Quasar-ViT: Hardware-Oriented Quantization-Aware Architecture Search for Vision Transformers

Keyphrases

• Vision Transformer 100%
• Quasars 100%
• Architecture Search 100%
• Supernet 100%
• Hardware Efficiency 100%
• Inference Speed 66%
• Entanglement 33%
• Adaptive Design 33%
• Edge Devices 33%
• Efficient Hardware 33%
• Quantization Scheme 33%
• Hardware Implementation 33%
• Limited Resources 33%
• Frames per Second 33%
• Xilinx 33%
• Theoretical Computation 33%
• Computation-intensive 33%
• Scaling Technique 33%
• Efficient Deployment 33%
• Convolutional Neural Network 33%
• FPGA Platform 33%
• Computational Reduction 33%
• Model Adaptive 33%
• Computer Vision Tasks 33%
• Mixed Precision 33%
• Superior Accuracy 33%
• ImageNet Dataset 33%
• Inference Latency 33%
• ZCU102 33%
• Oriented Search Algorithm 33%
• Latency Modeling 33%
• Practical Inference 33%
• Resource Modelling 33%
• Mixed Precision Quantization 33%

Computer Science

• Vision Transformer 100%
• Computer Hardware 100%
• Field Programmable Gate Arrays 50%
• Hardware Implementation 25%
• Intensive Computation 25%
• Scaling Technique 25%
• Searching Algorithm 25%
• Convolutional Neural Network 25%
• Computer Vision Task 25%
• Subnet 25%