Heterogeneous Sensor Data Fusion via Inferred Significance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Relevance Vector Machines (RVMs) have limited flexibility and struggle to adequately model uncertainty for inputs far away from the training data, particularly when dealing with heterogeneous data sources that use different terminologies, units, and data types.
Innovation Solution
The method involves inferring the significance of pre-processing methods used for data points from heterogeneous sources by performing a machine learning algorithm, specifically using a regression technique like Gaussian Processes, which applies kernel functions to combine data from diverse sources, allowing for the identification of the most significant pre-processor and data source.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If conventional RVM is used to process data from heterogeneous sources, then computational cost is reduced, but flexibility and ability to model uncertainty are limited
Solution Approach 1:
The patent segments the data processing approach by treating each data source and pre-processing method as separate, identifiable components. The system infers significance values for individual pre-processing methods and data sources, allowing selective weighting and combination of heterogeneous data without requiring uniform treatment, thus maintaining flexibility while managing computational complexity.
Solution Approach 2:
The patent introduces significance values as dynamic parameters that quantify the reliability and relevance of each data source and pre-processing method. By inferring these parameters from the data itself rather than using fixed conventional RVM assumptions, the system adapts to heterogeneous data sources while maintaining computational efficiency through parameterized uncertainty modeling.
2Device complexity
If conventional RVM is used with fixed basis functions, then model simplicity is maintained, but ability to adequately model uncertainty for inputs far from training data is reduced
Solution Approach 1:
The patent transforms the static basis function approach of conventional RVM into a dynamic system where significance values are inferred for each data point based on its context and distance from training data. This dynamic adaptation allows the model to adjust its uncertainty estimates and basis function weights according to the specific input, improving reliability for extrapolation while maintaining reasonable complexity through efficient inference algorithms.
3Adaptability or versatility
If multiple pre-processing methods are applied to heterogeneous data sources, then data fusion capability is improved, but difficulty in determining significance of each source increases
Solution Approach 1:
The patent implements a feedback mechanism where the system infers significance values for pre-processing methods and data sources based on the actual performance and characteristics of the fused data. This inferred significance information feeds back into the data fusion process, allowing the system to automatically adjust weights and prioritize reliable data sources, thereby reducing the difficulty of significance determination while maintaining high data fusion capability.
Data Source
AI summary
A method and apparatus for processing data, the data including: a set of one or more system inputs; and a set of one or more system outputs; wherein each system output corresponds to a respective system input; each system input includes a plurality of data points, a first data point in the plurality and a second data point in the plurality being from a same raw data source, and the first data point being pre-processed using a different pre-processing method relative to a pre-processing method used to pre-process the second data point, the method including: for each of the first and second data points, inferring a value indicative of a significance of the pre-processing method used to pre-process that data point; wherein the inferring includes performing a machine learning algorithm on a given system input from the data and a further system input.