Urban mobility is undergoing a profound transformation driven by multiple forces, including sustainability, autonomous vehicles, and safety. In this study, we focus on road safety, which remains a global concern with 1.19 million road traffic deaths reported annually by the World Health Organization. Among the key factors contributing to crash risk, infrastructure design has been recognized as one of the most actionable. This paper proposes an AI-based tool that provides automatic urban risk assessment through topological imagery, even in the absence of detailed crash records in a specific area of interest. We develop a regression model based on a modified ResNet-18 architecture, trained to predict a risk index from map-based images of urban areas and real-world telematics data, specifically harsh braking events. The model is trained on a dataset of over 80,000 harsh events collected from approximately 10,000 vehicles in the city of Milan during 2024.The results show that the model can effectively infer infrastructure-related road risk from topological inputs only, providing a lightweight and data-efficient prior for intelligent vehicles, and supporting enhanced global path planning in an autonomous driving context, even in scenarios with limited or no incident data.

High dynamic range (HDR) imaging aims to simultaneously capture a large range of illuminance levels, often observed in real world scenes. An HDR image is conventionally captured on a CMOS sensor by combining multiple frames with different exposure values. Ultimately, the dynamic range is limited by the read noise of the CMOS at low-light and by the full-well capacity at high-light. Singlephoton avalanche diode (SPAD) imagers display high HDR capabilities due to sensitivity from individual photons up to high photon fluxes. In this work, we demonstrate that fusing multi-frame raw-CMOS RGB and monochrome SPAD images can enhance the dynamic range and color accuracy in real-world settings. To fully leverage the SPAD data, we propose a Linearisation-Upsample processing block that linearises the intrinsically nonlinear response of SPAD-QIS and accounts for the low spatial resolution of current commercial SPAD technology. We demonstrate a clear advantage by including a SPAD image into a multiframe HDR pipeline with comprehensive evaluation on HDR datasets and real-world data. We also address the lack of public SPAD datasets by providing two raw CMOS-SPAD datasets for multi-frame HDR. Both datasets are available for download here: https://github.com/suonsivu/rawcmos-spad-hdr

Accurately localizing 3D objects like pedestrians, cyclists, and other vehicles is essential in Autonomous Driving. To ensure high detection performance, Autonomous Vehicles complement RGB cameras with LiDAR sensors, but effectively combining these data sources for 3D object detection remains challenging. We propose LCF3D, a novel sensor fusion framework that combines a 2D object detector on RGB images with a 3D object detector on LiDAR point clouds. By leveraging multimodal fusion principles, we compensate for inaccuracies in the LiDAR object detection network. Our solution combines two key principles: (i) late fusion , to reduce LiDAR False Positives by matching LiDAR 3D detections with RGB 2D detections and filtering out unmatched LiDAR detections; and (ii) cascade fusion , to recover missed objects from LiDAR by generating new 3D frustum proposals corresponding to unmatched RGB detections. Experiments show that LCF3D is beneficial for domain generalization, as it turns out to be successful in handling different sensor configurations between training and testing domains. LCF3D achieves significant improvements over LiDAR-based methods, particularly for challenging categories like pedestrians and cyclists in the KITTI dataset, as well as motorcycles and bicycles in nuScenes. Code can be downloaded from: https://github.com/CarloSgaravatti/LCF3D .

Most Federated Learning (FL) approaches assume a single global model is updated locally by clients and aggregated at the server. However, the single-model assumption is often too restrictive, especially in scenarios involving a large amount of different users, where adaptation to different domains and personalization are necessary to improve model performance. In this paper, we propose ClusterSSFDA, the first FL framework to leverage clustering for addressing domain shifts in Semi-Supervised Federated Learning (SSFL), where clients collect data without supervision. In particular, ClusterSSFDA clusters clients based on their models' agreement, and updates multiple models at the server, each tailored to a cluster of similar clients. ClusterSSFDA significantly improves adaptation in SSFL, striking a balance between a single global model, which may be suboptimal in the presence of domain shift among the clients, and fully personalized models, which would be trained on too small datasets. Our experiments on real-world scenarios with multiple levels of domain shift demonstrate that ClusterSSFDA outperforms existing methods, achieving superior performance in challenging SSFL settings.

Blind Spot Networks (BSNs) are powerful deep denoisers that can be trained in a self-supervised way, namely without access to clean images. The training scheme of BSN-based methods builds upon the assumption that noise is pixel-wise independent and zero-mean. However, this assumption is not satisfied in many real-world scenarios, where the noise exhibits spatial correlation that degrades BSN denoising performance. Training deep neural networks under spatially correlated noise has attracted a lot of interest. However, several approaches require complex network architectures or resource-demanding training procedures, resulting in less practical solutions than simple BSNs. In this work, we present Non-Local N2V (NL-N2V), an extension of the mainstream Noise2Void (N2V) method. It is designed to address spatially correlated noise while maintaining the simplicity and the network architecture of N2V. Our approach introduces a novel masking strategy inspired by the Non-Local Self-Similarity prior that masks small regions around blind spots and replaces these with values obtained from non-local similar patches. Tests performed on real-world datasets corrupted by correlated noise, such as SIDD and DND, show that NL-N2V considerably outperforms the traditional N2V.

