SudokuSens: Enhancing Deep Learning Robustness for IoT Sensing Applications using a Generative Approach

TL;DR

SudokuSens tackles data scarcity in IoT sensing by combining CVAE-based conditional interpolation to synthesize unseen intrinsic-attribute conditions with a session-aware temporal contrastive learning (SA-TCL) framework to distill disturbance-robust representations. The offline pre-training pipeline augments data and learns a disturbance-insensitive encoder, which is then used with a downstream classifier for deployment on edge devices such as a Raspberry Pi. Across seismic-acoustic vehicle detection, wearable HAR, and Wi-Fi HAR datasets, SudokuSens improves accuracy and F1 scores under unseen conditions, with larger gains at lower data coverage, and is validated in real-world outdoor deployments. Ablation studies and generalization analyses reveal factors influencing effectiveness, including condition coverage, temporal variability, and applicability to regression tasks. The work provides a practical, edge-friendly approach to robust IoT sensing with clear guidelines for deployment and known limits.

Abstract

This paper introduces SudokuSens, a generative framework for automated generation of training data in machine-learning-based Internet-of-Things (IoT) applications, such that the generated synthetic data mimic experimental configurations not encountered during actual sensor data collection. The framework improves the robustness of resulting deep learning models, and is intended for IoT applications where data collection is expensive. The work is motivated by the fact that IoT time-series data entangle the signatures of observed objects with the confounding intrinsic properties of the surrounding environment and the dynamic environmental disturbances experienced. To incorporate sufficient diversity into the IoT training data, one therefore needs to consider a combinatorial explosion of training cases that are multiplicative in the number of objects considered and the possible environmental conditions in which such objects may be encountered. Our framework substantially reduces these multiplicative training needs. To decouple object signatures from environmental conditions, we employ a Conditional Variational Autoencoder (CVAE) that allows us to reduce data collection needs from multiplicative to (nearly) linear, while synthetically generating (data for) the missing conditions. To obtain robustness with respect to dynamic disturbances, a session-aware temporal contrastive learning approach is taken. Integrating the aforementioned two approaches, SudokuSens significantly improves the robustness of deep learning for IoT applications. We explore the degree to which SudokuSens benefits downstream inference tasks in different data sets and discuss conditions under which the approach is particularly effective.

Figures (14)

• Figure 1: The differences of dataset collection methodology between CV/NLP and IoT sensing applications
• Figure 2: A case study to motivate SudokuSens. The average accuracy and the accuracy for each vehicle type during the development phase are presented.
• Figure 3: SudokuSens. The name highlights the design of conditional interpolation; it generates data for missing intrinsic attribute combinations much like solving a Sudoku puzzle.
• Figure 4: The CVAE architecture for conditional interpolation. A 2-modality input is taken as an example to illustrate the design.
• Figure 5: Session-aware temporal contrastive learning.