Speedup from a different parametrization within the Neural Network algorithm
This work addresses training efficiency for neural network practitioners, but it appears incremental as it modifies an existing parametrization without introducing a new paradigm.
The paper tackled the problem of slow training in neural networks by proposing a different parametrization of hyperplanes, which significantly outperformed the usual method on autoencoder examples, achieving lower training errors with fewer epochs.
A different parametrization of the hyperplanes is used in the neural network algorithm. As demonstrated on several autoencoder examples it significantly outperforms the usual parametrization, reaching lower training error values with only a fraction of the number of epochs. It's argued that it makes it easier to understand and initialize the parameters.