Abstract | ||
---|---|---|
Energy-constrained neural network processing is in high demanded for various mobile applications. Binarized neural network (BNN) aggressively enhances the computational efficiency, and in contrast, it suffers from degradation of accuracy due to its extreme approximation. We propose a neural network model using a new activation function "Delta" based on binarization of differences between weighted-sums. The "Delta" retains the magnitude relation between numerical values, and conveys richer information than ordinary binarization. We can design the hardware architecture for the proposed model with almost the same elements as BNN. The evaluation shows that it achieves higher recognition accuracy than a conventional BNN with almost the same hardware configuration. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1109/ASAP.2019.00-34 | 2019 IEEE 30th International Conference on Application-specific Systems, Architectures and Processors (ASAP) |
Keywords | Field | DocType |
Binary Neural Network (BNN), Activation Function, Differential Operation, Hardware Friendly DNN | Activation function,Computer science,Binary neural network,Parallel computing,Artificial neural network,Computer engineering,Hardware architecture | Conference |
Volume | ISSN | ISBN |
2160-052X | 2160-0511 | 978-1-7281-1602-0 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Yuka Oba | 1 | 3 | 2.82 |
Kota Ando | 2 | 24 | 6.81 |
Tetsuya Asai | 3 | 121 | 26.53 |
Masato Motomura | 4 | 91 | 27.81 |
Shinya Takamaeda-Yamazaki | 5 | 65 | 16.83 |