Tiny EDA Models Through Neural Network Search for Wearable Stress Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wearable sensor-based stress detection systems face challenges in achieving a balance between classification accuracy and computational resource efficiency, particularly on battery-powered devices, with manual design processes being time-consuming and inefficient.
Innovation Solution
A processor-implemented method utilizing a Neural Network Search Space (NNSS) for automated generation of tiny models that optimize feature selection and model generation directly on wearable devices, minimizing computational overhead and battery consumption while maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual design process is used for ML/DL models, then classification accuracy can be optimized, but computational resource consumption increases and design time is extended
Solution Approach 1:
The patent transforms the design problem from manual parameter tuning to automated neural architecture search, where the system automatically optimizes model parameters, architecture, and feature selection. This eliminates the need for manual design iterations while achieving comparable accuracy with reduced resource footprint through automated hyperparameter optimization and architecture search.
Solution Approach 2:
The system performs self-optimization by automatically generating and selecting optimal model configurations without human intervention. The automated neural network search space exploration enables the model to self-adjust its architecture and parameters to achieve the best trade-off between accuracy and resource consumption for the specific wearable device constraints.
2Measurement precision
If complex ML/DL models are deployed on wearable devices, then classification accuracy improves, but device complexity and resource requirements increase
Solution Approach 1:
The patent segments the model development process into distinct phases: feature extraction, neural architecture search, and model selection. This segmentation allows for systematic optimization at each stage, enabling the generation of simplified models that maintain accuracy while reducing complexity through automated exploration of the architecture search space.
Solution Approach 2:
The system dynamically adapts model complexity based on device constraints. Through automated neural architecture search, the model configuration is dynamically optimized to match the specific computational capabilities of the target wearable device, generating appropriately complex models rather than deploying uniformly complex architectures.
3Measurement precision
If manual model design is performed, then model performance can be optimized, but development time and productivity are reduced
Solution Approach 1:
The patent replaces the manual mechanical process of model design and tuning with an automated computational system. The neural architecture search algorithm automatically explores the model space and identifies optimal configurations, substituting human expert manual work with automated search and evaluation processes that are faster and more systematic.
Solution Approach 2:
The system performs preliminary automated exploration of the model architecture search space before final deployment. By pre-optimizing model configurations through automated search and evaluation, the system prepares ready-to-deploy models that require minimal manual intervention, significantly accelerating the overall development and deployment timeline.
Data Source
AI summary
Wearable sensor-based stress detection is a well explored area of research in affective computing domain and are performed with non-invasive sensing modalities like Electrodermal Activity (EDA). In recent years, with the increased availability of such wearable devices to end-users, these applications have become more pervasive and thus require a greater level of optimization for continuous usage on resource constrained and battery-powered devices. While several research works have focused on designing Machine Learning (ML) and Deep Learning (DL) models for these tasks, very few focus on resource footprint. The balance between classification accuracy and computational resources is difficult to achieve with manual design process and is time consuming. In the present disclosure, systems and methods apply Neural Network Search Space (NASS) technique on features for feature optimization, and generation of tiny models based on optimized features set suitable for round-the-clock inference, with minimal resource requirements, and low latency.