Air Conditioner Label Inference With Semi-Supervised Model Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning techniques face challenges in predicting with sufficient accuracy when the amount of labeled training data is insufficient, particularly in the context of air conditioning apparatuses.
Innovation Solution
A method involving the acquisition of combined data sets from air conditioning apparatuses, including measurement data and labels, to generate trained models that can infer labels for measurement data, with the option to create a lighter second trained model using mapping information and semi-supervised learning techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only ground-truth-labeled measurement data is used for training, then the training data requires high quality labels, but the amount of usable training data becomes insufficient
Solution Approach 1:
The patent combines ground-truth-labeled measurement data with predicted-labeled measurement data into a unified training data set. The labeled data generation unit creates training data by combining high-quality ground-truth labels with large quantities of predicted labels from the trained model, thereby merging the advantages of both data sources to achieve sufficient training data volume while maintaining acceptable label quality
Solution Approach 2:
The patent performs preliminary training with ground-truth-labeled data to create an initial trained model, which then generates predicted labels that are used to expand the training data set. This preliminary action enables the system to leverage the initial model's predictions to create additional training data, effectively bootstrapping the data generation process
2Measurement precision
If a large first trained model is generated to improve prediction accuracy, then prediction accuracy improves, but communication cost and processing load increase
Solution Approach 1:
The patent extracts and transmits only the essential trained model parameters from the large first trained model to the air conditioning apparatus, rather than transmitting the complete model. This extraction approach reduces communication cost while preserving the core predictive capabilities needed for accurate operation
Solution Approach 2:
The patent creates a simplified second trained model that replicates the essential predictive functionality of the large first trained model but with reduced complexity. This copying approach enables the air conditioning apparatus to perform accurate predictions locally without requiring the full original model, thereby reducing communication and processing requirements
Data Source
AI summary
This disclosure aims to provide a technique for improving the accuracy of prediction. A first trained model for inferring labels for measurement data is generated based on a first data set. The first data set includes: combined data that are a combination of first measurement data, which are related to a first air conditioning apparatus, and labels set for the first measurement data; and second measurement data related to the first air conditioning apparatus.