Method for supervised teaching of a recurrent artificial neural network
First Claim
1. A method for constructing a discrete-time recurrent neural network and training it in order to minimize its output error, comprising;
- constructing a recurrent neural network as a reservoir for excitable dynamics (dynamical reservoir network;
providing means of feeding input to the dynamical reservoir network;
attaching output units to the dynamical reservoir network through weighted connections; and
training the weights of the connections only from the dynamical reservoir network to the output units in a supervised training scheme.
1 Assignment
0 Petitions
Accused Products
Abstract
A method for the supervised teaching of a recurrent neutral network (RNN) is disclosed. A typical embodiment of the method utilizes a large (50 units or more), randomly initialized RNN with a globally stable dynamics. During the training period, the output units of this RNN are teacher-forced to follow the desired output signal. During this period, activations from all hidden units are recorded. At the end of the teaching period, these recorded data are used as input for a method which computes new weights of those connections that feed into the output units. The method is distinguished from existing training methods for RNNs through the following characteristics: (1) Only the weights of connections to output units are changed by learning—existing methods for teaching recurrent networks adjust all network weights. (2) The internal dynamics of large networks are used as a “reservoir” of dynamical components which are not changed, but only newly combined by the learning procedure—existing methods use small networks, whose internal dynamics are themselves completely re-shaped through learning.
77 Citations
41 Claims
-
1. A method for constructing a discrete-time recurrent neural network and training it in order to minimize its output error, comprising;
-
constructing a recurrent neural network as a reservoir for excitable dynamics (dynamical reservoir network; providing means of feeding input to the dynamical reservoir network; attaching output units to the dynamical reservoir network through weighted connections; and training the weights of the connections only from the dynamical reservoir network to the output units in a supervised training scheme. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38)
-
-
39. A neural network for constructing a discrete-time recurrent neural network and training it in order to minimize its output error comprising;
-
a recurrent neural network as a reservoir for excitable dynamics dynamical reservoir network; means for feeding input to the dynamical reservoir network; output units attached to the dynamical reservoir network through weighted connections; and wherein the weights of the connections are trained only from the dynamical reservoir network to the output units in a supervised training scheme. - View Dependent Claims (40, 41)
-
Specification