Single-Photon Avalanche Diodes (SPADs) are an emerging pixel technology able to detect the arrival of single photons. Arrays of SPADs can be used to image a scene in binary frames, indicating whether there has been a photon detection in a pixel during a very short frame exposure (e.g., few microseconds). Like many imaging technologies, SPAD sensors suffer from false photon hits, called Dark Count (DC), which increase the photon count, resulting in abnormally bright pixels, called hot pixels. Correction methods used for common CCD/CMOS sensors, such as dark frame subtraction, are ineffective for SPADs, which have non-linear response in observed counts due to their 1-bit quantization. Thus, we derive an analytical model of the count bias introduced by DC, which is signal-dependent, and provide a pixel-wise, closed-form DC correction algorithm. Remarkably, our correction is optimal in expectation, meaning that on average it removes perfectly the distortion. We validate our method on real SPAD acquisitions and synthetic data, showing that it significantly improves image reconstruction.

Semi-Supervised Federated Learning (SSFL) aims to improve a pretrained model using unlabeled data from clients. Traditional SSFL solutions relying on pseudo-labels or autoencoders often struggle in the presence of domain shift, i.e. a difference in data distributions between the server and the clients. In this paper we present SemiFDA, the first solution to effectively handle domain shift in SSFL. After training an initial classifier on the server's labeled data, we establish an unsupervised learning process at clients to train feature extractors based on encoders. This process adopts a custom unsupervised loss function that promotes the clients' encoders to align their feature distributions with those extracted by the encoder at server. The updated encoders are then aggregated at the server using Federated Aver-aging and sent back for the next iteration, while the classification head remains frozen to preserve the benefits of aligning features locally. Furthermore, we design an experimental framework to mimic various levels of domain shift and test SSFL methods in real-world scenarios, including HAR and Digit Classification. Our results also demonstrate the detrimental effects of domain shift in SSFL and show that SemiFDA outperforms other solutions under these challenging conditions.

For more than a decade, deep learning models have been dominating in various 2D imaging tasks. Their application is now extending to 3D imaging, with 3D Convolutional Neural Networks (3D CNNs) being able to process LIDAR, MRI, and CT scans, with significant implications for fields such as autonomous driving and medical imaging. In these critical settings, explaining the model’s decisions is fundamental. Despite recent advances in Explainable Artificial Intelligence, however, little effort has been devoted to explaining 3D CNNs, and many works explain these models via inadequate extensions of 2D saliency methods. A fundamental limitation to the development of 3D saliency methods is the lack of a benchmark to quantitatively assess these on 3D data. To address this issue, we propose SE3D: a framework for Saliency method Evaluation in 3D imaging. We propose modifications to ShapeNet, ScanNet, and BraTS datasets, and evaluation metrics to assess saliency methods for 3D CNNs. We evaluate both state-of-the-art saliency methods designed for 3D data and extensions of popular 2D saliency methods to 3D. Our experiments show that 3D saliency methods do not provide explanations of sufficient quality, and that there is margin for future improvements and safer applications of 3D CNNs in critical fields.

The anomaly detection literature is abundant with offline methods, which require repeated access to data in memory, and impose impractical assumptions when applied to a streaming context. Existing online anomaly detection methods also generally fail to address these constraints, resorting to periodic retraining to adapt to the online context. We propose Online-iForest, a novel method explicitly designed for streaming conditions that seamlessly tracks the data generating process as it evolves over time. Experimental validation on real-world datasets demonstrated that Online-iForest is on par with online alternatives and closely rivals state-of-the-art offline anomaly detection techniques that undergo periodic retraining. Notably, Online-iForest consistently outperforms all competitors in terms of efficiency, making it a promising solution in applications where fast identification of anomalies is of primary importance such as cybersecurity, fraud and fault detection.

Advances in multi-modal embeddings, and in particular CLIP, have recently driven several breakthroughs in Computer Vision (CV). CLIP has shown impressive performance on a variety of tasks, yet, its inherently opaque architecture may hinder the application of models employing CLIP as backbone, especially in fields where trust and model explainability are imperative, such as in the medical domain. Current explanation methodologies for CV models rely on Saliency Maps computed through gradient analysis or input perturbation. However, these Saliency Maps can only be computed to explain classes relevant to the end task, often smaller in scope than the backbone training classes. In the context of models implementing CLIP as their vision backbone, a substantial portion of the information embedded within the learned representations is thus left unexplained.In this work, we propose Concept Visualization (ConVis), a novel saliency methodology that explains the CLIP embedding of an image by exploiting the multi-modal nature of the embeddings. ConVis makes use of lexical information from WordNet to compute task-agnostic Saliency Maps for any concept, not limited to concepts the end model was trained on. We validate our use of WordNet via an out of distribution detection experiment, and test ConVis on an object localization benchmark, showing that Concept Visualizations correctly identify and localize the image’s semantic content. Additionally, we perform a user study demonstrating that our methodology can give users insight on the model’s functioning.

