Radar Sleep Monitoring Model Training Using Simulated Dimensions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Obtaining large datasets for training machine learning models, especially those related to sleep monitoring, is time-consuming and expensive, and datasets from different sensors with varying dimensions cannot be directly applied to each other, making it challenging to transfer knowledge across sensor types.
Innovation Solution
A method is described where a first dataset with fewer dimensions is enhanced by simulating an additional dimension, allowing it to be used to train a machine learning model that can classify data from a second dataset with more dimensions, using noise to simulate the additional dimension and leveraging weights from a pre-trained model to fine-tune a new model for different sensor types like chest straps and radar sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large polysomnography dataset is created using chest straps to train a machine learning model, then the model can accurately classify sleep states based on chest movement data, but the dataset cannot be directly applied to other sensor types such as radar sensors due to dimensional mismatches
Solution Approach 1:
The patent applies dimensionality change by transforming the 2D chest strap dataset (magnitude over time) into a 3D representation that includes distance as an additional dimension. This is achieved by simulating distance information through noise addition and data transformation, allowing the model to process radar-like 3D input data. The transformation enables the chest strap dataset to be adapted for training radar-based sleep monitoring models without requiring new data collection.
2Measurement precision
If a new clinical study is conducted to create a large dataset for radar sensor training, then the model can be trained on radar data, but the process becomes exceedingly time-consuming and cost prohibitive
Solution Approach 1:
The patent applies preliminary action by pre-processing and transforming the existing chest strap dataset into a format that simulates radar data characteristics before training begins. The dataset is enhanced with simulated distance dimensions and noise patterns that mimic radar sensor output. This preliminary transformation allows the model to be trained on radar-like data without actually conducting a new clinical study, saving significant time and resources while maintaining training effectiveness.
3Device complexity
If the first dataset with fewer dimensions is used directly for training, then the training process is simpler, but the model cannot effectively process the additional dimension present in the second dataset
Solution Approach 1:
The patent applies parameter changes by systematically modifying the dimensional parameters of the first dataset. Noise is added to simulate the additional dimension, and data transformation techniques are applied to convert 2D magnitude-time data into 3D magnitude-distance-time data. These parameter changes enable the model to learn from multi-dimensional radar-like input while maintaining reasonable processing complexity, bridging the gap between simple data processing and reliable multi-dimensional model performance.
Data Source
AI summary
Various arrangements are presented for training and using a machine learning model. A first training data set may be created that has more samples but fewer dimensions than a second dataset. A second set of training data, created from the second dataset, has at least one additional dimension of data than the first set of training data. An additional dimension of data can then be simulated for the first set of training data. The simulated additional dimension of data can be incorporated with the first set of training data. A first machine learning model can be trained based on the first set of training data that comprises the simulated additional dimension of data to obtain various weights. A second machine learning model can then be trained based on the second set of training data and the obtained plurality of weights from the first trained machine learning model.