Towards efficient full 8-bit integer DNN online training on resource-limited devices without batch normalization

Computer Science

• Deep Neural Network 100%
• Batch Normalization 100%
• Quantization (Signal Processing) 40%
• Computational Cost 30%
• Processing Speed 30%
• Floating Point 20%
• Deep Learning 10%
• Performance Model 10%
• Residual Neural Network 10%
• Inference Process 10%
• Power Consumption 10%
• Root Operation 10%
• Convolution Operation 10%

Chemical Engineering

• Deep Neural Network 100%
• Deep Learning 33%

Engineering

• Bit Floating Point 40%
• Isometry 20%