Simulating human thinking on computer systems by means of mathematical models is one of the most challenging prolems in the field of Computer Science. A recently developed such model is the artificial neural network, inspired by the neuron structure in the human brain. A limiting factor for the performance of this model is the depth of neural connections that can be established while retaining learning ability. We attempt to remove this limitation in the field of automated reading comprehension using a recent neural network model known for achieving high theoretical efficiency. We further optimize the implementation and architecture of that model and exhibit speed improvement up to 5 times better than the original implementation. In addition, we use our improved model for real time text analysis, speech recognition, and cryptoanalysis, and achieve higher performance than the current state-of-the-art model.
Key words: neural network, recurrent, LSTM, unitary matrix, automatic text understaning, speech recognition, cryptography, vigenère
This paper provides a comparison between metaheuristic algorithms, designed to find solutions of the Graph Coloring Problem. Such comparison can only be found in a generalized form without any detailed information. A Genetic Algorithm, an Ant Colony Optimization Algorithm, a Simulated Annealing Algorithm, and a Tabu Search Algorithm have been implemented, optimized and evaluated on standard tests, created by the Center for Discrete Mathematics and Theoretical Computer Science (DIMACS). The goal of the research is to contribute with a detailed comparison and analysis of the results.