Image Classification on CIFAR-10 (Accuracy and Emissions)
84.67Top-1 AccuracyMFMO (POS 2)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MFMO (POS 2)Batch Size=102, Optimizer=SGD, Learning Rate=0.0798, Momentum=0.6738, Epochs=91, Weight Bits=7, Activation Integer Bits=29, Activation Fraction Bits=10, Shift Depth=16, Shift Type=Q, Rounding=deterministic, Weight Decay=0.000122024.04 | 84.67 | 0.1673 | |
| MFMO (POS 1)Batch Size=97, Optimizer=Ranger, Learning Rate=0.0847, Momentum=0.5016, Epochs=67, Weight Bits=4, Activation Integer Bits=4, Activation Fraction Bits=4, Shift Depth=5, Shift Type=PS, Rounding=stochastic, Weight Decay=0.00262024.04 | 83.5 | 0.1661 |