Showing 1–1 of 1 results for author: Vila, O P
-
Solving internal covariate shift in deep learning with linked neurons
Authors:
Carles Roger Riera Molina,
Oriol Pujol Vila
Abstract:
This work proposes a novel solution to the problem of internal covariate shift and dying neurons using the concept of linked neurons. We define the neuron linkage in terms of two constraints: first, all neuron activations in the linkage must have the same operating point. That is to say, all of them share input weights. Secondly, a set of neurons is linked if and only if there is at least one memb…
▽ More
This work proposes a novel solution to the problem of internal covariate shift and dying neurons using the concept of linked neurons. We define the neuron linkage in terms of two constraints: first, all neuron activations in the linkage must have the same operating point. That is to say, all of them share input weights. Secondly, a set of neurons is linked if and only if there is at least one member of the linkage that has a non-zero gradient in regard to the input of the activation function. This means that for any input in the activation function, there is at least one member of the linkage that operates in a non-flat and non-zero area. This simple change has profound implications in the network learning dynamics. In this article we explore the consequences of this proposal and show that by using this kind of units, internal covariate shift is implicitly solved. As a result of this, the use of linked neurons allows to train arbitrarily large networks without any architectural or algorithmic trick, effectively removing the need of using re-normalization schemes such as Batch Normalization, which leads to halving the required training time. It also solves the problem of the need for standarized input data. Results show that the units using the linkage not only do effectively solve the aforementioned problems, but are also a competitive alternative with respect to state-of-the-art with very promising results.
△ Less
Submitted 7 December, 2017;
originally announced December 2017.
