An Online Kernel Scalar Quantization Scheme for Signal Classification

Nonparametric Decentralized Detection and Sparse Sensor Selection via Multi-Sensor Online Kernel Scalar Quantization

“…In the past, authors have approached this topic by leveraging kernelbased learning techniques, e.g., [12]- [16]. Kernel methods are commonly employed to reduce computational complexity in problems that require mapping non-linearly separable datasets to higher dimensional spaces for decision learning [12]- [16]. The scenarios studied in [12]- [14] most closely align with our focus in this paper.…”

“…This widespread use of WSNs is driving research in a number of directions. Of particular continuing interest is leveraging WSNs to provide detection/estimation of one or more underlying phenomena, specifically for signal classification tasks [12]- [16]. Several works have studied wireless signal classification and related problems in non-WSN settings as well, i.e., where the observations are centralized [17], [18].…”

“…In this paper, we focus on decentralized detection [12]- [16], [19]- [24] exploiting a compress-and-forward sensor network [2]- [10], [12]. In this setting, each sensor forwards information resulting from a local decision rule to a fusion center for further processing and decision-making.…”

“…In this setting, each sensor forwards information resulting from a local decision rule to a fusion center for further processing and decision-making. Decentralized detection problems can be divided into parametric [20]- [22] and non-parametric categories [12]- [16], [23], [24]. A parametric decentralized detection problem assumes that a fusion center knows or partially knows the joint distribution of X (a random vector) and Y (a random variable), denoted as P (X, Y ), where X is the observation vector made by the sensors and Y is the class label of X.…”

“…We are interested specifically in non-parametric decentralized detection for signal classification tasks. In the past, authors have approached this topic by leveraging kernelbased learning techniques, e.g., [12]- [16]. Kernel methods are commonly employed to reduce computational complexity in problems that require mapping non-linearly separable datasets to higher dimensional spaces for decision learning [12]- [16].…”