Evolving Deep DenseBlock Architecture Ensembles for Image Classification

Automatic deep architecture generation is a challenging task, owing to the large number of controlling parameters inherent in the construction of deep networks. The combination of these parameters leads to the creation of large, complex search spaces that are feasibly impossible to properly navigate without a huge amount of resources for parallelisation. To deal with such challenges, in this research we propose a Swarm Optimised DenseBlock Architecture Ensemble (SODBAE) method, a joint optimisation and training process that explores a constrained search space over a skeleton DenseBlock Convolutional Neural Network (CNN) architecture. Specifically, we employ novel weight inheritance learning mechanisms, a DenseBlock skeleton architecture, as well as adaptive Particle Swarm Optimisation (PSO) with cosine search coefficients to devise networks whilst maintaining practical computational costs. Moreover, the architecture design takes advantage of recent advancements of the concepts of residual connections and dense connectivity, in order to yield CNN models with a much wider variety of structural variations. The proposed weight inheritance learning schemes perform joint optimisation and training of the architectures to reduce the computational costs. Being evaluated using the CIFAR-10 dataset, the proposed model shows great superiority in classification performance over other state-of-the-art methods while illustrating a greater versatility in architecture generation.

Genetic Fuzzy Systems Applied to Ports and Coasts Engineering

Fuzzy Logic (FL) and fuzzy sets in a wide interpretation of FL (in terms in which fuzzy logic is coextensive with the theory of fuzzy sets, that is, classes of objects in which the transition from membership to non membership is gradual rather than abrupt) have placed modelling into a new and broader perspective by providing innovative tools to cope with complex and ill-defined systems. The area of fuzzy sets has emerged following some pioneering works of Zadeh (Zadeh, 1965 and 1973) where the first fundamentals of fuzzy systems were established. Rule based systems have been successfully used to model human problem-solving activity and adaptive behaviour. The conventional approaches to knowledge representation are based on bivalent logic. A serious shortcoming of such approaches is their inability to come to grips with the issue of uncertainty and imprecision. As a consequence, the conventional approaches do not provide an adequate model for modes of reasoning. Unfortunately, all commonsense reasoning falls into this category. The application of FL to rule based systems leads us to fuzzy systems. The main role of fuzzy sets is representing Knowledge about the problem or to model the interactions and relationships among the system variables. There are two essential advantages for the design of rule-based systems with fuzzy sets and logic: • The key features of knowledge captured by fuzzy sets involve handling uncertainty. • Inference methods become more robust and flexible with approximate reasoning methods of fuzzy logic. Genetic Algorithms (GAS) are a stochastic optimization technique that mimics natural selection (Holland, 1975). GAs are intrinsically robust and capable of determining a near global optimal solution. The use of GAS is usually recommended for optimization in high-dimensional, multimodal complex search spaces where deterministic methods normally fail. GAs explore a population of solutions in parallel. The GA is a searching process based on the laws of natural selections and genetics. Generally, a simple GA contains three basic operations: selection, genetic operations and replacement. A typical GA cycle is shown in Fig. 1. In this paper it is shown how a genetic algorithm can be used in order to optimize a fuzzy system which is used in wave reflection analysis at submerged breakwaters.

Swarm Intelligence and the Taguchi Method for Identification of Fuzzy Models

Nature is a wonderful source of inspiration for building models and techniques for solving difficult problems in design, optimisation, and control. More specifically, the study of evolution, the human immune system, and the collective behaviour of insects/birds have guided the origin of evolutionary algorithms, artificial immune systems, and optimisation techniques based on swarm intelligence, respectively. In this chapter, we present the use of particle swarm optimisation (PSO) and the Taguchi method for the identification of optimised fuzzy models from the available data. PSO is a member of the broad category of swarm intelligence (SI) techniques based on the metaphor of social interaction. It has been used for finding promising solutions in complex search spaces through the interaction of particles in a swarm, and is especially useful when dealingwith a high number of dimensions and situations where problem-specific information is not available. However, caution needs to be exercised in selecting PSO, as the performance of PSO largely depends on their values. In this chapter, a systematic reasoning approach based on the Taguchi method is also presented to quickly identify PSO parameters. The Taguchi method is a robust design approach that helps in optimisation, and which requires relatively few experiments. Although we focus here on the use of PSO and the Taguchi method for fuzzy model identification, these techniques have much broader use and application. In order to validate our approach, data from the rapid Nickel-Cadmium (Ni-Cd) battery charger developed by the authors were used. The results are based on real data and illustrate the viability and efficiency of the approach.