Variable-Output-Space Prediction Models Using Contextual Input Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive data analysis solutions face inefficiencies and reliability issues due to the need to perform a large number of computational operations for classification tasks, particularly when dealing with a large number of candidate predictive associations.
Innovation Solution
The use of an output space refinement machine learning model to filter dynamically-preselected candidate predictive associations, combined with an isolated input embedding and contextual input embedding model, reduces the computational load by generating a subset of relevant associations for the variable-output-space prediction model, thereby improving computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of candidate predictive associations are evaluated for classification, then the accuracy and completeness of prediction results are improved, but the computational operations and processing resources required increase substantially
Solution Approach 1:
The patent segments the classification process into two distinct stages: (1) an output space refinement model that filters candidate predictive associations to create a reduced subset, and (2) a final classification model that performs detailed analysis only on the filtered subset. This segmentation allows the system to maintain high prediction accuracy by thoroughly analyzing relevant candidates while avoiding computational waste on obviously irrelevant associations.
Solution Approach 2:
The output space refinement model performs preliminary filtering of candidate predictive associations before the main classification process. By pre-identifying and ranking the most relevant candidates based on initial feature matching, the system prepares a reduced set of high-priority associations that the final classification model can process in detail, thereby improving overall computational efficiency without sacrificing accuracy.
2Reliability
If all B candidate predictive associations are processed to generate classification scores, then comprehensive prediction coverage is achieved, but the processing operations and resource utilization increase significantly
Solution Approach 1:
The patent extracts and isolates the most promising candidate predictive associations from the full set of B candidates using the output space refinement model. This extraction process identifies and separates the top C candidates (where C < B) that are most likely to be relevant, allowing the system to focus computational resources on these extracted candidates while maintaining comprehensive prediction coverage for the final classification.
Solution Approach 2:
The patent applies different processing quality levels to different candidates: high-level filtering and ranking is applied to all B candidates to identify relevant ones, while detailed classification analysis is applied only to the top C candidates. This local quality approach ensures that computational resources are allocated efficiently, with intensive processing reserved for candidates that have already been pre-qualified as potentially relevant.
3Stability of the object's composition
If the classification model processes a fixed large set of candidate associations, then consistent output space is maintained, but the model complexity and processing time increase for each prediction
Solution Approach 1:
The patent introduces dynamic adaptability into the classification process by allowing the output space to vary based on the input data. The output space refinement model dynamically determines the appropriate subset of candidates to process based on the specific characteristics of each prediction input, creating a variable output space that adapts to each case rather than processing a fixed large set for all inputs.
Solution Approach 2:
The patent changes the parameter of output space size from a fixed constant to a variable determined by the refinement model. By adjusting the effective number of candidates processed based on input features and refinement scores, the system maintains consistency in its approach while varying the actual processing load, thereby reducing average processing time without sacrificing output reliability.
Data Source
AI summary
As described herein, various embodiments of the present invention use an output space refinement machine learning model to filter C of the B candidate predictive associations that are referred to herein as dynamically-preselected candidate predictive associations for each prediction input data object, with every prediction input data object being associated with a different subset of the dynamically-preselected candidate predictive associations, where C is less than B and is in some embodiments typically much less than B. As described in greater detail below, this approach reduces the number of computational operations that need to be performed by a final classification machine learning model (referred to herein as a variable-output-space prediction machine learning model), and leads to substantial computational efficiency advantages relative to naïve implementations.