Hybrid Machine Learning Approaches A Method to Improve Expected Output of Semi-structured Sequential Data

This paper proposes an intuitive yet simple machine learning (ML) approach that consist of two generic algorithms augmenting one another to solve problems they are not designed to solve. Since most machine learning algorithms are designed for a particular dataset or task, combining multiple ML algorithms can greatly improve the overall result by either helping tune one another, generalize, or adapt to unknown tasks. In this paper, we attempt to augment the architecture of traditional Artificial Neural Network (ANN) with a state machine acting as a form of short term memory in addition to help divide the work amongst multiple modular ANNs through transitioning from state to state. The result is a larger non-stochastic network that is able to self adjust as it is fed input. We train and test the work on data that is outside either an Artificial Neural Network or a state-machine's normal capability with simplified music notation extracted from midi files. The extracted data are used to simulate inherently sequential data to test the principle. Finally, while we find many large improvements in the augmentation of the ANN's architecture, but discuss further approaches to the system to improve generalization for new data.