TY - GEN
T1 - Circuit Optimization of Ternary Sparse Neural Net
AU - Megumi, Taichi
AU - Kawahara, Takayuki
N1 - Publisher Copyright:
© 2024 IEEE.
PY - 2024
Y1 - 2024
N2 - In the coming years, AIoT, which combines AI and IoT, is expected to become a reality. To realize it, an AI chip with small size, high speed, and high recognition accuracy is required. Binarized Neural Networks (BNNs), a binarization method, are known for miniaturization. Ternary Sparse XNOR-Net (TSXN), a ternary method, is expected to significantly reduce the circuit size while improving recognition accuracy compared with BNN. In this study, we proposed a further circuit optimization method for TSXN and evaluated its implementation on an FPGA. Compared with BNN, the proposed method reduced the circuit size by 81.0% and increased the operation speed by 33.3%, while improving the recognition accuracy by 3.7%.
AB - In the coming years, AIoT, which combines AI and IoT, is expected to become a reality. To realize it, an AI chip with small size, high speed, and high recognition accuracy is required. Binarized Neural Networks (BNNs), a binarization method, are known for miniaturization. Ternary Sparse XNOR-Net (TSXN), a ternary method, is expected to significantly reduce the circuit size while improving recognition accuracy compared with BNN. In this study, we proposed a further circuit optimization method for TSXN and evaluated its implementation on an FPGA. Compared with BNN, the proposed method reduced the circuit size by 81.0% and increased the operation speed by 33.3%, while improving the recognition accuracy by 3.7%.
KW - Deep Learning
KW - FPGA
KW - Neural Network
KW - Sparse
KW - Ternary
UR - http://www.scopus.com/inward/record.url?scp=85186765024&partnerID=8YFLogxK
U2 - 10.1109/SAMI60510.2024.10432809
DO - 10.1109/SAMI60510.2024.10432809
M3 - Conference contribution
AN - SCOPUS:85186765024
T3 - 2024 IEEE 22nd World Symposium on Applied Machine Intelligence and Informatics, SAMI 2024 - Proceedings
SP - 53
EP - 58
BT - 2024 IEEE 22nd World Symposium on Applied Machine Intelligence and Informatics, SAMI 2024 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2024 IEEE 22nd World Symposium on Applied Machine Intelligence and Informatics, SAMI 2024
Y2 - 25 January 2024 through 27 January 2024
ER -