Deep neural networks require specific layers to process point clouds, as the scattered and irregular location of 3D points prevents the use of conventional convolutional filters. We introduce the composite layer, a flexible and general alternative to the existing convolutional operators that process 3D point clouds. We design our composite layer to extract and compress the spatial information from the 3D coordinates of points and then combine this with the feature vectors. Compared to mainstream point-convolutional layers such as ConvPoint and KPConv, our composite layer guarantees greater flexibility in network design and provides an additional form of regularization. To demonstrate the generality of our composite layers, we define both a convolutional composite layer and an aggregate version that combines spatial information and features in a nonlinear manner, and we use these layers to implement CompositeNets. Our experiments on synthetic and real-world datasets show that, in both classification, segmentation, and anomaly detection, our CompositeNets outperform ConvPoint, which uses the same sequential architecture, and achieve similar results as KPConv, which has a deeper, residual architecture. Moreover, our CompositeNets achieve state-of-the-art performance in anomaly detection on point clouds. Our code is publicly available at https://github.com/sirolf-otrebla/CompositeNet.

Despite the remarkable progress of fiber optics in communication, little attention has been devoted to the automatic detection of anomalies in optical spectra, i.e., poorly transmitted channels. This task is typically addressed by ad-hoc heuristics that fall short in spectra presenting heavy distortions caused by optical amplifiers during transmission. We propose a method based on a joint optimization procedure for estimating the major trends that characterize the spectrum, enabling the detection of anomalies even in the presence of few channels and heavy distortions. Our experiments have shown that the proposed method can successfully localize anomalies achieving more than 98% accuracy, outperforming all competitors.

Traditional and deep Structure-from-Motion (SfM) methods typically operate under the assumption that the scene is rigid, i.e., the environment is static or consists of a single moving object. Few multi-body SfM approaches address the reconstruction of multiple rigid bodies in a scene but suffer from the inherent scale ambiguity of SfM, such that objects are reconstructed at inconsistent scales. We propose a depth and camera pose estimation framework to resolve the scale ambiguity in multi-body scenes. Specifically, starting from disorganized images, we present a novel multi-view scale estimator that resolves the camera pose ambiguity and a multi-body plane sweep network that generalizes depth estimation to dynamic scenes. Experiments demonstrate the advantages of our method over state-of-the-art SfM frameworks in multi-body scenes and show that it achieves comparable results in static scenes. The code and dataset are available at https://github.com/andreadalcin/MultiBodySfM.

We present Kernel QuantTree (KQT), a non-parametric change detection algorithm that monitors multivariate data through a histogram. KQT constructs a nonlinear partition of the input space that matches pre-defined target probabilities and specifically promotes compact bins adhering to the data distribution, resulting in a powerful detection algorithm. We prove two key theoretical advantages of KQT: i) statistics defined over the KQT histogram do not depend on the stationary data distribution $\phi_0$, so detection thresholds can be set a priori to control false positive rate, and ii) thanks to the kernel functions adopted, the KQT monitoring scheme is invariant to the roto-translation of the input data. Consequently, KQT does not require any preprocessing step like PCA. Our experiments show that KQT achieves superior detection power than non-parametric state-of-the-art change detection methods, and can reliably control the false positive rate.

We address the problem of online change detection in multivariate datastreams, and we introduce QuantTree Exponentially Weighted Moving Average (QT-EWMA), a nonparametric change-detection algorithm that can control the expected time before a false alarm, yielding a desired Average Run Length (ARL$_{0}$). Controlling false alarms is crucial in many applications and is rarely guaranteed by online change-detection algorithms that can monitor multivariate datastreams without knowing the data distribution. Like many change-detection algorithms, QT-EWMA builds a model of the data distribution, in our case a QuantTree histogram, from a stationary training set. To monitor datastreams even when the training set is extremely small, we propose QT-EWMA-update, which incrementally updates the QuantTree histogram during monitoring, always keeping the ARL$_{0}$under control. Our experiments, performed on synthetic and real-world datastreams, demonstrate that QT-EWMA and QT-EWMA-update control the ARL$_{0}$and the false alarm rate better than state-of-the-art methods operating in similar conditions, achieving lower or comparable detection delays.

We address the problem of detecting distribution changes in a novel batch-wise and multimodal setup. This setup is characterized by a stationary condition where batches are drawn from potentially different modalities among a set of distributions in [Formula: see text] represented in the training set. Existing change detection (CD) algorithms assume that there is a unique-possibly multipeaked-distribution characterizing stationary conditions, and in batch-wise multimodal context exhibit either low detection power or poor control of false positives. We present MultiModal QuantTree (MMQT), a novel CD algorithm that uses a single histogram to model the batch-wise multimodal stationary conditions. During testing, MMQT automatically identifies which modality has generated the incoming batch and detects changes by means of a modality-specific statistic. We leverage the theoretical properties of QuantTree to: 1) automatically estimate the number of modalities in a training set and 2) derive a principled calibration procedure that guarantees false-positive control. Our experiments show that MMQT achieves high detection power and accurate control over false positives in synthetic and real-world multimodal CD problems. Moreover, we show the potential of MMQT in Stream Learning applications, where it proves effective at detecting concept drifts and the emergence of novel classes by solely monitoring the input distribution.

We introduce Class Distribution Monitoring (CDM), an effective concept-drift detection scheme that monitors the class-conditional distributions of a datastream. In particular, our solution leverages multiple instances of an online and nonparametric change-detection algorithm based on QuantTree. CDM reports a concept drift after detecting a distribution change in any class, thus identifying which classes are affected by the concept drift. This can be precious information for diagnostics and adaptation. Our experiments on synthetic and real-world datastreams show that when the concept drift affects a few classes, CDM outperforms algorithms monitoring the overall data distribution, while achieving similar detection delays when the drift affects all the classes. Moreover, CDM outperforms comparable approaches that monitor the classification error, particularly when the change is not very apparent. Finally, we demonstrate that CDM inherits the properties of the underlying change detector, yielding an effective control over the expected time before a false alarm, or Average Run Length (ARL0).

