Federated Feature Selection for Cyber-Physical Systems of Systems

Abstract

Autonomous vehicles (AVs) generate a massive amount of multi-modal data that once collected and processed through Machine Learning algorithms, enable AI-based services at the Edge. In fact, only a subset of the collected data present informative attributes to be exploited at the Edge. Therefore, extracting such a subset is of utmost importance to limit computation and communication workloads. Doing that in a distributed manner imposes the AVs to cooperate in finding an agreement on which attributes should be sent to the Edge. In this work, we address such a problem by proposing a federated feature selection (FFS) algorithm where the AVs collaborate to filter out, iteratively, the less relevant attributes in a distributed manner, without any exchange of raw data, thought two different components: a Mutual-Information-based feature selection algorithm run by the AVs and a novel aggregation function based on the Bayes theorem executed on the Edge. The FFS algorithm has been tested on two reference datasets: MAV with images and inertial measurements of a monitored vehicle, WESAD with a collection of samples from biophysical sensors to monitor a relative passenger. The numerical results show that the AVs converge to a minimum achievable subset of features with both the datasets, i.e., 24 out of 2166 (99%) in MAV and 4 out of 8 (50%) in WESAD, respectively, preserving the informative content of data.

Wrapper-Based Federated Feature Selection for IoT Environments

“…Independent and identically distributed (iid) and non-iid configurations are used for data distribution. Therefore, the results are comparable with results in [5], [10], [11], and [12]. The number of programs and iterations in local GSP are 30 and 5, respectively.…”

“…The results are demonstrated in Table III and IV. In the first scenario, filter-based FFS method proposed in [5] (Fed-FS-CE) can select a few number of features and reduce communication cost, but lost a lot of information. However, our method can construct multiple high-level features and achieve good accuracy with low computational complexity of local learning models.…”

“…However, our method can construct multiple high-level features and achieve good accuracy with low computational complexity of local learning models. For example, for ACC dataset, although Fed-FS-CE [5] and MFPSO [10] methods remove 80 % and 56 % of features respectively, they cannot provide accurate results. Moreover, FSHFL [12] and Fed-FS-GSA [11] methods obtain the same or about the same accuracy of the FL classifier with no feature selection (No-FS).…”

“…However, in horizontal FFS, clients have different instances with the same feature sets. The horizontal FFS in [5] used filter-based Cross-Entropy FS method in both clients and edge server. A few number of informative features were selected with information loss.…”

“…To construct new features in IoT environments with distributed datasets, a collaborative FC method is needed. There are a number of federated FS (FFS) methods in the literature that are inspired by federated learning (FL) procedure to select informative features in a collaborative manner [5], [6]. However, according to our best knowledge, FC has not been investigated from this perspective.…”