Multitask Machine Learning Model Training for Question Answering Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine-learning models for natural language understanding, particularly in question answering, face challenges with long questions due to inaccuracies caused by superfluous language, leading to hindered performance in computing devices.
Innovation Solution
Multitask machine-learning model training with techniques such as question summarization and recognizing question entailment, along with a configurable weighted loss function and data augmentation, improves accuracy by synthesizing question pairs and sharing parameters between tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine-learning models are used for question answering, then the models can process relatively short inputs, but accuracy deteriorates when confronted with long questions containing multiple sentences
Solution Approach 1:
The patent applies segmentation by dividing the question processing into multiple tasks: question summarization (creating a condensed version) and recognizing question entailment (determining if the question can be answered from given text). This multi-task approach allows the model to handle long questions effectively by breaking down the complex processing into manageable components, thereby improving accuracy without being hindered by input length
Solution Approach 2:
The patent implements multi-functionality through a universal machine-learning model that performs multiple functions simultaneously: it can summarize questions, determine entailment relationships, and answer questions. This single model handles diverse question types and lengths, improving adaptability while maintaining accuracy across different scenarios
2Measurement precision
If multitask machine-learning model training is implemented, then accuracy of each task improves, but device complexity increases
Solution Approach 1:
The patent merges multiple task objectives (question summarization and recognizing question entailment) into a single unified training process. By combining these tasks and sharing parameters between them, the model achieves improved accuracy for each individual task while managing complexity through parameter sharing rather than maintaining completely separate models for each function
3Measurement precision
If data augmentation techniques are employed to synthesize question pairs, then model training accuracy improves, but processing time increases
Solution Approach 1:
The patent applies preliminary action by synthesizing question pairs and augmenting training data before the main model training process. This pre-processing step creates enhanced training datasets that improve model accuracy, while the synthesis is performed automatically without user intervention, managing the time investment efficiently
Data Source
AI summary
Multitask machine-learning model training and training data augmentation techniques are described. In one example, training is performed for multiple tasks simultaneously as part of training a multitask machine-learning model using question pairs. Examples of the multiple tasks include question summarization and recognizing question entailment. Further, a loss function is described that incorporates a parameter sharing loss that is configured to adjust an amount that parameters are shared between corresponding layers trained for the first and second tasks, respectively. In an implementation, training data augmentation techniques are also employed by synthesizing question pairs, automatically and without user intervention, to improve accuracy in model training.