NeuLens: Spatial-based Dynamic Acceleration of Convolutional Neural Networks on Edge

Keyphrases

• Latency 100%
• Microflow 100%
• Convolutional Neural Network 100%
• Dynamic Acceleration 100%
• Convolutional Neural Network Accelerator 66%
• Supernet 66%
• Region-aware 66%
• Accuracy Improvement 33%
• Computational Cost 33%
• Mobile Platform 33%
• Edge Computing System 33%
• Single Model 33%
• Computing Platform 33%
• Vision-based 33%
• Application Requirements 33%
• Object Detection 33%
• Convolutional Neural Network Model 33%
• State-of-the-art Techniques 33%
• Object Classification 33%
• Spatial Channel 33%
• Redundant Operation 33%
• Memory Constraints 33%
• Inference Accuracy 33%
• Accuracy Loss 33%
• Mobile Computing Systems 33%
• Acceleration Framework 33%
• Spatial Redundancy 33%
• Dynamic Inference 33%
• Hardware Level 33%
• Content Requirements 33%
• Inference Mechanisms 33%
• Operation Reduction 33%
• Edge Platform 33%
• GPU 33%
• Latency Reduction 33%
• Neural Network Inference 33%
• Computing Latency 33%

Computer Science

• Convolutional Neural Network 100%
• Speed-up 50%
• Convolution 33%
• Mobile Computing 16%
• Computational Cost 16%
• Computing Platform 16%
• Edge Computing 16%
• Memory Latency 16%
• Neural Network Model 16%
• Application Requirement 16%
• Reduction Operation 16%
• Network Inference 16%
• Baseline Method 16%
• Latency Reduction 16%
• Critical Region 16%
• Spatial Redundancy 16%
• Hardware Level 16%
• Graphics Processing Unit 16%