperative co-evolution has been a major approach in neuro-evolution. Memetic computing approach... more Cooperative co-evolution has been a major approach in neuro-evolution. Memetic computing approaches employ local refinement to selected individuals in a population. The use of crossover-based local refinement has gained attention in memetic computing. This work proposes a cooperative co-evolutionary framework that utilises the strength of local refinement from memetic computing. It employs a crossover-based local search for refinement. The proposed framework is used for training feedforward neural networks on pattern classification problems. The results show that the proposed approach can achieve better performance than the standard cooperative coevolution framework.
Neurocomputing, 2012
Cooperative coevolution divides an optimisation problem into subcomponents and employs evolutiona... more Cooperative coevolution divides an optimisation problem into subcomponents and employs evolutionary algorithms for evolving them. Problem decomposition has been a major issue in using cooperative coevolution for neuro-evolution. Efficient problem decomposition methods group interacting variables into the same subcomponents. It is important to find out which problem decomposition methods efficiently group subcomponents and the behaviour of neural network during training in terms of the interaction among the synapses. In this paper, the interdependencies among the synapses are analysed and a problem decomposition method is introduced for feedforward neural networks on pattern classification problems. We show that the neural network training problem is partially separable and that the level of interdependencies changes during the learning process. The results confirm that the proposed problem decomposition method has improved performance compared to its counterparts.
Neurocomputing, 2012
Cooperative coevolution decomposes a problem into subcomponents and employs evolutionary algorith... more Cooperative coevolution decomposes a problem into subcomponents and employs evolutionary algorithms for solving them. Cooperative coevolution has been effective for evolving neural networks. Different problem decomposition methods in cooperative coevolution determine how a neural network is decomposed and encoded which affects its performance. A good problem decomposition method should provide enough diversity and also group interacting variables which are the synapses in the neural network. Neural networks have shown promising results in chaotic time series prediction. This work employs two problem decomposition methods for training Elman recurrent neural networks on chaotic time series problems. The Mackey-Glass, Lorenz and Sunspot time series are used to demonstrate the performance of the cooperative neuro-evolutionary methods. The results show improvement in performance in terms of accuracy when compared to some of the methods from literature.
International Journal of Applied Mathematics and …
We present a hybrid architecture of recurrent neural networks (RNNs) inspired by hidden Markov mo... more We present a hybrid architecture of recurrent neural networks (RNNs) inspired by hidden Markov models (HMMs). We train the hybrid architecture using genetic algorithms to learn and represent dynamical systems. We train the hybrid architecture on a set of deterministic finite-state automata strings and observe the generalization performance of the hybrid architecture when presented with a new set of strings which were not present in the training data set. In this way, we show that the hybrid system of HMM and RNN can learn and represent deterministic finite-state automata. We ran experiments with different sets of population sizes in the genetic algorithm; we also ran experiments to find out which weight initializations were best for training the hybrid architecture. The results show that the hybrid architecture of recurrent neural networks inspired by hidden Markov models can train and represent dynamical systems. The best training and generalization performance is achieved when the hybrid architecture is initialized with random real weight values of range -15 to 15.
Advanced Intelligent Mechatronics, …, Jan 1, 2009
This article examines an optimization method to solve the forward kinematics problem (FKP) applie... more This article examines an optimization method to solve the forward kinematics problem (FKP) applied to parallel manipulators. Based on Genetic Algorithms (GA), a non-linear equation system solving problem is converted into an optimization one. The majority of truly parallel manipulators can be modeled by the 6-6 which is an hexapod constituted by a fixed base and a mobile platform attached to six kinematics chains with linear (prismatic) actuators located between two ball joints. Parallel manipulator kinematics are formulated using the explicit Inverse Kinematics Model (IKM). The position based equation system is implemented. In order to implement GA, the objective function is formulated specifically for the FKP using one squared error performance criteria applied on the leg lengths augmented by three platform joint distances. The proposed approach implements an elitist selection process where a new mutation operator for Real-Coded GA is analyzed. These experiments are the first to converge towards several exact solutions on a general Gough platform manipulator with fast convergence.
Proc. of International Conference on Artificial …, Jan 1, 2007
We present a training approach for recurrent neural networks by combing evolutionary and gradient... more We present a training approach for recurrent neural networks by combing evolutionary and gradient descent learning. We train the weights of the network using genetic algorithms. We then apply gradient descent learning on the knowledge acquired by genetic training to further refine the knowledge. We also use genetic neural learning and gradient descent learning for training on the same network topology for comparison. We apply these training methods to the application of speech phoneme classification. We use Mel frequency cepstral coefficients for feature extraction of phonemes read from the TIMIT speech database. Our results show that the combined genetic and gradient descent learning can train recurrent neural networks for phoneme classification; however, their generalization performance does not show significant difference when compared to the performance of genetic neural learning and gradient descent alone. Genetic neural learning has shown the best training performance in terms of training time.
… . of the IASTED International Conference in …, Jan 1, 2006
We present an artificial intelligence method for the development of decision support systems for ... more We present an artificial intelligence method for the development of decision support systems for environmental management and demonstrate its strengths using an example from the domain of biodiversity and conservation biology. Renosterveld vegetation is unique to South Africa; it is under threat of extinction as a result of rapidly growing agricultural activities. We use artificial neural networks and decision trees for knowledge discovery on the Renosterveld domain. We train artificial neural networks on a dataset of the existing plant species and show their generalization performance with gradient descent learning. We then extract knowledge from trained neural networks in the form of decision trees and obtain rules which describe the existence of the remaining Renosterveld vegetation. These rules will be used as a contribution for the conservation of Renosterveld. The rules demonstrate a prediction of 78% that a Renosterveld plant will grow in a particular environment given its environmental conditions. The general paradigm can hence be applied to other plant species for knowledge discovery and the development of decision support systems.
