Deep learning models excel in computer vision tasks with large labeled datasets but often struggle with performance degradation under domain shifts to unlabeled target domains. Unsupervised Domain Adaptation (UDA) mitigates this challenge by transferring knowledge from a labeled source domain to an unlabeled target domain. In this paper, we propose Self-Contrastive Learning for Domain Adaptation (SCoDA), a novel framework that combines self-supervised contrastive learning with distributional alignment techniques to learn domain-invariant representations. SCoDA jointly optimizes domain alignment and classification by using contrastive loss to reduce feature gaps while preserving task performance through source supervision. Experiments on UDA benchmarks demonstrate that SCoDA outperforms traditional baselines, achieving accuracy improvements without relying on target labels. These results highlight SCoDA’s effectiveness in addressing complex domain shifts and its potential for real-world applications. Our code is available at https://github.com/viniavena/SCoDA.

Wi-Fi datasets play a crucial role in wireless networking research. They are often the result of extensive, demanding measurement campaigns. Unfortunately, researchers lack a clear assessment of where to place probes and when they have collected enough data and still obtain a representative view of the environment. Our goal is to make this process more efficient while preserving its rigor. We propose a framework that incorporates a calibration phase to evaluate the representativeness of a dataset. To this end, we use Earth Mover’s Distance (EMD) as a similarity metric to quantify data distribution differences and avoid redundant data captures. Through experimental campaigns in three distinct environments, we demonstrate that achieving a significant reduction in data collection effort is possible without compromising measurement reliability.

This work presents a proactive computer vision-based ADAS system designed to enhance VRU safety in risk zones at the rear of the vehicle and filter lanes. The system combines object detection, multi-object tracking, and risk assessment to generate real-time proximity alerts. Compared to a detection-only approach, the proposed method shows a 19% improvement in precision and a 10% increase in F1-score, while introducing a latency overhead of less than 1 ms. Evaluated on edge devices, the system maintains efficient performance across resource-constrained platforms. To support research in this domain, we introduce FilterLane-VRU, a new dataset of rearview urban traffic scenarios with temporal proximity annotations. Results validate the feasibility of deploying the system in real-world ADAS applications, offering a reliable and cost-effective solution for VRU safety.

Deep Neural Networks (DNNs) have demonstrated exceptional performance in diverse tasks. However, deploying DNNs on resource-constrained devices presents challenges due to energy consumption and delay overheads. To mitigate these issues, early-exit DNNs (EE-DNNs) incorporate exit branches within intermediate layers to enable early inferences. These branches estimate prediction confidence and employ a fixed threshold to determine early termination. Nonetheless, fixed thresholds yield suboptimal performance in dynamic contexts, where context refers to distortions caused by environmental conditions, in image classification, or variations in input distribution due to concept drift, in NLP. In this article, we introduce Upper Confidence Bound in EE-DNNs (UCBEE), an online algorithm that dynamically adjusts early exit thresholds based on context. UCBEE leverages confidence levels at intermediate layers and learns without the need for true labels. Through extensive experiments in image classification and NLP, we demonstrate that UCBEE achieves logarithmic regret, converging after just a few thousand observations across multiple contexts. We evaluate UCBEE for image classification and text mining. In the latter, we show that UCBEE can reduce cumulative regret and lower latency by approximately 10%–20% without compromising accuracy when compared to fixed threshold alternatives. Our findings highlight UCBEE as an effective method for enhancing EE-DNN efficiency.

The emergence of software-defined vehicles has brought machine learning into the vehicular domain. To support these data-driven applications, techniques to incentivize users to share their vehicle data are crucial. Federated learning trains machine learning models in a distributed manner, leveraging client data without compromising its privacy. Nonetheless, in vehicular networks, the dynamic behavior of nodes affects client availability and the global model's performance. Accordingly, this paper evaluates federated learning (FL) in a realistic vehicular network topology, accounting for real vehicle traffic in two Brazilian urban areas. The network simulation covers 3.7 km2with road speeds and 1,290 vehicles per hour, based on real data. We observe a performance decay in urban areas with longer vehicle permanence. Interestingly, longer vehicle participation in FL training leads to a biased model, with reduced generalization. We then improve our investigation based on the Dice-Sorensen coefficient to enhance vehicle variability over time. With 47% fewer vehicles per round, we achieve faster learning, higher convergence in the first 15 rounds, and equivalent final accuracy of 93 %.

