Data Class Separability Analysis for Reliable Data Collection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data collection methods often result in incorrect assumptions due to inadequate preliminary testing, leading to resource-intensive and potentially inconclusive results, as seen in projects like tracking hand motions in sign language, where distinguishing symbols using inertial measurement units may not generalize to a larger group, causing costly delays or project cancellation.
Innovation Solution
A computing device apparatus and method that determines data class separability by calculating average intra-class similarity and inter-class similarity, allowing for the removal of highly variable classes and combination of inseparable ones, utilizing a hardware acceleration module with artificial neurons to refine data collection and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If preliminary testing is conducted with a small sample size, then initial assumptions can be made quickly, but the results may not generalize to larger groups leading to incorrect conclusions
Solution Approach 1:
The patent performs preliminary separability analysis using a small pilot dataset before full-scale data collection. This preliminary action allows researchers to assess whether the chosen features and classes are separable, preventing waste of resources on doomed projects while maintaining reliability through the mathematical separability criteria.
Solution Approach 2:
The patent collects more data than the minimum required for preliminary testing, using an excess amount to robustly estimate separability metrics. This partial action approach ensures that the preliminary results are reliable enough to predict full project outcomes while still being much less resource-intensive than complete data collection.
2Quantity of substance
If data collection is performed without preliminary separability analysis, then all available resources can be used for data collection, but resources are wasted on collecting data that cannot be effectively used
Solution Approach 1:
The patent performs separability analysis before full data collection to determine whether the research question is answerable with the proposed methodology. This preliminary action prevents wasting resources on collecting large amounts of data that cannot be effectively used due to poor separability.
Solution Approach 2:
The patent uses the results of preliminary separability analysis to provide feedback on whether to proceed with full data collection, modify the experimental design, or abandon the project. This feedback mechanism ensures resources are allocated efficiently based on objective criteria.
3Ease of operation
If incorrect assumptions are made during data collection planning, then the project can proceed with confidence, but the results may be inconclusive leading to project cancellation
Solution Approach 1:
The patent performs preliminary separability analysis during the planning stage to validate assumptions before full project execution. This preliminary action maintains ease of operation by providing a clear go/no-go decision criterion while ensuring reliability through objective mathematical assessment of data separability.
Solution Approach 2:
The patent proactively identifies potential failures in data separability before they occur by analyzing pilot data. This preliminary anti-action prevents the harmful outcome of inconclusive results by detecting separability issues early, allowing corrective actions or project abandonment before significant resources are committed.
Data Source
AI summary
Methods, apparatus, and system determine if a data class in a plurality of data classes is separable, such as by determining an average intra-class similarity within each data class, inter-class similarity across all data classes in the plurality of data classes, and determining separability based on the average intra-class similarity relative to the inter-class similarity. Data classes determined to be highly variable may be removed. Pair(s) of data classes not separable from one another may be combined into one class or one of the data classes may be dropped. A hardware accelerator, which may comprise artificial neurons, accelerate performance of the data analysis.