A 21mW low-power recurrent neural network accelerator with quantization tables for embedded deep learning applications.
Jinmook LeeDongjoo ShinHoi-Jun YooPublished in: A-SSCC (2017)
Keyphrases
- low power
- deep learning
- recurrent neural networks
- power consumption
- low cost
- feed forward
- unsupervised learning
- unsupervised feature learning
- neural network
- complex valued
- recurrent networks
- high speed
- low power consumption
- single chip
- embedded systems
- mental models
- machine learning
- artificial neural networks
- weakly supervised
- image processing
- real time
- mixed signal
- logic circuits
- power dissipation
- hidden layer
- cmos technology
- genetic algorithm
- computer vision
- denoising