Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sensor Calibration Model Balancing Accuracy, Real-time, and Efficiency

Jinyong Yun, Hyungjin Kim, Seokho Ahn, Euijong Lee, Young-Duk Seo

TL;DR

This work addresses the deployment gap in on-device sensor calibration by decomposing the traditional accuracy, real-time, and resource-efficiency triad into eight microscopic requirements. It introduces Scare, an ultra-compressed transformer featuring Sequence Lens Projector (SLP) for logarithmic sequence compression, Efficient Bitwise Attention (EBA) based on hash codes with sign-based operations, and a hash-function optimization strategy with dynamic sampling to enable end-to-end training without auxiliary losses. Across large-scale air-quality datasets and real MCU deployments, Scare achieves state-of-the-art calibration accuracy while maintaining low latency, minimal memory footprint, and strong hardware compatibility, effectively meeting all eight microscopic requirements. The approach enables robust, real-time, energy-efficient on-device calibration suitable for resource-constrained IoT systems, with potential for lightweight on-device continual learning and server-edge collaboration in future work.

Abstract

Most on-device sensor calibration studies benchmark models only against three macroscopic requirements (i.e., accuracy, real-time, and resource efficiency), thereby hiding deployment bottlenecks such as instantaneous error and worst-case latency. We therefore decompose this triad into eight microscopic requirements and introduce Scare (Sensor Calibration model balancing Accuracy, Real-time, and Efficiency), an ultra-compressed transformer that fulfills them all. SCARE comprises three core components: (1) Sequence Lens Projector (SLP) that logarithmically compresses time-series data while preserving boundary information across bins, (2) Efficient Bitwise Attention (EBA) module that replaces costly multiplications with bitwise operations via binary hash codes, and (3) Hash optimization strategy that ensures stable training without auxiliary loss terms. Together, these components minimize computational overhead while maintaining high accuracy and compatibility with microcontroller units (MCUs). Extensive experiments on large-scale air-quality datasets and real microcontroller deployments demonstrate that Scare outperforms existing linear, hybrid, and deep-learning baselines, making Scare, to the best of our knowledge, the first model to meet all eight microscopic requirements simultaneously.

Sensor Calibration Model Balancing Accuracy, Real-time, and Efficiency

TL;DR

This work addresses the deployment gap in on-device sensor calibration by decomposing the traditional accuracy, real-time, and resource-efficiency triad into eight microscopic requirements. It introduces Scare, an ultra-compressed transformer featuring Sequence Lens Projector (SLP) for logarithmic sequence compression, Efficient Bitwise Attention (EBA) based on hash codes with sign-based operations, and a hash-function optimization strategy with dynamic sampling to enable end-to-end training without auxiliary losses. Across large-scale air-quality datasets and real MCU deployments, Scare achieves state-of-the-art calibration accuracy while maintaining low latency, minimal memory footprint, and strong hardware compatibility, effectively meeting all eight microscopic requirements. The approach enables robust, real-time, energy-efficient on-device calibration suitable for resource-constrained IoT systems, with potential for lightweight on-device continual learning and server-edge collaboration in future work.

Abstract

Most on-device sensor calibration studies benchmark models only against three macroscopic requirements (i.e., accuracy, real-time, and resource efficiency), thereby hiding deployment bottlenecks such as instantaneous error and worst-case latency. We therefore decompose this triad into eight microscopic requirements and introduce Scare (Sensor Calibration model balancing Accuracy, Real-time, and Efficiency), an ultra-compressed transformer that fulfills them all. SCARE comprises three core components: (1) Sequence Lens Projector (SLP) that logarithmically compresses time-series data while preserving boundary information across bins, (2) Efficient Bitwise Attention (EBA) module that replaces costly multiplications with bitwise operations via binary hash codes, and (3) Hash optimization strategy that ensures stable training without auxiliary loss terms. Together, these components minimize computational overhead while maintaining high accuracy and compatibility with microcontroller units (MCUs). Extensive experiments on large-scale air-quality datasets and real microcontroller deployments demonstrate that Scare outperforms existing linear, hybrid, and deep-learning baselines, making Scare, to the best of our knowledge, the first model to meet all eight microscopic requirements simultaneously.

Paper Structure

This paper contains 29 sections, 4 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Microscopic Requirements
  4. Time-series Forecasting Models
  5. Hash-based Models
  6. Sensor Calibration Models
  7. Efficient On-Device Sensor Calibration Model: Scare
  8. Definition of Sensor Calibration Problem
  9. Overview of Scare
  10. Sequence Lens Projector
  11. Efficient Bitwise Attention
  12. Hash Function Optimization
  13. Experiment
  14. Evaluation setup.
  15. Baselines.
  16. ...and 14 more sections

Figures (3)

  • Figure 1: Overview of Scare, composing (i) Sequence Lens Projector (SLP), (ii) Efficient Bitwise Attention (EBA), and (iii) Hash Function Optimization.
  • Figure 2: RMSE comparisons within ±30 time points around change points, under error change thresholds of 5%, 10%, 20% on the Ant. and Zag. datasets.
  • Figure 3: Performance on Arduino Nano 33 BLE Sense. The graphs show inference latency and flat-buffer size across varying windows (15, 60, 360, 720, 1440). Missing points indicate that on-device measurement was not possible.