Multi-Task Learning with Incomplete Data via Augmented Matrices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-task learning methods face challenges in handling missing data, as they either delete observations or impute values, which can lead to biased models and neglect data variability, especially in applications like clinical studies, bioinformatics, and image processing.
Innovation Solution
The system computes an updated weight matrix by using an augmented observation matrix, plug-in autocovariance matrix, and plug-in covariance vector, incorporating a noise value and predefined probability of missing data, and applies graph regularization to handle missing features within the learning process, avoiding separate handling of missing values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If observations with missing features are deleted entirely, then the machine learning model can be trained without imputation bias, but the dataset becomes statistically non-representative and biased, resulting in considerable error in the trained model
Solution Approach 1:
The patent introduces an intermediary mechanism (augmented observation matrix with probability indicators) that bridges the gap between incomplete data and the learning algorithm. Instead of directly deleting or imputing missing values, the system creates an augmented representation that preserves both the original data structure and the uncertainty information, allowing the learning algorithm to work with the full dataset without bias introduction.
Solution Approach 2:
The patent transforms the data representation by changing parameters: it creates an augmented observation matrix that includes probability indicators alongside original features. This parameter transformation allows the system to maintain dataset size while providing the learning algorithm with explicit information about data completeness, resolving the contradiction between using full data and avoiding imputation bias.
2Quantity of substance
If a zero or constant value is used to provide the missing feature value, then the observation vector is preserved, but the model becomes biased and data variability is neglected
Solution Approach 1:
Instead of using constant values as intermediaries for missing data, the patent introduces probability indicators as the intermediary mechanism. These indicators explicitly represent the uncertainty and missingness information, allowing the learning algorithm to distinguish between observed and missing values without introducing bias, thereby preserving both dataset size and prediction accuracy.
Solution Approach 2:
The patent inverts the traditional approach by not trying to fill in missing values with constants, but rather by explicitly marking them with probability indicators. This inversion allows the system to preserve the full dataset while maintaining data variability and avoiding imputation bias, as the learning algorithm can directly utilize the probability information.
3Measurement precision
If a mean value or recent observation value is imputed for each missing feature, then the observation vector is completed, but uncertainty of missing values is neglected, inducing bias and underrating data variability
Solution Approach 1:
The patent uses probability indicators as intermediaries that preserve uncertainty information rather than eliminating it through imputation. These indicators serve as a bridge between the incomplete data and the learning algorithm, maintaining data variability and preventing the loss of information about missingness that occurs with traditional imputation methods.
Solution Approach 2:
The patent maintains continuity of useful action by preserving the full dataset with its inherent variability rather than disrupting it through imputation. The probability indicators allow the learning algorithm to continuously utilize all available data while respecting the uncertainty structure, preventing the breakdown of data variability that occurs with mean or recent value imputation.
4Measurement precision
If matrix completion method is used to provide the missing feature value, then the observation vector is completed, but assumptions that are not valid in many cases are made, inducing bias
Solution Approach 1:
The patent introduces probability indicators as a universal intermediary that works across different scenarios without requiring scenario-specific assumptions. This intermediary mechanism is adaptable to various data types and missingness patterns, making the approach versatile and broadly applicable while avoiding the restrictive assumptions of matrix completion methods.
Solution Approach 2:
The patent creates a universal solution that handles missing data across multiple scenarios and applications. The probability indicator mechanism is multi-functional, working with different data types, missingness patterns, and learning algorithms without requiring task-specific customization, thereby achieving broad adaptability while maintaining prediction accuracy.
Data Source
AI summary
A computing device computes a weight matrix to compute a predicted value. For each of a plurality of related tasks, an augmented observation matrix, a plug-in autocovariance matrix, and a plug-in covariance vector are computed. A weight matrix used to predict the characteristic for each of a plurality of variables and each of a plurality of related tasks is computed. (a) and (b) are repeated with the computed updated weight matrix as the computed weight matrix until a convergence criterion is satisfied: (a) a gradient descent matrix is computed using the computed plug-in autocovariance matrix, the computed plug-in covariance vector, the computed weight matrix, and a predefined relationship matrix, wherein the predefined relationship matrix defines a relationship between the plurality of related tasks, and (b) an updated weight matrix is computed using the computed gradient descent matrix.