BRIAN: a basic regimen for intelligent analysis using networks

• US 8,768,869 B1
• Filed: 04/19/2012
• Issued: 07/01/2014
• Est. Priority Date: 04/19/2012
• Status: Active Grant

• Current Assignee: the united states of america as represented by the secretary of the navy
• Original Assignee: the united states of america as represented by the secretary of the navy
• Inventors: Rubin, Stuart Harvey
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Zidanic, Michael
• Application Number:
  US13/451,348
• Time in Patent Office: 803 Days
• Field of Search: 706/12
• US Class Current: 706/12
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

A method for machine-learning by analogy, comprising: providing a network of component computer systems, each component having sequential layers, each layer having parallel segments, each segment comprises a processor and a memory store, each memory store is configured to store a domain-specific case base, which is defined by a situation-action pair; stochastically transforming cases in each case base through automatic deterministic generalization and analogy when the corresponding processor is in a dream mode to create transformed cases; providing a user-entered situation; modifying the user-entered situation by expanding its contextual mnemonics; searching each case base for cases and transformed cases that include contextual subsets of the modified, user-entered situation; for a given case base, mapping the modified, user-entered situation to a matched action within the given case base; creating a new case comprising the user-entered situation and the matched action.

Claims

1. A method for creative machine-learning by analogy, the method comprising the following steps:

providing a network of component computer systems, each component comprises a plurality of sequential layers, each layer comprises a plurality of parallel segments, each segment comprises a processor and a memory store, each memory store is configured to store a domain-specific case base, wherein each case is defined by a situation-action pair;

stochastically transforming cases in each case base through automatic deterministic generalization and analogy when the corresponding processor is in a dream mode to create transformed cases, and selecting pairs of cases for generalization and analogy with a 3-2-1 skew during the transforming;

providing a user-entered situation, wherein the situations are sets of parametized natural language questions, and wherein each situation comprises a vertically-stacked, AND-operator-connected set of questions;

modifying the user-entered situation by expanding its contextual mnemonics;

searching each case base for cases and transformed cases that include contextual subsets of the modified, user-entered situation;

for a given case base, mapping the modified, user-entered situation to a matched action within the given case base, wherein the actions are sequences of parametized natural language statements, and wherein each action comprises a vertically-stacked sequence of action statements;

using the 3-2-1 skew to independently find candidate macro situations and macro actions from within the most-frequently-used (MFU) segments; and

using the 3-2-1 skew to iteratively replace macros within each segment with a corresponding macro definition when found during dream mode;

creating a new case comprising the user-entered situation and the matched action, wherein the 3-2-1 skew moves newly acquired cases to a logical head of a case base and progressively and stochastically selects from the logical head until the 3-2-1 skew uniformly selects from the entire case base; and

calculating a possibility for a fired action in a static domain by using the heuristic

2. The method of claim 1, further comprising mapping the modified, user-entered situation to a new action if no match can be found in a given case base.

3. The method of claim 1, wherein one of the situations comprises an illustration.

4. The method of claim 1, further comprising determining a sequence of applied transforms using a k-limited transformation such that the number of transformations in any derivation are limited by k so the method does not go into infinite cycles.