Deep Neural Networks (DNNs) are nowadays widely used in low-cost accelerators, characterized by limited computational resources. These models, and in particular DNNs for image classification, are becoming increasingly popular in safety-critical applications, where they are required to be highly reliable. Unfortunately, increasing DNNs reliability without computational overheads, which might not be affordable in low-power devices, is a non-trivial task. Our intuition is to detect network executions affected by faults as outliers with respect to the distribution of normal network’s output. To this purpose, we propose to exploit Open-Set Recognition (OSR) techniques to perform Fault Detection in an extremely low-cost manner. In particuar, we analyze the Maximum Logit Score (MLS), which is an established Open-Set Recognition technique, and compare it against other well-known OSR methods, namely OpenMax, energy-based outof-distribution detection and ODIN. Our experiments, performed on a ResNet-20 classifier trained on CIFAR-10 and SVHN datasets, demonstrate that MLS guarantees satisfactory detection performance while adding a negligible computational overhead. Most remarkably, MLS is extremely convenient to conFigure and deploy, as it does not require any modification or re-training of the existing network. A discussion of the advantages and limitations of the analysed solutions concludes the paper.

Abstract Optical fiber links are customarily monitored by Optical Time Domain Reflectometer (OTDR), an optoelectronic instrument that measures the scattered or reflected light along the fiber and returns a signal, namely the OTDR trace . OTDR traces are typically analyzed by experts in laboratories or by hand-crafted algorithms running in embedded systems to localize critical events occurring along the fiber. In this work, we address the problem of automatically detecting optical events in OTDR traces through a deep learning model that can be deployed in embedded systems. In particular, we take inspiration from Faster R-CNN and present the first 1D object-detection neural network for OTDR traces. Thanks to an ad-hoc preprocessing pipeline for OTDR traces, we can also identify unknown events , namely events that are not represented in training data but that might indicate rare and unforeseen situations that need to be reported. The resulting network brings several advantages with respect to existing solutions, as these typically classify fixed-size windows of OTDR traces, thus are less accurate in the localization. Moreover, existing solutions do not report events that cannot be safely associated to any label in the training set. Our experiments, performed on real OTDR traces, show very promising performance, and can be directly executed on embedded OTDR devices.

Classical redundancy-based fault detection techniques, such as Duplication with Comparison (DWC), rely on replicating the computation and comparing the replicas’ output at a bit-wise granularity. In many application environments these costs are prohibitive, especially when applications are characterized by an intrinsic level of tolerance. This article presents a novel fault-detection approach for the specific context of image filtering. Peculiarity of the proposed approach is that it estimates the impact of the fault on the processed output, in order to determine whether the image is usable or should be re-processed. To limit overheads, the proposed solution exploits Approximate Computing (AC), allowing the definition of disciplined AC strategies to trade-off between accuracy and costs. Core of our solution is the successful combination of Image Quality Assessment metrics and Machine Learning models to assess the visual impact of the fault in a lightweight manner. Extensive experimental campaigns demonstrate the effectiveness of the solution, achieving achieving a reduction in terms of execution time up to 44 percent with respect to the classical DWC, with a fault detection precision ranging from 94.58 to 96.70 percent, and recall ranging from 88.2 to 97.8 percent, depending on the adopted level of approximation.

We address the problem of recovering multiple structures of different classes in a dataset contaminated by noise and outliers. In particular, we consider geometric structures defined by a mixture of underlying parametric models (e.g. planes and cylinders, homographies and fundamental matrices), and we tackle the robust fitting problem by preference analysis and clustering. We present a new algorithm, termed MultiLink, that simultaneously deals with multiple classes of models. MultiLink combines on-the-fly model fitting and model selection in a novel linkage scheme that determines whether two clusters are to be merged. The resulting method features many practical advantages with respect to methods based on preference analysis, being faster, less sensitive to the inlier threshold, and able to compensate limitations deriving from hypotheses sampling. Experiments on several public datasets demonstrate that MultiLink favourably compares with state of the art alternatives, both in multi-class and single-class problems. Code is publicly made available for download1.

Synchronization refers to the problem of inferring the unknown values attached to vertices of a graph where edges are labelled with the ratio of the incident vertices, and labels belong to a group. This paper addresses the synchronization problem on multi-graphs, that are graphs with more than one edge connecting the same pair of nodes. The problem naturally arises when multiple measures are available to model the relationship between two vertices. This happens when different sensors measure the same quantity, or when the original graph is partitioned into sub-graphs that are solved independently. In this case, the relationships among sub-graphs give rise to multi-edges and the problem can be traced back to a multi-graph synchronization. The baseline solution reduces multi-graphs to simple ones by averaging their multi-edges, however this approach falls short because: i) averaging is well defined only for some groups and ii) the resulting estimator is less precise and accurate, as we prove empirically. Specifically, we present MULTISYNC, a synchronization algorithm for multi-graphs that is based on a principled constrained eigenvalue optimization. MULTISYNC is a general solution that can cope with any linear group and we show to be profitably usable both on synthetic and real problems.

