The Efficiency and the Robustness of Natural Gradient Descent Learning Rule
نویسندگان
چکیده
The inverse of the Fisher information matrix is used in the natural gradient descent algorithm to train single-layer and multi-layer perceptrons. We have discovered a new scheme to represent the Fisher information matrix of a stochastic multi-layer perceptron. Based on this scheme, we have designed an algorithm to compute the natural gradient. When the input dimension n is much larger than the number of hidden neurons, the complexity of this algorithm is of order O(n). It is confirmed by simulations that the natural gradient descent learning rule is not only efficient but also robust.
منابع مشابه
Nips*97 the Eeciency and the Robustness of Natural Gradient Descent Learning Rule Sub-category: Dynamics of Learning Algorithms Category: Theory
We have discovered a new scheme to represent the Fisher information matrix of a stochastic multi-layer perceptron. Based on this scheme, we have designed an algorithm to compute the inverse of the Fisher information matrix. When the input dimension n is much larger than the number of hidden neurons, the complexity of this algorithm is of order O(n 2) while the complexity of conventional algorit...
متن کاملExtensions of the Hestenes-Stiefel and Polak-Ribiere-Polyak conjugate gradient methods with sufficient descent property
Using search directions of a recent class of three--term conjugate gradient methods, modified versions of the Hestenes-Stiefel and Polak-Ribiere-Polyak methods are proposed which satisfy the sufficient descent condition. The methods are shown to be globally convergent when the line search fulfills the (strong) Wolfe conditions. Numerical experiments are done on a set of CUTEr unconstrained opti...
متن کاملMulti-robot Reinforcement Learning Based On Learning Classifier System with Gradient Descent Methods
This paper proposed a robot reinforcement learning method based on learning classifier system. A Learning Classifier System is a accuracy-based machine learning system with gradient descent that combines reinforcement learning and rule discovery system. The genetic algorithm and the covering operator act as innovation discovery components which are responsible for discovering new better reinfor...
متن کاملPosition Control of a Pulse Width Modulated Pneumatic Systems: an Experimental Comparison
In this study, a new adaptive controller is proposed for position control of pneumatic systems. Difficulties associated with the mathematical model of the system in addition to the instability caused by Pulse Width Modulation (PWM) in the learning-based controllers using gradient descent, motivate the development of a new approach for PWM pneumatics. In this study, two modified Feedback Error L...
متن کاملMatrix momentum for practical natural gradient learning
An on-line learning rule, based on the introduction of a matrix momentum term, is presented, aimed at alleviating the computational costs of standard natural gradient learning. The new rule, natural gradient matrix momentum, is analysed in the case of two-layer feed-forward neural network learning viamethods of statistical physics. It appears to provide a practical algorithm that performs as we...
متن کامل