Fine-Grained Categorization Using a Mixture of Transfer Learning Networks

Abstract

In this paper, we apply a mixture of experts approach to further enhance the accuracy of transfer learning networks on a fine-grained categorization problem, expanding on the work of Firsching and Hashem [4]. Mixture of experts approaches may help to improve accuracy on categorization problems. Likewise, transfer learning is a highly effective tech nique for solving problems in machine learning of varying complexities. We here illustrate that mixtures of trained transfer learning networks, when applied properly, may further improve categorization accuracy.