Abstract | ||
---|---|---|
This article extends neural networks to the case of an uncountable number of hidden units, in several ways. In the rst approach proposed, a nite parametrization is possi- ble, allowing gradient-based learning. While having the same number of parameters as an ordinary neural network, its internal struc- ture suggests that it can represent some smooth functions much more compactly. Un- der mild assumptions, we also nd better er- ror bounds than with ordinary neural net- works. Furthermore, this parametrization may help reducing the problem of satura- tion of the neurons. In a second approach, the input-to-hidden weights are fully non- parametric, yielding a kernel machine for which we demonstrate a simple kernel for- mula. Interestingly, the resulting kernel ma- chine can be made hyperparameter-free and still generalizes in spite of an absence of ex- plicit regularization. |
Year | Venue | Keywords |
---|---|---|
2007 | AISTATS | neural network |
Field | DocType | Volume |
Mathematical optimization,Physical neural network,Computer science,Stochastic neural network,Algorithm,Recurrent neural network,Probabilistic neural network,Time delay neural network,Types of artificial neural networks,Kernel method,Artificial neural network | Journal | 2 |
Citations | PageRank | References |
4 | 0.42 | 1 |
Authors | ||
4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Nicolas Le Roux | 1 | 1684 | 145.19 |
universite de montreal | 2 | 15 | 3.20 |
montreal quebec | 3 | 4 | 0.42 |
Yoshua Bengio | 4 | 42677 | 3039.83 |