Search

The hybridization of evolutionary technology has been extensively used to enhance the performance of recurrent type neural networks (RTNN) for storing patterns and their recalling. Several experiments have been done to link evolutionary processes such as genetic algorithm (GA) with RTNN regarding the connection weight among the processing elements. This integration strengthens the efficiency of the Recurrent neural network (RNN) to effectively recall the increased capacity and patterns of sample storage to reduce the flaw of local minima. Bipolar product rule (BPR) has been applied predominantly for pattern storage, and GA are further used to develop the weight matrix to explore the global optimal solution reflecting the correct invocation of the storage pattern. Here, Edge Detection (ED) and self-organizing map (SOM) methods are applied for the purpose of feature extraction. The modified BPR and GA have been employed to store patterns, and recalling respectively. The proposed hybrid RTNN performance is examined for the handwritten Greek symbols. [ABSTRACT FROM AUTHOR]

Byline: Somesh Kumar, Manu Pratap Singh In this paper, we are analysing the performance of Hopfield model of feedback neural networks (NNs) with general Hebbian learning rule and genetic algorithm (GA) for pattern recognition. In the Hopfield type of NNs, the weighted code of input patterns provides an auto-associative function in the network, which exhibits its associative memory feature. The objective is to determine the optimal weight matrix for efficient recalling of any approximate input pattern. For this, we explore the population generation technique (mutation and elitism), crossover and setting up of proper fitness evaluation functions to generate the new population of the weight matrices. This process will continue until the last weight matrix has been selected. The experiments consider a neural network architecture that stores all letters of English alphabets (capitals only) using Hebbian rule and then accomplishes the recalling of these stored patterns on presentation of any prototype input pattern of the already stored patterns using both conventional Hebbian rule and evolutionary algorithm. The simulated results demonstrate the better performance of network for recalling of the stored letters of English alphabets using GA and minimise the randomness from the GA.

Carrying out the classification of patterns in a two-qubit system by separately using Grover’s and Ventura’s algorithms on different possible superposition, it has been shown that the exclusion superposition and the phase-invariance superposition are the most suitable search states obtained from two-pattern start-states and one-pattern start-states, respectively, for the simultaneous classifications of patterns. The higher effectiveness of Grover’s algorithm for large search states has been verified but the higher effectiveness of Ventura’s algorithm for smaller data base has been contradicted in two-qubit systems and it has been demonstrated that the unknown patterns (not present in the concerned data-base) are classified more efficiently than the known ones (present in the data-base) in both the algorithms. It has also been demonstrated that different states of Singh-Rajput MES obtained from the corresponding self-single- pattern start states are the most suitable search states for the classification of patterns |00>,|01 >, |10> and |11> respectively on the second iteration of Grover’s method or the first operation of Ventura’s algorithm. [ABSTRACT FROM AUTHOR]