This paper leverages transfer learning~(TL) on a mixture of experts~(MoE) model for radio map estimation. The proposed MoE combines location-based and location-free experts through a gating network exploiting their complementary benefits. To estimate the radio map in a new wireless environment, the learned model of another sufficiently similar environment is transferred and fine-tuned with additional data from the new environment. The proposed data-driven similarity measure predicts the amount of data needed for TL. Results demonstrate that the proposed method only requires 20-40\% of measurements to adapt to several varying environments, and as expected, the proposed MoE method outperforms both experts.

Predictive linear and nonlinear models based on kernel machines or deep neural networks have been used to discover dependencies among time series. This paper proposes an efficient nonlinear modeling approach for multiple time series, with a complexity comparable to linear vector autoregressive (VAR) models while still incorporating nonlinear interactions among different time-series variables. The modeling assumption is that the set of time series is generated in two steps: first, a linear VAR process in a latent space, and second, a set of invertible and Lipschitz continuous nonlinear mappings that are applied per sensor, that is, a component-wise mapping from each latent variable to a variable in the measurement space. The VAR coefficient identification provides a topology representation of the dependencies among the aforementioned variables. The proposed approach models each component-wise nonlinearity using an invertible neural network and imposes sparsity on the VAR coefficients to reflect the parsimonious dependencies usually found in real applications. To efficiently solve the formulated optimization problems, a custom algorithm is devised combining proximal gradient descent, stochastic primal-dual updates, and projection to enforce the corresponding constraints. Experimental results on both synthetic and real data sets show that the proposed algorithm improves the identification of the support of theVAR coefficients in a parsimonious manner while also improving the time-series prediction, as compared to the current state-of-the-art methods.

A document by Baltasar Beferull-Lozano . Click on the document to view its contents.

Estimating accurate radio maps is important for various tasks in wireless communications, such as channel modeling, resource allocation, and network planning, to name a few. Due to the changes in the propagation characteristics of the wireless environments, a radio map model learned under a particular wireless environment cannot be directly used in a new wireless environment. Moreover, learning a new model for every environment requires, in general, a large amount of data and is computationally demanding. In this work, we design an effective novel data-driven transfer learning method that transfers and fine-tunes a deep neural network (DNN)-based radio map model learned from an original wireless environment to other wireless environments with a certain level of similarity, allowing the radio map to be estimated with less amount of training data. As opposed to other widely used similarity measures that do not take into account the wireless propagation characteristics, we design a data-driven similarity measure that predicts the mean square error (MSE) and the amount of training data needed when learning a radio map in a new wireless environment. The proposed solution is corroborated by extensive simulations over a range of wireless environments, achieving savings of approximately 60-70% in sensor measurement data.

An online topology estimation algorithm for nonlinear structural equation models (SEM) is proposed in this paper, addressing the nonlinearity and the non-stationarity of real-world systems. The nonlinearity is modeled using kernel formulations, and the curse of dimensionality associated with the kernels is mitigated using random feature approximation. The online learning strategy uses a group-lasso-based optimization framework with a prediction-corrections technique that accounts for the model evolution. The proposed approach has three properties of interest. First, it enjoys node-separable learning, which allows for scalability in large networks. Second, it offers privacy in SEM learning by replacing the actual data with node-specific random features. Third, its performance can be characterized theoretically via a dynamic regret analysis, showing that it is possible to obtain a linear dynamic regret bound under mild assumptions. Numerical results with synthetic and real data corroborate our findings and show competitive performance w.r.t. state-of-the-art alternatives.

An online nonlinear topology identification algorithm termed RFNL-TIRSO is proposed in the paper. The multivariate time series data received in sequential form is processed online to estimate time-varying nonlinear dependencies