We present Augmented Grad-CAM, a general framework to provide a high-resolution visual explanation of CNN outputs. Our idea is to take advantage of image augmentation to aggregate multiple low-resolution heat-maps - in our experiments Grad-CAMs - computed from augmented copies of the same input image. We generate the high-resolution heat-map through super-resolution, and we formulate a general optimization problem based on Total Variation regularization. This problem is entirely solved on the GPU at inference time, together with image augmentation. Augmented Grad-CAM outperforms Grad-CAM in weakly supervised localization on Imagenet dataset, and provides more detailed heat-maps. Moreover, Augmented Grad-CAM turns to be particularly useful in monitoring the production of silicon wafers, where CNNs are employed to classify defective patterns on the wafer surface to detect harmful faults in the production line.

We address the problem of detecting anomalies with respect to structured patterns. To this end, we conceive a novel anomaly detection method called PIF, that combines the advantages of adaptive isolation methods with the flexibility of preference embedding. Specifically, we propose to embed the data in a high dimensional space where an efficient tree-based method, PI-Forest, is employed to compute an anomaly score. Experiments on synthetic and real datasets demonstrate that PIF favorably compares with state-of-the-art anomaly detection techniques, and confirm that PI-Forest is better at measuring arbitrary distances and isolate points in the preference space.

In a wide range of applied problems involving fluid flows, Computational Fluid Dynamics (CFD) provides detailed quantitative information on the flow field, at variable level of fidelity and computational cost. However, CFD alone cannot predict high-level functional properties that are not easily obtained from the equations of fluid motion. In this work, we present a data-driven framework to extract these additional information, such as medical diagnostic output, from CFD solutions. This is a challenging task because of the huge data dimensionality of CFD, and the limited training data that can be typically gathered due to the large computational cost of CFD. By pursuing a traditional Machine Learning (ML) pipeline of pre-processing, feature extraction, and model training, we demonstrate that informative features can be extracted from CFD data. Two experiments, pertaining to different application domains, support our claim that the convective properties implicit into a CFD solution can be leveraged to retrieve functional information that does not admit an analytical definition. Despite the preliminary nature of our study and the relative simplicity of both the geometrical and CFD models, for the first time we demonstrate that the combination of ML and CFD can diagnose a complex system in terms of high-level functional properties.

We address the problem of detecting distribution changes in multivariate data streams by means of histograms. Histograms are very general and flexible models, which have been relatively ignored in the change-detection literature as they often require a number of bins that grows unfeasibly with the data dimension. We present QuantTree, a recursive binary splitting scheme that adaptively defines the histogram bins to ease the detection of any distribution change. Our design scheme implies that i) we can easily control the overall number of bins and ii) the bin probabilities do not depend on the distribution of stationary data. This latter is a very relevant aspect in change detection, since thresholds of tests statistics based on these histograms (e.g., the Pearson statistic or the total variation) can be numerically computed from univariate and synthetically generated data, yet guaranteeing a controlled false positive rate. Our experiments show that the proposed histograms are very effective in detecting changes in high dimensional data streams, and that the resulting thresholds can effectively control the false positive rate, even when the number of training samples is relatively small.

We present a prototype wearable device able to perform online and long-term monitoring of ECG signals, and detect anomalous heartbeats such as arrhythmias. Our solution is based on user-specific dictionaries which characterizes the morphology of normal heartbeats and are learned every time the device is positioned. Anomalies are detected via an optimized sparse coding procedure, which assesses the conformance of each heartbeat to the user-specific dictionary. The dictionaries are adapted during online monitoring, to track heart rate variations occurring during everyday activities. Perhaps surprisingly, dictionary adaptation can be successfully performed by transformations that are user-independent and learned from large datasets of ECG signals.

Detecting frauds in credit card transactions is perhaps one of the best testbeds for computational intelligence algorithms. In fact, this problem involves a number of relevant challenges, namely: concept drift (customers' habits evolve and fraudsters change their strategies over time), class imbalance (genuine transactions far outnumber frauds), and verification latency (only a small set of transactions are timely checked by investigators). However, the vast majority of learning algorithms that have been proposed for fraud detection rely on assumptions that hardly hold in a real-world fraud-detection system (FDS). This lack of realism concerns two main aspects: 1) the way and timing with which supervised information is provided and 2) the measures used to assess fraud-detection performance. This paper has three major contributions. First, we propose, with the help of our industrial partner, a formalization of the fraud-detection problem that realistically describes the operating conditions of FDSs that everyday analyze massive streams of credit card transactions. We also illustrate the most appropriate performance measures to be used for fraud-detection purposes. Second, we design and assess a novel learning strategy that effectively addresses class imbalance, concept drift, and verification latency. Third, in our experiments, we demonstrate the impact of class unbalance and concept drift in a real-world data stream containing more than 75 million transactions, authorized over a time window of three years.

