Presentation
WinoTrain: Winograd-Aware Training for Accurate Full 8-bit Convolution Acceleration
DescriptionEfficient inference is critical in realizing a real-time implementation of convolutional neural networks (CNNs) on compute and memory-constrained embedded platforms. Using quantization techniques and fast convolutional algorithms like Winograd, CNN inference can achieve benefits in latency and in energy consumption. While previous works exclusively investigate quantization methods or Winograd algorithms, combining them introduces numerical instability. We address this by proposing a Winograd and Quantization aware training scheme and an FPGA design employing tiling and unrolling methods to highlight the performance benefits of using Winograd algorithm. We achieved 1.48x reduction in inference time with minimal accuracy degradation for ResNet18 model.
Event Type
Research Manuscript
TimeWednesday, July 12th10:55am - 11:10am PDT
Location3004, 3rd Floor
AI
AI/ML Algorithms