Machine Learning Loss Function for Non-Functional Performance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing language models take a long time to generate prediction results that satisfy desired non-functional performance, as they primarily focus on functional requirements and do not consider non-functional aspects like execution speed and accuracy during training.
Innovation Solution
Incorporating a term that reflects non-functional performance into the loss function during machine learning, using a parameter that indicates the ratio of non-functional performance, allowing the model to generate predictions that meet required non-functional criteria in a shorter time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If language models are trained using conventional loss functions that focus only on functional requirements, then the model can generate predictions based on basic functionality, but the time required to generate prediction results that satisfy desired non-functional performance becomes excessively long
Solution Approach 1:
The patent applies preliminary action by incorporating non-functional performance considerations into the loss function during the training phase. The loss function includes a first term for functional accuracy and a second term that penalizes violations of non-functional requirements (execution speed, memory usage). This allows the model to learn non-functional performance characteristics in advance, so that when generating predictions, the model can satisfy both functional and non-functional requirements without requiring extensive post-generation optimization or retraining, thus reducing the time needed to produce satisfactory results.
2Productivity
If language models prioritize functional requirements during training, then the model achieves basic prediction capability, but the non-functional performance (execution speed, accuracy) does not meet desired levels
Solution Approach 1:
The patent applies parameter changes by modifying the loss function to include a second term that specifically addresses non-functional performance parameters. The loss function becomes: L = L_functional + λ * L_non-functional, where L_non-functional penalizes predictions that violate non-functional requirements such as execution speed thresholds, memory usage limits, or accuracy constraints. The hyperparameter λ controls the weight of non-functional performance considerations. This parameter change in the training objective enables the model to simultaneously optimize for both productivity (prediction generation speed) and reliability (non-functional performance quality) by balancing these competing objectives during training.
Data Source
AI summary
A non-transitory computer-readable recording medium stores a machine learning program for causing a computer to execute processing including: measuring, for each data, a non-functional performance that represents a performance for a requirement that excludes a function of each data; and by machine learning that uses divided data obtained by dividing each data into a first portion of the data and a second portion that is correct answer data as training data, executing machine learning processing of training a prediction model that predicts the second portion of the data in response to an input of the first portion of the data, wherein the machine learning processing uses a loss function that includes a parameter determined according to a measurement result of the non-functional performance that is the parameter that indicates a ratio of reflecting the non-functional performance in the prediction model.