PROCESSOR NODE, ARTIFICIAL NEURAL NETWORK AND METHOD OF OPERATION OF AN ARTIFICIAL NEURAL NETWORK

• US 20120215728A1
• Filed: 01/11/2012
• Published: 08/23/2012
• Est. Priority Date: 01/12/2011
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: ISAIADIS, Stavros
• Granted Patent: US 8,892,486 B2
• Time in Patent Office: Days
• US Class Current: 706/25
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

There is provided a temporal processor node for use as an input node in the input layer of a class network in an artificial neural network, the class network being operable to generate an output signal based on a network input vector component received by the input layer, the temporal processor node being operable to receive observation data representing the observed state of a monitored entity as a component of the network input vector. The temporal processor node comprises a memory module operable to store a most recently observed state of the monitored entity in the memory module as a current state, a modification module having a timer, the timer being operable to output a value representing time elapsed since observation of the current state, the modification module being operable to modify the current state with a modification factor dependent on the value output by the timer, wherein when triggered, the temporal processor node is operable to output the modified current state as a representation of the current state.

Claims

1. A temporal processor node for use as an input node in the input layer of a class network in an artificial neural network, the class network being operable to generate an output signal based on a network input vector component received by the input layer, the temporal processor node being operable to receive observation data representing the observed state of a monitored entity as a component of the network input vector, the temporal processor node comprising:

a memory module operable to store a most recently observed state of the monitored entity in the memory module as a current state;

a modification module having a timer, the timer being operable to output a value representing time elapsed since observation of the current state, the modification module being operable to modify the current state with a modification factor dependent on the value output by the timer;

wherein

when triggered, the temporal processor node is operable to output the modified current state as a representation of the current state.

2. An artificial neural network having a class network, the class network operable to generate an output signal based on a network input vector received by an input layer of the class network, the class network comprising:

a plurality of input nodes forming the input layer, each input node having a memory module and being operable receive observation data representing the observed state of a monitored entity as a component of the network input vector, to store a most recently observed state of the monitored entity in the memory module as a current state, and, when triggered, to output a representation of the current state;

wherein

one or more of the plurality of input nodes is a temporal processor node according to claim 1.

3. The artificial neural network according to claim 2, wherein

the class network is operable to function in a training mode, in which

the input layer is operable to receive a training input vector as a network input vector with sample states of monitored entities as components for treatment as observed states of monitored entities in the input nodes, and

the modification factor for each of a plurality of values of elapsed time is set for each temporal processor node.

4. The artificial neural network according to claim 3, wherein each time a training input vector is received:

the training procedure is performed with the timers of the one or more temporal processor nodes outputting an elapsed time of zero;

and for each temporal processor node, the training procedure is performed with the timer of the temporal processor node outputting each of a plurality of values of elapsed time.

5. The artificial neural network according to claim 4, wherein

the class network is operable to function in a recognition mode, in which

when a component of the network input vector changes, the plurality of input nodes forming the input layer are triggered and are each operable to output a representation of the current state, and

if the input node at which the received component of the network input vector has changed is a temporal processor node, the timer of that temporal processor node is reset.

6. The artificial neural network according to claim 5, the class network further comprising:

a processing layer including a plurality of processing nodes operable to receive one or more representations of current states from the input layer, and to combine and modify the received representations to output as a processed signal; and

an output layer, operable to receive processed signals from one or more processing nodes and to combine and modify the received processed signals to output as the output signal.

7. The artificial neural network according to claim 6, wherein the monitored entities whose states are observed belong to a system, and the class network is operable to generate a decision on whether or not a particular problem has occurred in the system based on the network input vector.

8. The artificial neural network according to claim 7, further comprising one or more additional class networks, each additional class network being operable to generate a decision on whether or not a different particular problem has occurred in the same system based on a network input vector.

9. A method for generating an output signal from a network input vector received by an input layer in a class network forming part of an artificial neural network, the class network comprising a plurality of input nodes forming the input layer, each input node having a memory module, wherein one or more of the input nodes are temporal processor nodes, each comprising a modification module having a timer,

the method comprising:

at each of the input nodes, receiving observation data representing the observed state of a monitored entity as a component of the network input vector, and storing a most recently observed state of the monitored entity in the memory module as a current state, and, when triggered, outputting a representation of the current state; and

at each of the temporal processing nodes, the method further comprises;

outputting a value from the timer representing time elapsed since observation of the current state;

modifying the current state with a modification factor dependent on the value output by the timer; and

when triggered, outputting the modified current state as the representation of the current state.

10. The method according to claim 9, further comprising, in a training mode of the class network:

inputting to the input layer as a network input vector, a training input vector with sample states of monitored entities for treatment as observed states of monitored entities in the input nodes;

for each training input vector, resetting each of the timers when the training sample is received, and for each of a plurality of values of elapsed time, performing a training procedure to define a modification factor for each temporal processor node.

11. The method according to claim 10, further comprising, in a recognition mode of the class network:

when a component of the network input vector changes, triggering each of the plurality of input nodes forming the input layer to output a representation of the current state, and

if the input node at which the received component of the network input vector has changed is a temporal processor node, resetting the timer.