Image Classification using Reinforcement Learning

Abstract

The addition of newer labels to neural models without modifying the architecture or storing previous datasets is one of the most difficult problems that deep learning faces these days. The issues usually arise from the fact that storing data over time causes the system memory to bloat, which in turn increases the training time. The most common approach we find is to use reinforcement learning enables an agent to learn through the consequences of actions in a specific environment and allows us to use the weights and biases of trained models to train custom models for image classification. RL algorithm provides data analysis feedback, directing the user to the best result and the catch is that the previous labels are not carried forward. To improve image classification accuracy, several machine learning techniques, and models, such as Random Forest, SVM, and Logistic Regression, may be considered. The FDD agile framework will be used to apply all these methods in Python.