O-RAN architecture introduces a new level of flexibility in managing Radio Access Networks (RANs), facilitating the development of different applications. One of these applications is spectrum sharing, in which cellular traffic can share the unlicensed band with WLAN technologies, such as Wi-Fi. A key component of this application is a spectrum classification unit that identifies the communication technology used in the medium to support decision making in the RAN. This classification can be performed using Deep Neural Networks (DNNs) that receive I/Q samples and infer which communication technology is generating the traffic. Despite the high accuracy of DNNs in this task, the inference must be performed quickly to allow timely action to avoid interference. One promising approach to enhancing the performance of DNNs is to use early-exit DNNs (EE-DNNs), which are designed to reduce computations by allowing the inference process to terminate at intermediate layers when a certain confidence level is achieved. In this paper, we explore the application of EE-DNNs for spectrum classification by applying early exits to the Convolutional Neural Network (CNN) used by the ChARM (Channel-Aware Reacting Mechanism) framework. Using the ChARM dataset, we show that an EE-DNN can accelerate inference by 10% and even achieve higher accuracy than a conventional CNN by approximately 2%.

Deep Neural Networks (DNNs) are widely used to solve a growing number of tasks, such as image classification. However, their deployment at resource-constrained devices still poses challenges related to energy consumption and delay over-heads. Early-Exit DNNs (EE-DNNs) address the challenges by adding side branches through their architecture. Under an edge-cloud co-inference, if the confidence at a side branch is larger than a fixed confidence threshold, the inference is performed completely at the edge device, saving computation for more difficult observations. Otherwise, the edge device offloads the inference task to the cloud, incurring overhead. Despite its success, EE-DNNs for image classification have to cope with distorted images. The baseline distortion level depends on the environmental context, e.g., time of the day, lighting, and weather conditions. To cope with varying distortion, we propose Adaptive Early-Exit in Deep Neural Networks (AdaEE), a novel algorithm to dynamically adjust the confidence threshold based on context, leveraging the Upper Confidence Bound (UCB) for that matter. AdaEE provably achieves logarithmic regret under mild conditions. We experimentally verify that 1) convergence occurs after collecting a few thousand observations for images with different distortion levels and overhead values, and 2) AdaEE obtains a lower cumulative regret when compared against alternatives using the Caltech-256 dataset subject to varying distortion.

Image recognition is an essential task in several mobile applications. For instance, a smartphone can process a landmark photo to gather more information about its location. If the device does not have enough computational resources available, it offloads the processing task to a cloud infrastructure. Although this approach solves resource shortages, it introduces a communication delay. Image-recognition caches on the Internet’s edge can mitigate this problem. These caches run on servers close to mobile devices and stores information about previously recognized images. If the server receives a request with a photo stored in its cache, it replies to the device, avoiding cloud offloading. The main challenge for this cache is to verify if the received image matches a stored one. Furthermore, for outdoor photos, it is difficult to compare them if one was taken in the daytime and the other at nighttime. In that case, the cache might wrongly infer that they refer to different places, offloading the processing to the cloud. This work shows that a well-known generative adversarial network, called ToDayGAN, can solve this problem by generating daytime images using nighttime ones. We can thus use this translation to populate a cache with synthetic photos that can help image matching. We show that our solution reduces cloud offloading and, therefore, the application’s latency.

Edge offloading for deep neural networks (DNNs) can be adaptive to the input's complexity by using early-exit DNNs. These DNNs have side branches throughout their architecture, allowing the inference to end earlier in the edge. The branches estimate the accuracy for a given input. If this estimated accuracy reaches a threshold, the inference ends on the edge. Otherwise, the edge offloads the inference to the cloud to process the remaining DNN layers. However, DNNs for image classification deals with distorted images, which negatively impact the branches' estimated accuracy. Consequently, the edge offloads more inferences to the cloud. This work introduces expert side branches trained on a particular distortion type to improve robustness against image distortion. The edge detects the distortion type and selects appropriate expert branches to perform the inference. This approach increases the estimated accuracy on the edge, improving the offloading decisions. We validate our proposal in a realistic scenario, in which the edge offloads DNN inference to Amazon EC2 instances.

