I relish in how they perform complex computations with ease, and long to make them even more efficient. Through my studies, I have found that neural networks are particularly useful in classification tasks, as in my undergraduate project where my tea and I set up a system to classify an input image (a face) into a category. In this endeavor, we used supervised learning, which is a technique wherein we provide the system with objects (images) and the groups they 're classified into. For example, we would give the system ten images of a person X and train the system that this is person X. Although training a system to perform given pattern classification tasks may seem simplistic to some, it actually involved complex feed-forward and feed-back non-linear structures, along with implementation of non-linear systems theory and various types of training algorithms. In addition, we needed to develop a sufficiently diverse and comprehensive set of training samples for classes in order to ensure that the network could perform facial recognition on samples not present in the training set. I found it a think of beauty how the neural network adjusted and modified itself as we crafted it to recognize more and more people. Although, my greatest source of price came when, upon testing under different environments, it provide to be accurate 97% of the time, a virtually unheard of rate for such as undergraduate
I relish in how they perform complex computations with ease, and long to make them even more efficient. Through my studies, I have found that neural networks are particularly useful in classification tasks, as in my undergraduate project where my tea and I set up a system to classify an input image (a face) into a category. In this endeavor, we used supervised learning, which is a technique wherein we provide the system with objects (images) and the groups they 're classified into. For example, we would give the system ten images of a person X and train the system that this is person X. Although training a system to perform given pattern classification tasks may seem simplistic to some, it actually involved complex feed-forward and feed-back non-linear structures, along with implementation of non-linear systems theory and various types of training algorithms. In addition, we needed to develop a sufficiently diverse and comprehensive set of training samples for classes in order to ensure that the network could perform facial recognition on samples not present in the training set. I found it a think of beauty how the neural network adjusted and modified itself as we crafted it to recognize more and more people. Although, my greatest source of price came when, upon testing under different environments, it provide to be accurate 97% of the time, a virtually unheard of rate for such as undergraduate