NEUROPLASTICITY OF ARTIFICIAL NEURAL NETWORKS: AN INVESTIGATION USING ENGLISH AND DEVANAGARI CHARACTER RECOGNITION

The human brain is the most complex system known to us. The most intriguing of its awe-inspiring powers is neuroplasticity – the brain's ability to learn whatever it is trained to do. This paper investigates the neuroplasticity of artificial neural networks for character recognition of English and Devanagari scripts. An artificial neural network that recognizes characters from a gray-scale image is developed. Two sets of character images are used to train this neural network: images of the first four English characters and images of the first four Devanagari characters. First, the network was trained on images of the English characters, and it recognized those characters with 98 percent accuracy. However, after this training the neural network did not recognize Devanagari characters. Then, the network was trained on images of Devanagari characters, and it recognized Devanagari characters with 97 percent accuracy. However, the neural network did not recognize English characters at all after retraining. This demonstrates that there is nothing in the data structures or the learning algorithm of the neural network that is specific to English or Devanagari characters. The artificial neural network learns from the training data set presented to it, thus exhibiting neuroplasticity.