Intelligent buildings are typically endowed with sensing devices that are able to measure the concentration of specific contaminants in relevant zones. The collected measurements are subsequently processed by intelligent algorithms in order to enable the prompt detection and isolation of contaminant sources inside the building. Unfortunately, in real-world conditions, these sensing devices may suffer from faults affecting the sensors or the embedded electronics. Such faults, generally result in perturbed or missed data in the acquired data-stream, that can induce false alarms (or possibly missed alarms) and compromise the contaminant detection and isolation ability. This paper proposes a three-layer cognitive monitoring system for the detection and isolation of both contaminants and sensor faults in intelligent buildings. The first two layers are designed for the prompt detection of small variations in the concentration of a specific contaminant, while reducing the possible occurrence of false alarms. At the third layer, a cognitive mechanism employing a propagation model for the contaminant, which is based on the airflows between the building zones, allows to isolate the source zone and discriminate between sensor faults and the presence of a contaminant source. The proposed method is validated using a realistic 14-zone building scenario.

Successful ECG monitoring algorithms often rely on learned models to describe the heartbeats morphology. Unfortunately, when the heart rate increases the heartbeats get transformed, and a model that can properly describe the heartbeats of a specific user in resting conditions might not be appropriate for monitoring the same user during everyday activities. We model heartbeats by dictionaries yielding sparse representations and propose a novel domain-adaptation solution which transforms user-specific dictionaries according to the heart rate. In particular, we learn suitable linear transformations from a large dataset containing ECG tracings, and we show that these transformations can successfully adapt dictionaries when the heart rate changes. Remarkably, the same transformations can be used for multiple users and different sensing apparatus. We investigate the implications of our findings in ECG monitoring by wearable devices, and present an efficient implementation of an anomaly-detection algorithm leveraging such transformations.

We address change-detection problems in the challenging conditions where data are multivariate and no a priori information or experimental evidence suggests a specific family of distributions to match stationary data. In such nonparametric settings, one typically resorts to computing an empirical model for the distribution of stationary data in the form of histograms. We here analyze two ways for building histograms in the change-detection context. In particular, we consider histograms following a uniformity criterion: uniformity in the volume and uniformity in the density. In the former case the input domain is divided into a regular grid, while in the latter the input domain is adaptively partitioned to yield subsets having the same probability to contain stationary data. For both histograms we discuss nonparametric monitoring procedures which implement likelihood-based and distance-based approaches to detect changes in the distribution. In our experiments, performed both on synthetic and real-world datasets, we show that the combination of uniform density histograms and distance-based approaches achieves the best change-detection performance.

Denoising is often addressed via sparse coding with respect to an overcomplete dictionary. There are two main approaches when the dictionary is composed of translates of an orthonormal basis. The first, traditionally employed by techniques such as wavelet cycle spinning, separately seeks sparsity w.r.t. each translate of the orthonormal basis, solving multiple partial optimizations and obtaining a collection of sparse approximations of the noise-free image, which are aggregated together to obtain a final estimate. The second approach, recently employed by convolutional sparse representations, instead seeks sparsity over the entire dictionary via a global optimization. It is tempting to view the former approach as providing a suboptimal solution of the latter. In this letter, we analyze whether global sparsity is a desirable property, and under what conditions the global optimization provides a better solution to the denoising problem. In particular, our experimental analysis shows that the two approaches attain comparable performance in case of natural images and global optimization outperforms the simpler aggregation of partial estimates only when the image admits an extremely sparse representation. We explain this phenomenon by separately studying the bias and variance of these solutions, and by noting that the variance of the global solution increases very rapidly as the original signal becomes less and less sparse.

Nanoproducts represent a potential growing sector and nanofibrous materials are widely requested in industrial, medical, and environmental applications. Unfortunately, the production processes at the nanoscale are difficult to control and nanoproducts often exhibit localized defects that impair their functional properties. Therefore, defect detection is a particularly important feature in smart-manufacturing systems to raise alerts as soon as defects exceed a given tolerance level and to design production processes that both optimize the physical properties and control the defectiveness of the produced materials. Here, we present a novel solution to detect defects in nanofibrous materials by analyzing scanning electron microscope images. We employ an algorithm that learns, during a training phase, a model yielding sparse representations of the structures that characterize correctly produced nanofiborus materials. Defects are then detected by analyzing each patch of an input image and extracting features that quantitatively assess whether the patch conforms or not to the learned model. The proposed solution has been successfully validated over 45 images acquired from samples produced by a prototype electrospinning machine. The low computational times indicate that the proposed solution can be effectively adopted in a monitoring system for industrial production.

We present hierarchical change-detection tests (HCDTs), as effective online algorithms for detecting changes in datastreams. HCDTs are characterized by a hierarchical architecture composed of a detection layer and a validation layer. The detection layer steadily analyzes the input datastream by means of an online, sequential CDT, which operates as a low-complexity trigger that promptly detects possible changes in the process generating the data. The validation layer is activated when the detection one reveals a change, and performs an offline, more sophisticated analysis on recently acquired data to reduce false alarms. Our experiments show that, when the process generating the datastream is unknown, as it is mostly the case in the real world, HCDTs achieve a far more advantageous tradeoff between false-positive rate and detection delay than their single-layered, more traditional counterpart. Moreover, the successful interplay between the two layers permits HCDTs to automatically reconfigure after having detected and validated a change. Thus, HCDTs are able to reveal further departures from the postchange state of the data-generating process.

