This research demonstrates the use of TensorFlow to build a Hierarchical Neural Network (HNN). Constructing and engineering neural networks to maximize accuracy and efficiency is an active field of research in machine learning. HNN, along with several other applications of split networks have been developed as recently as 2017. However, implementations thus far have required custom-built and coded HNNs. The research conducted here uses TensorFlow to validate this structure by building entirely separate neural nets with logical relations between the output of one net and the inputs of the nets that are downstream. Research has shown that Hierarchical Neural Networks can increase training speed and reduce compute resources. The validation results of HNN using the Fashion-MNIST dataset demonstrate a prediction accuracy of 99.49%, 95.96%, and 88.84% for coarse, medium, and fine level classification, respectively, in which the fine level classification accuracy is greater than the baseline model.
Fontenot, Rick; Lazarus, Joseph; Rudick, Puri; and Sgambellone, Anthony
"Hierarchical Neural Networks (HNN): Using TensorFlow to build HNN,"
SMU Data Science Review: Vol. 6:
2, Article 4.
Available at: https://scholar.smu.edu/datasciencereview/vol6/iss2/4
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License