Mobile devices can offload deep neural network (DNN)-based inference to the cloud, overcoming local hardware and energy limitations. However, offloading adds communication delay, thus increasing the overall inference time, and hence it should be used only when needed. An approach to address this problem consists of the use of adaptive model partitioning based on early-exit DNNs. Accordingly, the inference starts at the mobile device, and an intermediate layer estimates the accuracy: If the estimated accuracy is sufficient, the device takes the inference decision; Otherwise, the remaining layers of the DNN run at the cloud. Thus, the device offloads the inference to the cloud only if it cannot classify a sample with high confidence. This offloading requires a correct accuracy prediction at the device. Nevertheless, DNNs are typically miscalibrated, providing overconfident decisions. This work shows that the employment of a miscalibrated early-exit DNN for offloading via model partitioning can significantly decrease inference accuracy. In contrast, we argue that implementing a calibration algorithm prior to deployment can solve this problem, allowing for more reliable offloading decisions.

Deep Neural Network (DNN) inference requires high computation power, which generally involves a cloud infrastructure. However, sending raw data to the cloud can increase the inference time due to the communication delay. To reduce this delay, the first DNN layers can be executed at an edge infrastructure and the remaining ones at the cloud. Depending on which layers are processed at the edge, the amount of data can be highly reduced. However, executing layers at the edge can increase the processing delay. A partitioning problem tries to address this trade-off, choosing the set of layers to be executed at the edge to minimize the inference time. In this work, we address the problem of partitioning a BranchyNet, which is a DNN type where the inference can stop at the middle layers. We show that this partitioning can be treated as the shortest path problem, and thus solved in polynomial time.

A hot topic in data center design is to envision geo-distributed architectures spanning a few sites across wide area networks, allowing more proximity to the end users and higher survivability, defined as the capacity of a system to operate after failures. As a shortcoming, this approach is subject to an increase of latency between servers, caused by their geographic distances. In this paper, we address the trade-off between latency and survivability in geo-distributed data centers, through the formulation of an optimization problem. Simulations considering realistic scenarios show that the latency increase is significant only in the case of very strong survivability requirements, whereas it is negligible for moderate survivability requirements. For instance, the worst-case latency is less than 4 ms when guaranteeing that 80% of the servers are available after a failure, in a network where the latency could be up to 33 ms.

The network infrastructure plays an important role for datacenter applications. Therefore, datacenter network architectures are designed with three main goals: bandwidth, latency and reliability. This work focuses on the last goal and provides a comparative analysis of the topologies of prevalent datacenter architectures. Those architectures use a network based only on switches or a hybrid scheme of servers and switches to perform packet forwarding. We analyze failures of the main networking elements (link, server, and switch) to evaluate the tradeoffs of the different datacenter topologies. Considering only the network topology, our analysis provides a baseline study to the choice or design of a datacenter network with regard to reliability. Our results show that, as the number of failures increases, the considered hybrid topologies can substantially increase the path length, whereas servers on the switch-only topology tend to disconnect more quickly from the main network.

Xen is a tool for hardware virtualization often used to build virtual routers. Xen, however, does not assure the fundamental requirement of network isolation among these routers. This work proposes XTC (Xen Throughput Control) to fill this gap, and therefore, to guarantee multiple network coexistence without interference. XTC sets the amount of CPU allocated to each virtual router according to the maximum throughput allowed. Xen behavior is modeled by using experimental data, and based on these data, XTC is designed using feedback control. Results obtained in a testbed demonstrate the XTC ability to isolate virtual network capacities and to adapt to system changes.

As of today, many routing protocols for wireless mesh networks have been proposed. Nevertheless, quite a few take the high loss rate of control packets into account. This work analyzes the problem of consistent routing information among wireless network nodes. To accomplish this, we propose a metric to evaluate the level of inconsistency among routing tables. Our experimental analysis demonstrates that the high loss rates seen in indoor environments negatively influence route computation. In addition, we demonstrate that the high network dynamics leads to severe instability in next hop selection. Results show that the effect of loss is significant and that the simple manipulation of routing protocol configuration parameters may be not enough to cope with the problem.