The automatic detection of anomalies, defined as patterns that are not encountered in representative set of normal images, is an important problem in industrial control and biomedical applications. We have shown that this problem can be successfully addressed by the sparse representation of individual image patches using a dictionary learned from a large set of patches extracted from normal images. Anomalous patches are detected as those for which the sparse representation on this dictionary exceeds sparsity or error tolerances. Unfortunately, this solution is not suitable for many real-world visual inspection-systems since it is not scale invariant: since the dictionary is learned at a single scale, patches in normal images acquired at a different magnification level might be detected as anomalous. We present an anomaly-detection algorithm that learns a dictionary that is invariant to a range of scale changes, and overcomes this limitation by use of an appropriate sparse coding stage. The algorithm was successfully tested in an industrial application by analyzing a dataset of Scanning Electron Microscope (SEM) images, which typically exhibit different magnification levels.

In recent years, Radio frequency (RF) sensor networks have been used to localize people indoor without requiring them to wear invasive electronic devices. These wireless mesh networks, formed by low-power radio transceivers, continuously measure the received signal strength (RSS) of the links. Radio Tomographic Imaging (RTI) is a technique that generates, starting from these RSS measurements, 2D images of the change in the electromagnetic field inside the area covered by the radio transceivers to spot the presence and movements of animates (e.g., people, large animals) or large metallic objects (e.g., cars). Here, we present a RTI system for localizing and tracking people outdoors. Differently than in indoor environments where the RSS does not change significantly with time unless people are found in the monitored area, the outdoor RSS signal is time-variant, e.g., due to rainfall or wind-driven foliage. We present a novel outdoor RTI method that, despite the nonstationary noise introduced in the RSS data by the environment, achieves high localization accuracy and dramatically reduces the energy consumption of the sensing units. Experimental results demonstrate that the system accurately detects and tracks a person in real-time in a large forested area under varying environmental conditions, significantly reducing false positives, localization error and energy consumption compared to state-of-the-art RTI methods.

We address the problem of detecting anomalies in images, specifically that of detecting regions characterized by structures that do not conform those of normal images. In the proposed approach we exploit convolutional sparse models to learn a dictionary of filters from a training set of normal images. These filters capture the structure of normal images and are leveraged to quantitatively assess whether regions of a test image are normal or anomalous. Each test image is at first encoded with respect to the learned dictionary, yielding sparse coefficient maps, and then analyzed by computing indicator vectors that assess the conformance of local image regions with the learned filters. Anomalies are then detected by identifying outliers in these indicators. Our experiments demonstrate that a convolutional sparse model provides better anomaly-detection performance than an equivalent method based on standard patch-based sparsity. Most importantly, our results highlight that monitoring the local group sparsity, namely the spread of nonzero coefficients across different maps, is essential for detecting anomalous regions.

Most fraud-detection systems (FDSs) monitor streams of credit card transactions by means of classifiers returning alerts for the riskiest payments. Fraud detection is notably a challenging problem because of concept drift (i.e. customers' habits evolve) and class unbalance (i.e. genuine transactions far outnumber frauds). Also, FDSs differ from conventional classification because, in a first phase, only a small set of supervised samples is provided by human investigators who have time to assess only a reduced number of alerts. Labels of the vast majority of transactions are made available only several days later, when customers have possibly reported unauthorized transactions. The delay in obtaining accurate labels and the interaction between alerts and supervised information have to be carefully taken into consideration when learning in a concept-drifting environment. In this paper we address a realistic fraud-detection setting and we show that investigator's feedbacks and delayed labels have to be handled separately. We design two FDSs on the basis of an ensemble and a sliding-window approach and we show that the winning strategy consists in training two separate classifiers (on feedbacks and delayed labels, respectively), and then aggregating the outcomes. Experiments on large dataset of real-world transactions show that the alert precision, which is the primary concern of investigators, can be substantially improved by the proposed approach.

We propose a novel approach to performing change-detection based on sparse representations and dictionary learning. We operate on observations that are finite support signals, which in stationary conditions lie within a union of low dimensional subspaces. We model changes as perturbations of these subspaces and provide an online and sequential monitoring solution to detect them. This approach allows extension of the change-detection framework to operate on streams of observations that are signals, rather than scalar or multi-variate measurements, and is shown to be effective for both synthetic data and on bursts acquired by rockfall monitoring systems.

Intelligent buildings are equipped with sensing systems able to measure the contaminant concentration in the different building zones for safety purposes. The aim of these systems is to promptly detect the presence of a contaminant so that appropriate actions can be taken to ensure the safety of the people. At the same time, these sensing systems, which operate in real-world conditions, suffer from noise and sensor degradation faults. Both noise and faults can induce false alarms (resulting in unnecessary disruptive actions such as building evacuation) or missed alarms (when the presence of a contaminant is not detected). This paper proposes a novel cognitive monitoring system for performing contaminant detection in intelligent buildings with real-time point-trigger sensors. The proposed system reduces the occurrence of false alarms by means of a three-layered architecture, which employs cognitive mechanisms to validate possible detections and discriminate between the presence of a real contaminant source and a degradation fault affecting the sensors of the sensing system. In addition, the proposed system is able to isolate the building zone containing the contaminant source (or the faulty sensor) and estimate the onset time of the release (or the fault).

