CIFAR-10 IOC-NN (PyTorch)

Architecture: 3072 -> 128 -> 64 -> 10; ELU on the two hidden layers, no activation on the output (logits)
Convexity: the first layer is unconstrained; the weights of all later layers are kept non-negative (negative values are replaced by exp(w - 5) after every optimiser step)
Parameters: 402,250
Preprocessing: flatten the 32x32x3 image to 3072 values, divide by 255, then standardise each column with the train-set mean and std stored in preprocessing.npz (columns with std 0 use std 1)
Loading: m = IOCNN(MLP()); m.load_state_dict(torch.load("model_ioc_p0.pt"))
Test accuracy: 0.4841

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support