Neural networks: Efficient implementations and applications

Abstract

This paper serves as an overview to the special session “Neural Networks: Efficient Implementations and Applications” on IEEE International Conference on ASIC (ASICON) 2017. Focusing on the state-of-the-art research progresses on neural networks, our introduction consists of two main parts, namely efficient hardware implementations and the latest applications. Deep neural network (DNN) provides superior performance for complex tasks, which pushes one step further in many fields. However, in the development of DNN, a lot of challenges emerge and remain to be settled. In this overview paper, latest research progress of neural network in both implementation and application will be summarized to open the discussion on some important to-be-solved problems and limitations, especially from the circuits-and-systems design perspective. This motivates the organization of this special session, which includes this overview paper and four technical papers. An outline of these four invited papers constructs the last part of this overview, sketching their targeted research problems, proposed methodologies and solutions, as well as the main contribution and results.