CNN-MERP: An FPGA-Based Memory-Efficient Reconfigurable Processor for Forward and Backward Propagation of Convolutional Neural Networks.
Xushen HanDajiang ZhouShihao WangShinji KimuraPublished in: CoRR (2017)
Keyphrases
- memory efficient
- forward and backward
- convolutional neural networks
- hardware implementation
- field programmable gate array
- digital signal
- convolutional network
- video processing
- cellular neural networks
- smart camera
- systolic array
- multithreading
- parallel architecture
- external memory
- parallel processing
- dynamic programming
- functional units
- embedded systems
- greedy search
- low cost
- processing elements
- hardware architecture
- iterative deepening
- application specific
- general purpose
- parallel computing
- combinatorial optimization
- objective function
- genetic algorithm
- data structure
- integral image
- search algorithm
- image processing algorithms