Proc. of 8th IEEE International …, Jan 1, 2009
The forward kinematic of the 3-RPR parallel manipulator is solved using a hybrid meta-heuristic t... more The forward kinematic of the 3-RPR parallel manipulator is solved using a hybrid meta-heuristic technique where the simulated annealing algorithm replaces the mutation operator in a genetic algorithm. The results from the hybrid meta-heuristic approach is compared with the standard simulated annealing and genetic algorithm. The results show that the simulated annealing algorithm outperforms genetic algorithm in terms of computation time and overall accuracy of the solution. The hybrid meta-heuristic search algorithm shows better performance in comparision to standard genetic algorithm.
Computational Intelligence in …, Jan 1, 2009
The forward kinematics of the general Gough platform, namely the 6-6 parallel manipulator is solv... more The forward kinematics of the general Gough platform, namely the 6-6 parallel manipulator is solved using hybrid meta-heuristic techniques in which the simulated annealing algorithm replaces the mutation operator in a genetic algorithm. The results are compared with the standard simulated annealing and genetic algorithm. It shows that the standard simulated annealing algorithm outperforms standard genetic algorithm in terms of computation time and overall accuracy of the solution on this problem. However, the hybrid meta-heuristic paradigm shows the best performance in terms of accuracy and success rate.
2009 IEEE International …, Jan 1, 2009
The forward kinematics of the 6-6 leg parallel manipulator is solved using hybrid meta-heuristic ... more The forward kinematics of the 6-6 leg parallel manipulator is solved using hybrid meta-heuristic techniques in which the simulated annealing algorithm replaces the mutation operator in a genetic algorithm. The results are compared with the standard simulated annealing and genetic algorithm. It shows that the standard simulated annealing algorithm outperforms standard genetic algorithm in terms of computation time and overall accuracy of the solution on this problem. However, the hybrid meta-heuristic paradigm shows the best performance in terms of accuracy and success rate.
ecs.victoria.ac.nz
Memetic algorithms and cooperative coevolution are emerging fields in evolutionary computation wh... more Memetic algorithms and cooperative coevolution are emerging fields in evolutionary computation which have shown to be powerful tools for real-world application problems and for training neural networks. Cooperative coevolution decomposes a problem into subcomponents that evolve independently. Memetic algorithms provides further enhancement to evolutionary algorithms with local refinement. The use of crossover-based local refinement has gained attention in memetic computing. This paper employs a cooperative coevolutionary framework that utilises the strength of local refinement via crossover. The framework is evaluated by training recurrent neural networks on grammatical inference problems. The results show that the proposed approach can achieve better performance than the standard cooperative coevolution framework.
Cooperative coevolution decomposes a large problem into its subcomponents and uses evolutionary a... more Cooperative coevolution decomposes a large problem into its subcomponents and uses evolutionary algorithms for solving them in order to gradually solve the large problem. This paper uses cooperative coevolution framework for training recurrent neural networks for grammatical inference problems. In the past, different encoding schemes were used to build subcomponents from the neural network for the cooperative coevolution framework. This work proposes a new encoding scheme for building subcomponents which is based on the functional properties of a neuron and compares it with the best encoding scheme from literature. All subcomponents in their respective cooperative coevolution framework employ the G3-PCX evolutionary algorithm.
gmdh.net
This work presents the comparison and combination of neural networks with decision trees on the a... more This work presents the comparison and combination of neural networks with decision trees on the application of wine classification. Neural networks are first trained and then combined with decision trees in order to extract knowledge learnt in the training process. Artificial neural networks are used for the classification of Italian wines obtained from a region which has three different wine cultivars. Wines are classified according to their respective cultivar using the chemical analysis of the thirteen major chemical constituents. The trained network classifies a sample of wine according to the knowledge the network acquired by learning from previous wine samples. After successful training, knowledge is extracted from these trained networks using decision trees in the form of 'if-then' rules. We then use decision trees to train on the same dataset and compare the performance of neural networks, and decision trees in both knowledge extraction from neural networks and classification of wines on their own. Our results show that artificial neural networks perform better when compared to decision trees however, the extraction of knowledge from neural networks do not outperform the performance of decision trees alone. The general paradigm can be applied to other categories of food classification and processing.
gmdh.net
This work presents the application of neural networks to real-life prediction problems. Neural ne... more This work presents the application of neural networks to real-life prediction problems. Neural networks are trained to predict three real world application problems given data for training and testing. We pre-process the actual output of the dataset by converting the values to integers and later translate them to binary strings. The sigmoidal output neurons of the feed-forward architecture predicts the output as binary values which are then translated back to integers to further compare the predicted output values of the network with the desired or actual output values from the dataset. The results for two, out of the three problems solved show that neural networks can be used to predict real-life problems similar to other inductive modeling approaches. A neural network can therefore be classified as an inductive modeling approach since it can be used for prediction, given the desired output . Finally, the performance of neural networks are then compared to GMDH.
Uploads
Papers by Rohitash Chandra