4 models have been trained on the Cifar10 dataset. Each model has some improvements wrt the previous one to make it faster on the GAP9 architecture.
Complete the following table:
| Model | Float Accuracy | Quant Accuracy | Ops | Parameters | Coeff Size deployment* | Cyc | Op/Cyc | Why is this better than previous? |
|---|---|---|---|---|---|---|---|---|
| v1 | ||||||||
| v2 | ||||||||
| v3 | ||||||||
| v4 |
*Size of the NN coefficients to deploy (flash usage).
Apply optimizations to the deployment scripts and make the previous table faster.
Make the best possible model (the fastest / energy efficient) with an accuracy of at least 50% on the cifar small (the one used in the deployment script) testset. You can train from scratch a new model, or play with NNTool optimizations.
Best model wins a prize :)