wk-7
1
4
Neural Network Project
Name: Shrey Bavaria
Student ID: 002843842
Institution: University of the Cumberlands
Course: ITS531 – Business Intelligence
Date: 11/2/2020
Neural networks consist of algorithms used to interpret data and recognize data patterns by the use of cluttering and labelling mechanisms (Au, 2020). Neural networks are used to recognize underlying relationships in various datasets presented by mimicking the function of human brain. There are different neural network projects that can be implemented to enhance the understanding different network architecture and how they work. There are different frameworks that can be used to develop neural network project and these include Tensorflow, Keras, PyTorch or Numpy (Au, 2020).
The first step of developing neural network is initialization. In this step, the framework that is needed for the development of the network is downloaded and set up ready for use. In the second step, data generation is done. There are various datasets that are available online but one can come up with own generated data depending on the sets that need to be analyzed. In the third step, train-test splitting is carried out. Any percentage of the data can be classified into test or train depending on the objectives set buy the user.
For example, 70% of the data can be split into training while the remaining 30% used for testing. The training set is tuned for neural network development and the testing set for performance evaluation. In data standardization, the data split for training is standardized to feature zero mean and unit variance. This scaler can also be used in the evaluation of the test data (Sharda, et al, 2020). The four steps are referred as the preprocessing steps. The fifth step involves neural net construction and, in this step, a layer is created using a programming class and in this case python.
Each layer has a matrix, vector and activation functions. Step six, forward propagation involves definition of a function to be used for forward propagation and it is given certain set of weights and biases. In step seven, back propagation, partial derivatives are used to calculate analytically the loss of metric change. In step eight, iterative optimization, the sensitivities of the weights have been known and minimization of loss of metric iteratively is focused on. The final step which is testing involves generalization of the testing and training loss and how effective is the neural project (Sharda, et al, 2020).
References
Au E. (2020). Building Neural Networks from Scratch in 9 Steps. Retrieved from https://medium.com/dataseries/building-neural-networks-from-scratch-in-9-steps-a5d3e1f8c711