Learning Model Training via Partial Summary Variation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods face challenges in accurately learning a model for generating summary sentences due to the presence of inappropriate training data, which can lead to the generation of incorrect summary sentences.
Innovation Solution
A learning method that acquires an input sentence and its summary, generates a partially changed version of the summary, and uses error backpropagation to learn a model that prioritizes the original summary over the changed version, ensuring the model generates the correct summary sentence with higher probability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data including inappropriate summary sentences is used for learning, then the learning process can proceed with available data, but the model generates incorrect summary sentences with high probability
Solution Approach 1:
The system performs preliminary action by generating candidate summary sentences before the main learning process, then evaluating and selecting the most appropriate ones. This preliminary filtering and ranking of training data ensures that only high-quality summaries are used for teaching the model, thereby improving reliability without significantly increasing overall system complexity
Solution Approach 2:
The system implements feedback by evaluating generated summary sentences against the original input sentences and using this evaluation information to adjust the learning process. The model receives feedback on the quality of its predictions and continuously improves by prioritizing training examples that demonstrate correct summarization, thus resolving the contradiction between reliability and complexity
2Reliability
If the model is trained to prioritize correct summary sentences, then the accuracy of generated summaries improves, but the learning time and computational resources increase
Solution Approach 1:
The system applies partial action by focusing the enhanced learning process only on critical aspects of summary generation rather than uniformly processing all training data with maximum computational effort. By selectively applying complex evaluation and prioritization mechanisms only where needed, the system achieves high accuracy while controlling learning time
Solution Approach 2:
The system changes parameters by dynamically adjusting the importance weights of different training examples based on their quality and relevance. Instead of using fixed learning rates and uniform processing, the model adapts parameter values during training to prioritize informative examples, thereby achieving high reliability with reduced overall computational cost and time
Data Source
AI summary
A non-transitory computer-readable recording medium storing a learning program that causes a computer to execute a process, a process includes acquiring a first input sentence and a first summary sentence into which the first input sentence is summarized, generating a second summary sentence to which the first summary sentence is partially-changed, and executing machine learning for a model to generate at least one summary sentence that corresponds to at least one input sentence, respectively, in response to an input of the at least one input sentence, such that a first probability of generating the first summary sentence in response to the input of the first input sentence becomes higher than a second probability of generating the second summary sentence in response to the input of the first input sentence, based on the first input sentence, the first summary sentence, and the second summary sentence.