Abstract
Recently, we proposed to transform the outputs of each hidden neuron in a multilayer perceptron network to have zero output and zero slope on average, and use separate shortcut connections to model the linear dependencies instead. We continue the work by firstly introducing a third transformation to normalize the scale of the outputs of each hidden neuron, and secondly by analyzing the connections to second order optimization methods. We show that the transformations make a simple stochastic gradient behave closer to second-order optimization methods and thus speed up learning. This is shown both in theory and with experiments. The experiments on the third transformation show that while it further increases the speed of learning, it can also hurt performance by converging to a worse local optimum, where both the inputs and outputs of many hidden neurons are close to zero.
Original language | English (US) |
---|---|
State | Published - 2013 |
Event | 1st International Conference on Learning Representations, ICLR 2013 - Scottsdale, United States Duration: May 2 2013 → May 4 2013 |
Conference
Conference | 1st International Conference on Learning Representations, ICLR 2013 |
---|---|
Country/Territory | United States |
City | Scottsdale |
Period | 5/2/13 → 5/4/13 |
ASJC Scopus subject areas
- Education
- Computer Science Applications
- Linguistics and Language
- Language and Linguistics