BiPSA: an inferential methodology and a computational tool

• US 20070276723A1
• Filed: 04/27/2007
• Published: 11/29/2007
• Est. Priority Date: 04/28/2006
• Status: Abandoned Application

• Current Assignee: Gideon Samid
• Original Assignee: Gideon Samid
• Inventors: Samid, Gideon
• Application Number:
  US11/790,876
• Publication Number:
  US 20070276723A1
• Time in Patent Office: Days
• US Class Current: 705/12
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

BiPSA is a novel inferential methodology characterized by: (1) breaking down all issues of unknown and uncertainty to a cascade of binary questions, (2) identifying all available sources of knowledge, and polling each source individually with respect to each binary question in its turn. Each binary answer is associated with a measure of confidence, and is expressed in a range {N:−N}, where N is a natural number. These answers are integrated through a novel minimum-arbitrariness mathematical operation to an output of the same format, that can be treated as input to a subsequent integration thereby allowing for a construction of a network that is capable of re-configuration, responding to feedback, and hence improving the merit and the credibility of the integrated answer. Useful for various situations challenged by uncertainty and partial knowledge, e.g.: R&D, pattern-recognition, inferential image and data technology, human/machine decision-making, and management procedures.

Claims

1. a method to represent data via a collection of n entities each associated with a “

vote”

expressed as an integer between −

N and +N where N is a natural number, and to reduce (integrate) these votes to a single vote also expressed in the same range, such that the reduction will be helpful for a variety of applications, included, but not limited to inferential challenges, learning, innovation, handling uncertainty, computational tasks, cryptographic primitives, and games, the method comprising

(1.1) a data processing mechanism conducive both to reduction and expansion of information

(1.2) providing a self-organizing, adaptive integration process to integrate the said votes into a reduced summary vote operating via,

(1.3.1) A Unit Integrator complying with the following terms;

(1.3.1.1). single output term.

(1.3.1.2) Permutation invariance.

(1.3.1.3) Symmetry

(1.3.1.4) Monotony

(1.3.1.5) Full-range terms for same sign instances.

(1.3.1.6) Full-range terms for mixed signs instances.

(1.3.1.7) N-invariance, and

(1.3.2) a network comprising threaded unit integrators

(1.4) mapping the integration process in (1) into a form reflective of matrix algebra providing commensurate mathematical computations,

(1.5) reflecting relative impact of n voters by designated weights in the form of natural numbers, such that the value of the weight will correspond to the number of times that vote is counted in a network comprised of Wmax unit integrators, where Wmax is the highest weight designation, and unit integrator i uses as input all the voters with weight designation i or above, feeding the resultant w voted into another unit integrator to produce the final weighted vote.

2. The method in (1) applied to handle uncertainty, exercise learning, and rendering unknown into known, the method comprising (2.1) dividing any uncertainty, or unknown information into a cascade of binary questions, each of which, in turn, is handled via (1) above.

(2.2) unifying all sources of wisdom and knowledge with regard to an issue in question, and integrating them fairly and usefully

(2.3) providing, or identifying a collection of voter entities (human, or data elements) issuing a binary vote over a binary question, and

(2.4) representing the binary vote with a confidence measure expressed via a range of ordinal numbers from −

N (highest confidence negative answer) to +N (highest confidence positive answer), featuring 2N+1 options where N is an arbitrary natural number,

(2.5) integrating the votes via a network that adapts itself via feedback wherein voters exhibiting strong correlation (regular or reverse) with what eventually turns out to be the correct vote are respectively endowed with greater impact on the integrated result, and that impact is commensurate with the voter'"'"'s expressed confidence such that correct high-confidence votes gain more impact on the integrated vote than the lower-confidence votes while incorrect high-confidence votes lose greater impact on the integrated vote than incorrect low-confidence votes, and where integration is otherwise enhanced via genetic algorithms designed to enhance the usefulness of the integrated vote.

3. The method in (1) applied to allowing a group voters associated with impact factors as in (1.5) to rank-order a target group, where the target group may be the same or different from the voters'"'"' group and where the ranking is accomplished by a series of binary questions comparing ranking favorability of two group members at a time, such that these binary questions are answered by the groups of voters according to the method in (1).

4. The method in (1) applied to allowing groups of individuals or organizations to serve their joint goal by providing operational and inferential flexibility between command hierarchy, and complete equality of members, the method comprising of

(4.1) allowing the members of the group to rank order the group members vs. every operational issue, or issue of decision, and

(4.2) using that ranking as impact designators within the integration process as in method (1),

5. The method in (1) applied to providing for a useful conclusion drawn from n sources of knowledge where k factors are identified to impact that conclusion, each in its own way, the method comprising,

(5.1) identifying for each source the degree of association with each conclusion factor, and using that association to govern the integration of the various sources'"'"' opinions, using the method in (1).

6. The method in (1) applied to providing for identifying image irregularities, distortions, and contamination by using the image data as voters over a cascade of binary questions as in (1) such that the answers would determine said irregularity, the method comprising

(6.1) a training process where images with and without the expected irregularities are processed by the method in (1), and the reality check effecting a re-configuration of the integration network as in (1), and

(6.2) a grid-tree, a hierarchy of grids that is superimposed on the image either as a set of Cartesian framework lines, or as polar elements anchored on a single anchor point on the image, or as a grid anchored on two or more points on the image, where each grid-cell is further grid-divided iteratively, and

(6.3) each grid-cell is identified by the ratio of pixels of two reference colors in the cell, mapping such ratio to a range of integers from “

−

N”

to “

+N”

, such that the higher the integer the greater than one the ratio between the first and second reference colors, and

(6.4) using these integer expressed cell contents as reduced expression of the image, where each cell will be regarded as a voter, and its contents will determine its vote on any binary component of the irregularity question of interest, where such determination is based on the probability of each cell contents figure to be found in an image with irregularity as opposed to images without irregularities, and where

(6.5) the cells data is aggregated to cell groups either comprehensively where n cells define 2ⁿ data elements, or alternatively using a genetic algorithm to couple effective cells into new voters, and thus re-configuring the network.