Keywords:Deep learning, Incremental development
Software development with deep learning has been spreading recently. However, in conventional software development methods with deep learning, generating models that meet the requirements often requires trial and error by software developers. With such ad hoc development methods, it is difficult to develop models efficiently and consistently. In this thesis, the author proposes a feature-based design method of learning model for deep learning. The proposed method controls learning of the model by selecting training data based on features iteratively. The author evaluates the accuracy and the speed of convergence in the proposed method by comparing it to the conventional method. From the evaluation results, the author confirmed that the proposed method improves the controllability of learning and the accuracy of the learning model. It can be expected that machine learning software developers can efficiently and consistently develop models that meet the requirements.
Authentication for paper PDF access
A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.