Differentiable Model Compression via Pseudo Quantization Noise.

Keyphrases

• Model Accuracy 100%
• Number of Bits 100%
• Quantization Noise 100%
• Differentiable Model 100%
• Model Compression 100%
• Hyperparameters 50%
• Image Classification 50%
• Compress 50%
• Audio Source Separation 50%
• Language Modeling 50%
• Gradient Approximation 50%
• Transformer Model 50%
• Model Size 50%
• ImageNet Dataset 50%
• Individual Weight 50%
• Bit Precision 50%
• Quantization Technique 50%
• Straight-through Estimator 50%
• Quantized Model 50%

Computer Science

• Model Compression 100%
• Quantization Noise 100%
• Approximation (Algorithm) 50%
• Image Classification 50%
• Language Modeling 50%
• Model Accuracy 50%
• Audio Source Separation 50%