Minimum-Cost Sensor Channel Selection For Wearable Computing

TL;DR

Addresses the problem of selecting a cost-efficient subset of sensor channels for wearable computing while meeting a performance constraint. Proposes MCCS with two backward-search algorithms: branch-and-bound for globally optimal subsets and a greedy method for efficient intermediate subsets, both incorporating a performance threshold $\lambda$ and a cost model $\{w_i\}$. Empirical evaluation on EEG and PAMAP2 time-series datasets shows cost savings up to $94.8\%$ (branch-and-bound) and $89.6\%$ (greedy) while meeting performance thresholds; PAMAP2 results illustrate robust performance under reduced channel sets. The framework is model-agnostic, supports runtime channel reconfiguration, and can be implemented with a single model masked to handle unavailable channels, enabling practical deployment in wearable sensing.

Abstract

Sensor systems are constrained by design and finding top sensor channel(s) for a given computational task is an important but hard problem. We define an optimization framework and mathematically formulate the minimum-cost channel selection problem. We then propose two novel algorithms of varying scope and complexity to solve the optimization problem. Branch and bound channel selection finds a globally optimal channel subset and the greedy channel selection finds the best intermediate subset based on the value of a score function. Proposed channel selection algorithms are conditioned with performance as well as the cost of the channel subset. We evaluate both algorithms on two publicly available time series datasets of human activity recognition and mental task detection. Branch and bound channel selection achieved a cost saving of up to 94.8% and the greedy search reduced the cost by 89.6% while maintaining performance thresholds.

Figures (2)

• Figure 1: Modular architecture of the CNN model with number of feature extractor equal to the number of input channels.
• Figure 2: Accuracy and cost of the selected channel subset using greedy search for both datasets at different values of alpha. The values on the line denotes the number of selected channels. For example at $\alpha=0.6$, $2$ channels were selected for the EEG mental arithmetic task and $20$ channels were selected for PAMAP activity recognition task.