Enhanced interface for a neural network engine

• US 5,361,326 A
• Filed: 12/31/1991
• Issued: 11/01/1994
• Est. Priority Date: 12/31/1991
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Miller, Patrice C., Aparicio, Manuel IV, Miller, Wade A.
• Primary Examiner(s): Downs, Robert W.
• Application Number:
  US07/816,460
• Time in Patent Office: 1,036 Days
• Field of Search: 395/22, 395/76
• US Class Current: 706/44
• CPC Class Codes: G06N 3/105 Shells for specifying net layout

Abstract

An interface for a neural network includes a generalized data translator and a certainty filter in the data path including the neural network for rendering a decision on raw data, possibly from a data processing application. The data translator is controlled with user-definable parameters and procedures contained in a property list in order to manipulate translation, truncation, mapping (including weighting) and other transformations of the raw data. The neuron to which the output of the data translator is applied is controlled by a code index contained in an action list. An external certainty threshold is also provided, preferably by the action list to filter the output of the neural network. The core program used with the ConExNS neurons for system maintenance also includes further core operations and size maintenance operations responsive to commands from the user of an application to cause operations to be performed with in the neural network as well as to create and update the property and action lists.

Claims

1. A method of interfacing between a data processing application and a neural network including the steps of

providing at least one label by said data processing application to an interface arrangement, wherein said label corresponds to an action which can be taken on an object of said application,

selecting, in accordance with said label, a neuron of said neural circuit from a list, said list containing at least an index of said neuron, for application of a value of said at least one label in response to said providing step, wherein said index of said neuron corresponds to said action, and

executing at least one command by said neural network in response to at least said value of said at least one label.

2. A method as recited in claim 1, wherein said command is defined by a user by providing an entry in a list in accordance with a label.

3. A method as recited in claim 2, wherein said label corresponds to an action which can be performed on a object in said data processing application.

4. A method as recited in claim 2, including the further step of deleting an entry in said list.

5. A method as recited in claim 2, including the further step of adding an entry to said list.

6. A method of interfacing between a data processing application and a neural network including the steps of

providing at least one label by said data processing application to an interface arrangement, wherein said label corresponds to an action which may be taken on an object of said application,

selecting, in accordance with said label, a neuron of said neural network, from at least one list accessible in accordance with said label, said at least one list containing at least an index of said neuron corresponding to said action, and a certainty threshold value and procedure including at least one of a translation, truncation and mapping of at least a value corresponding to said at least one label provided by said application,

performing said at least one of a translation, truncation and mapping of at least said value corresponding to said at least one label to develop bits,

applying bits developed by said performing step to predetermined inputs of said neural network, and

filtering an output of said neuron in accordance with a certainty threshold selected in said selecting step.

7. A method as recited in claim 6, wherein said procedure is defined by a user by providing an entry in a list in accordance with a label.

8. A method as recited in claim 7, wherein a property corresponds to said label.

9. A method as recited in claim 6, including the further step of deleting an entry in said list.

10. A method as recited in claim 6, including the further step of adding an entry to said list.

11. A method of operating a neural network including the steps of

obtaining an output and a certainty value from a neuron in said neural network, said neuron selected from a list containing at least an index of a neuron of said neural network in accordance with a label provided by an application, said label corresponding to an object on which an action may be taken an wherein said index of a neuron identifies a neuron corresponding to said action,

obtaining a certainty threshold value in response to a label in a data processing application, and

filtering said output of said neuron in response to a comparison of said certainty value and said certainty threshold value.

12. A method as recited in claim 11, including the further step of specification of said certainty threshold value by a user.

13. A method as recited in claim 11, wherein said certainty threshold value is obtained from a list in accordance with said label.

14. An interface arrangement between a data processing application and a neural network including

means for providing at least one label by said data processing application to said interface arrangement,

at least one list accessible in accordance with said label, said label corresponding to an object in said application on which an action may be taken, said list containing at least an index of a neuron in said neural network corresponding to said action,

means for selecting a confidence threshold and at least one of a procedure and an action, said procedure including at least one of a translation, truncation and mapping of at least a value corresponding to said at least one label provided by said application,

means for performing said at least one of an action based on at least one value corresponding to said at least one label and a translation, truncation and mapping of at least a value corresponding to said at least one label,

means for applying said at least one value corresponding to said label to predetermined inputs of said neural network, and

means for filtering an output of said neural network in accordance with said certainty threshold corresponding to said label.

15. An interface as recited in claim 14, wherein said label further corresponds to a property of an object and said list contains data indicating correspondence between an output of at least one sensor and an input of said neural network.

16. An interface as recited in claim 14, wherein said list contains a certainty threshold value corresponding to said action.

17. An interface as recited in claim 14, further including

means for changing said at least one list by at least one of adding and deleting entries therein.

18. An interface as recited in claim 17, wherein said means for changing said at least one list is controlled by a user.