Representable Matrices: Enabling High Accuracy Analog Computation for Inference of DNNs using Memristors

Keyphrases

• Deep Neural Network 100%
• Memristor 100%
• Analog Computing 100%
• Representable 100%
• Conductance 60%
• Scale Factor 60%
• Input Vector 40%
• Memristor Crossbar 40%
• State-of-the-art Techniques 20%
• Peripheral Circuits 20%
• Range of Values 20%
• Non-ideality 20%
• Mapping Method 20%
• Convolutional Neural Network 20%
• Classification Accuracy 20%
• Device Characteristics 20%
• Promising Solutions 20%
• Two-input 20%
• Computational Accuracy 20%
• Ranging Error 20%
• CIFAR-10 20%
• Arbitrary Matrix 20%
• Seven Layers 20%
• Conductance Matrix 20%
• Balanced Accuracy 20%

Engineering

• Scaling Factor 100%
• Deep Neural Network 100%
• Input Vector 66%
• Fundamental Problem 33%
• State Variable 33%
• State-of-the-Art Technique 33%
• Classification Accuracy 33%
• Convolutional Neural Network 33%

Computer Science

• Deep Neural Network 100%
• Memristor 100%
• Scaling Factor 42%
• Conductance Value 28%
• Classification Accuracy 14%
• Fundamental Problem 14%
• State Variable 14%
• Convolutional Neural Network 14%