An approach to deep learning service provision with elastic remote interfaces

Abstract

Deep learning has been widely applied for computer vision, natural language processing, and information retrieval etc. Using a deep learning framework can reduce learning curve of beginners facilitating them to get involved with deep learning algorithms. Current deep learning frameworks can mainly be divided into traditional local deployment and cloud-based platforms. However, the two forms cannot be considered at the same time in terms of debugging and remote access. This paper focuses on the logical isolation between deep learning algorithm design and actual business execution, and it proposes an elastic framework that can resolve the contradiction between internal improvement and external access, which can improve the efficiency of both algorithm design researchers and business requirements department engineers.