Francesco Malandrino

The diffusion of GPS-equipped devices has resulted in the generation of vast amounts of spatio-temporal data. This data represents a fundamental resource to conduct analysis on transportation networks. It is therefore of great interest to identify models capable of distinguishing and classifying trajectories to facilitate decision-making processes, congestion prediction and emissions monitoring. However, many existing algorithms necessitate a complex feature engineering process and domain knowledge. In this context, this work proposes a neural network-based approach, which eliminates the need for complicated hand-crafted features, using Gramian angular fields and leveraging possibly pre-trained convolutional neural networks. Therefore, we combine these tools to tackle the challenge of multiclass trajectory classification. We demonstrate the effectiveness of our method on an imbalanced dataset simulated with SUMO by classifying different means of transportation–private car, taxi, bus, pedestrian, motorcycle, bicycle–achieving good results in terms of accuracy and F1 score.

Distributed machine learning, including federated learning (FL), is increasingly taking place within the networks, in the so-called cloud-edge-device continuum. This creates significant challenges, e.g., the need to locate the best resources and data to leverage; at the same time, it also brings novel opportunities to boost the performance of learning and its resiliency to external attacks. In this paper, we present a novel architecture called SHIELDED that enables the decision-making entities in charge of learning orchestration, FL security, and model calibration to: 1) work in an integrated manner and 2) exploit the additional information available in programmable networking scenarios. Our performance evaluation, using network intrusion detection as a case study, shows that SHIELDED yields about 50% better accuracy and 35% lower calibration error compared to present-day alternatives.

Machine learning (ML) can be often distributed, owing to the need to harness more resources and/or to preserve privacy. Accordingly, distributed learning has received significant attention from the literature; however, most works focus on the expected learning quality (e.g., loss) attained and do not consider the distribution thereof. It follows that ML models are not dependable, and may fall short of the required performance in many real-world cases. In this work, we tackle this challenge and propose DepL, a framework attaining dependable learning orchestration: DepL efficiently makes joint, near-optimal decisions concerning (i) which data to use for learning, (ii) the ML models to use – chosen within a set of full-size models and compressed versions thereof – and when to switch from one model to another, and (iii) the clusters of physical nodes to use for the learning. DepL improves over previous works by guaranteeing that the learning quality target (e.g., a minimum loss) is achieved with a target probability, while minimizing the learning (e.g., energy) cost. DepL has provably low polynomial computational complexity and a constant competitive ratio. Further, experimental results using the CIFAR-10 and GTSRB datasets show that it consistently matches the optimum and outperforms state-of-the-art approaches (30% faster learning and 40–80% lower cost).

Extended reality (XR) devices, commonly known as wearables, must handle significant computational loads under tight latency constraints. To meet these demands, they rely on a combination of on-device processing and edge offloading. This letter focuses on offloading strategies for wearables and assesses the impact of offloading decisions over three distinct time scales: instantaneous power consumption, short-term temperature fluctuations, and long-term battery duration. We introduce a comprehensive system model that captures these temporal dynamics, and propose a stochastic and stationary offloading strategy, called TAO (for temperature-aware offloading), designed to minimize the offloading cost while adhering to power, thermal, and energy constraints. Our performance evaluation, leveraging COMSOL models of real-world wearables, confirms that TAO successfully avoids exceeding temperature limits while keeping additional edge offloading to a minimum. These results also highlight how properly accounting for all features of wearables allows fully exploiting edge offloading opportunities.

Next-generation mobile systems will support multiple AI-based applications, each leveraging heterogeneous sensors and data sources through deep neural network (DNN) architectures collaboratively executed within the network. In this context, to minimize the cost of the AI inference task subject to requirements on latency, quality, and – crucially – reliability of the inference process, it is vital to optimize (i) the set of sensors/data sources and (ii) the DNN architecture, (iii) the network nodes executing sections of the DNN, and (iv) the resources to use. To achieve these goals, we leverage dynamic gated neural networks with branches, and propose a novel algorithmic strategy called Quantile-constrained Inference (QIC), based upon quantile-Constrained policy optimization. QIC makes joint, high-quality, swift decisions on all the above aspects of the system, with the aim to minimize inference energy cost. We remark that this is the first contribution connecting gated dynamic DNNs with infrastructure-level decision making. We evaluate QIC using a dynamic gated DNN with stems and branches for optimal sensor fusion and inference, trained on the RADIATE dataset offering Radar, LiDAR, and Camera data, and real-world wireless measurements. Our results confirm that QIC closely matches the optimum and outperforms existing approaches in reducing energy consumption (compute, communication, and total) and application requirements failure by over 70%.

Pruning neural networks, i.e., removing some of their parameters whilst retaining their accuracy, is one of the main ways to reduce the latency of a machine learning pipeline, especially in resource- and/or bandwidth-constrained scenarios. In this context, the pruning technique, i.e., how to choose the parameters to remove, is critical to the system performance. In this paper, we propose a novel pruning approach, called FlexRel and predicated upon combining training-time and inference-time information, namely, parameter magnitude and relevance, in order to improve the resulting accuracy whilst saving both computational resources and bandwidth. Our performance evaluation shows that FlexRel is able to achieve higher pruning factors, saving over 35% bandwidth for typical accuracy targets.

An increasing number of mobile applications leverage deep neural networks (DNN) as an essential component to adapt to the operational context at hand and provide users with an enhanced experience. It is thus of paramount importance that network systems support the execution of DNN inference tasks in an efficient and sustainable way. Matching the diverse resources available at the mobile-edge-cloud network tiers with the applications requirements and the complexity of their, while minimizing energy consumption, is however challenging. A possible approach to the problem consists in exploiting the emerging concept of dynamic DNNs, characterized by multi-branched architectures with early exits enabling sample-based adaptation of the model depth. We leverage this concept and address the problem of deploying portions of DNNs with early exits across the mobile-edge-cloud system and allocating therein the necessary network, computing, and memory resources. We do so by developing a 3-stage graph-modeling method that allows us to represent the characteristics of the system and the applications as well as the possible options for splitting the DNN over the multi-tier network nodes. Our solution, called Feasible Inference Graph (FIN), can determine the DNN split, deployment, and resource allocation that minimizes the inference energy consumption while satisfying the nodes’ constraints and the requirements of multiple, co-existing applications. FIN closely matches the optimum and leads to over 89% energy savings with respect to state-of-the-art alternatives.

Next-generation networks are expected to adapt not only to a wide range of application demands but also to the specific needs of individual users, end device heterogeneity, and changing operational conditions. In this work, we discuss how to enhance the programmability of network devices (e.g., radio end devices, base stations, and routers) to achieve an unprecedented level of network adaptation and customization in 6G systems. We focus on two main innovations: full network programmability and morphable networking. The former breaks down traditional boundaries between protocol layers as well as domains (e.g., network segments, technological realms, and administrative domains); the latter enables the protocol stack of each network node to be dynamically and consistently reconfigured across different domains, enabled by embedded artificial intelligence (AI).

In this paper, we investigate the interplay between early exit mechanisms in deep neural networks and privacy preservation in the context of federated learning. Our primary objective is to assess how early exits impact privacy during the learning and inference phases. Through experiments, we demonstrate that models equipped with early exits perceivably boost privacy against membership inference attacks. Our findings suggest that the inclusion of early exits in neural models can serve as a valuable tool in mitigating privacy risks while, at the same time, retaining their original advantages of fast inference.