Adaptive Outlier Analysis for Crowd Intelligence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for amplifying the intelligence of crowds and swarms face challenges in identifying insightful participants without historical data, leading to inaccurate predictions and forecasts.
Innovation Solution
The system performs Adaptive Outlier Analysis by querying participants about prediction tasks, computing support values and outlier scores, and determining an outlier index to curate an optimized population for crowd-based or swarm-based intelligence generation, allowing for the culling of low-performing participants and weighting high-performing ones.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If adaptive outlier analysis is implemented to identify high-performing participants, then prediction accuracy is improved, but system complexity increases
Solution Approach 1:
The system performs preliminary outlier analysis on participant predictions before final aggregation. By computing outlier scores and indices in advance, the system identifies and weights high-performing participants beforehand, improving prediction accuracy without adding complexity during the actual prediction process.
Solution Approach 2:
The patent introduces an intermediary computational layer that processes participant predictions through outlier analysis. This intermediary step computes support values, outlier scores, and indices to transform raw predictions into weighted inputs for final aggregation, resolving the contradiction by structuring the complexity in a manageable intermediate stage.
2Measurement precision
If outlier analysis is performed on all participants, then identification of high-performers is improved, but computational time increases
Solution Approach 1:
The patent segments the participant population into different groups based on prediction patterns and outlier indices. By dividing participants into segments such as high-performers, low-performers, and moderate performers, the system can apply different levels of analysis to different segments, reducing overall computational time while maintaining identification accuracy.
Solution Approach 2:
The system performs full outlier analysis on a subset of participants who are most likely to be high-performers, while applying simplified analysis to others. This partial action approach focuses computational resources on the critical subset of participants, achieving good identification accuracy without analyzing every single participant in detail.
3Reliability
If the population is curated by culling low-performing participants, then prediction reliability is improved, but the diversity of the population decreases
Solution Approach 1:
The patent applies different quality standards to different participants based on their outlier indices. Instead of uniform culling, the system weights participants locally according to their performance characteristics, maintaining diversity in the population while giving appropriate influence to high-performing individuals. This local quality approach preserves population versatility while improving prediction reliability.
Solution Approach 2:
The system changes the parameter of participant influence from binary (included/excluded) to continuous (weighting factors). By transforming the culling parameter into a weighting parameter, the system maintains population diversity while improving reliability through differential weighting of participants based on their outlier indices and performance patterns.
Data Source
AI summary
System and method for amplifying the accuracy of forecasts generated by software systems that harness the collective intelligence of human populations by curating optimized sub-populations through an intelligent selection process. Participants predict event outcomes and/or provide evaluations of their confidence in their predictions. The system determines a score wherein the alignment score indicates how well that participant's prediction aligns with the predictions given by the baseline population. Participants can then be selected from the population based on the participant alignment scores.