Distributed nonlinear regression using in-network processing with multiple Gaussian kernels

Distributed Online Learning with Multiple Kernels

“…Motivated by the success of multiple kernel-based learnings [10], [12], [22], [23], our learned function in Section III will be assumed as a parameterized function in (6). We emphasize that the parameters to be optimized in the proposed algorithms are equivalent to those in OMKL [23] (e.g., { θ[t,p] , q[t,p] : p ∈ [P ]}), whereas the optimization technique (or learning algorithm) in Section III is completely different from that in OMKL.…”

“…A privacy-preserving distributed learning is witnessing an unprecedented interest, in which local data is kept at distributed edge devices without centralizing the data. According to the structures of a communication network, this can be categorized into a centralized federated learning [6], [7], [8], [9] and a fully distributed learning (a.k.a., a fully decentralized federated learning) [10], [11], [12], [13].…”

“…A kernel-based learning has been extended into a fully decentralized network [10], [11], [12] due to its necessity in various applications such as social networks, big data processing, and environmental monitoring. In [10], a consensus-based distributed MKL has been proposed for regression tasks, where alternating direction method of multipliers (ADMM) [19] is used as the underlying distributed optimization technique. A diffusion-based distributed learning with a single kernel has been developed in [11], where the cooperation of distributed nodes is performed via the diffusion rationale.…”

“…A diffusion-based distributed learning with a single kernel has been developed in [11], where the cooperation of distributed nodes is performed via the diffusion rationale. In addition, such diffusion-based approach has been incorporated into MKL framework [12], being able to outperform the previous works [10], [11] in distributed estimation tasks. However, none of the above kernel-based approaches cannot operate in distribute online learning frameworks.…”

Distributed Adaptive Learning With Multiple Kernels in Diffusion Networks

“…Regarding distributed kernel-based estimation algorithms, several schemes have been derived [33]- [42]. In [33] a distributed consensus-based regression algorithm based on kernel least squares has been proposed and extended by multiple kernels in [34]. Both schemes utilize alternating direction method of multipliers (ADMM) [43] for distributed consensusbased processing.…”