no code implementations • 15 Mar 2022 • Dongseok Im, Gwangtae Park, Zhiyong Li, Junha Ryu, Hoi-jun Yoo
This paper proposes energy-efficient signed bit-slice architecture which accelerates both high-precision and dense DNNs by exploiting a large number of zero values of signed bit-slices.
no code implementations • 24 Jan 2021 • JuHyoung Lee, Sangyeob Kim, Sangjin Kim, Wooyoung Jo, Hoi-jun Yoo
Deep reinforcement learning (DRL) has shown remarkable success in sequential decision-making problems but suffers from a long training time to obtain such good performance.
no code implementations • 23 Jun 2020 • Donghyeon Han, Gwangtae Park, Junha Ryu, Hoi-jun Yoo
Even though the DFA algorithm is biologically plausible and has a potential of high-speed training, it has not been considered as the substitute for back-propagation (BP) due to the low accuracy in the CNN and RNN training.
no code implementations • ICLR 2019 • Donghyeon Han, Hoi-jun Yoo
One of them was direct feedback alignment (DFA), but it showed low training performance especially for the convolutional neural network (CNN).