The following layers have been implemented
- Conv
- Fully connected
- Activation (Relu, sigmoid)
- Softmax
- Using Upsampled mnist dataset.
- Working offload/prefetch mechanism with min-heap implementation.
Results for main224
- 5 epochs
- 0.01 learning rate
- 1000 train
- 500 test
Batch Size | Val_acc | Test_acc |
---|---|---|
128 | 106 | 54 |
256 | NEM | NEM |
NEM - Not enough memory