Time-series data are often characterized by a large degree of self-similarity, which arises in application domains featuring periodicity or seasonality. While self-similarity has shown to be an effective prior for modeling real data in the signal and image-processing literature, it has received much less attention in time-series literature, where only few works leveraging the self-similarity for anomaly detection have been presented. Here we introduce a novel change-detection test to detect structural changes in time series by analyzing their self-similarity. The core of the proposed solution is the definition of a change indicator to quantitatively assesses the self-similarity of the time-series data over time. In particular, the change indicator is obtained by comparing each patch to be analyzed with its most similar counterpart in a change-free training set. Experimental results on the flow measurements in the water distribution network of the Barcelona city show the effectiveness of the proposed solution.

When our gaze fixates a point, the visual acuity is maximal at the fixation point (imaged by the fovea, i.e. the central part of the retina) and decreases rapidly towards the periphery of the visual field. This phenomenon is known as foveated vision or foveated imaging. We recently investigated the role of fovation in image filtering and we have shown that the foveated patch distance, i.e. the Euclidean distance between foveated patches, is a valuable feature for the assessment of nonlocal self-similarity. Foveation operators apply spatially variant blur, providing a compact multiscale representation of each image patch. Here, we introduce anisotropic foveation operators that embed directional point-spread functions, and we show that the operators providing the highest denoising quality are characterized by radial orientations. This result is coherent with the orientation preference in the human visual system.

Just-in-time (JIT) classifiers operate in evolving environments by classifying instances and reacting to concept drift. In stationary conditions, a JIT classifier improves its accuracy over time by exploiting additional supervised information coming from the field. In nonstationary conditions, however, the classifier reacts as soon as concept drift is detected; the current classification setup is discarded and a suitable one activated to keep the accuracy high. We present a novel generation of JIT classifiers able to deal with recurrent concept drift by means of a practical formalization of the concept representation and the definition of a set of operators working on such representations. The concept-drift detection activity, which is crucial in promptly reacting to changes exactly when needed, is advanced by considering change-detection tests monitoring both inputs and classes distributions.

Handling dynamic environments and building up algorithms operating at low supervised-sample rates are two main challenges for classification systems designed to operate in real-life scenarios. Here, changes in the probability density function of classes characterizing the data-generating process (also called concept drift) should be detected as soon as possible to prevent the classifier from becoming obsolete. Moreover, when the rate of supervised samples during the operational life is low (as in those situations where the sample inspection is costly or destructive) both detecting the change and re-training the classifier become even more critical aspects. We present an adaptive classifier that exploits both supervised and unsupervised data to monitor the process stationarity. The classifier follows the just-in-time (JIT) approach and relies on two different change-detection tests (CDTs) to reveal changes in the environment and reconfigure the classifier accordingly. The proposed solution assesses the stationary in both the joint probability density function (CDT at the classification error) and the distribution of the inputs (CDT on unlabeled data). In addition, we integrate in the JIT adaptive classifier a procedure able to handle recurrent concepts within an ensemble of classifiers framework. Experiments show that monitoring unsupervised samples and handling recurrent concepts is essential for classifying in non-stationary environments when few supervised samples are available.

Data streams from remote monitoring systems such as wireless sensor networks show immediately that the “you sample you get” statement is not always true. Not rarely, the data stream is interrupted by intermittent communication or sensors faults, resulting in missing data in the received sequence. This has a negative impact in many algorithms assuming continuous data stream; as such, the missing data must be suitably reconstructed, in order to guarantee continuous data availability. We suggest a general methodology for reconstructing missing data that exploits both temporal and spatial redundancy characterizing the phenomenon being monitored and the distributed system, a situation proper of many monitoring systems constituted by sensor and actuator networks. Temporal and spatial dependencies are learned through linear and non-linear non-parametric models, also encompassing neural -possibly recurrent- networks, which become the spatial transfer functions connecting the different views of the phenomenon under investigation. Missing data are finally reconstructed by exploiting the forecasting ability provided by such transfer functions. The experimental section shows the effectiveness of the proposed methodology.

When dealing with motion blur there is an inevitable trade-off between the amount of blur and the amount of noise in the acquired images. The effectiveness of any restoration algorithm typically depends on these amounts, and it is difficult to find their best balance in order to ease the restoration task. To face this problem, we provide a methodology for deriving a statistical model of the restoration performance of a given deblurring algorithm in case of arbitrary motion. Each restoration-error model allows us to investigate how the restoration performance of the corresponding algorithm varies as the blur due to motion develops. Our modeling treats the point-spread-function trajectories as random processes and, following a Monte-Carlo approach, expresses the restoration performance as the expectation of the restoration error conditioned on some motion-randomness descriptors and on the exposure time. This allows to coherently encompass various imaging scenarios, including camera shake and uniform (rectilinear) motion, and, for each of these, identify the specific exposure time that maximizes the image quality after deblurring.

We propose a powerful video filtering algorithm that exploits temporal and spatial redundancy characterizing natural video sequences. The algorithm implements the paradigm of nonlocal grouping and collaborative filtering, where a higher dimensional transform-domain representation of the observations is leveraged to enforce sparsity, and thus regularize the data: 3-D spatiotemporal volumes are constructed by tracking blocks along trajectories defined by the motion vectors. Mutually similar volumes are then grouped together by stacking them along an additional fourth dimension, thus producing a 4-D structure, termed group, where different types of data correlation exist along the different dimensions: local correlation along the two dimensions of the blocks, temporal correlation along the motion trajectories, and nonlocal spatial correlation (i.e., self-similarity) along the fourth dimension of the group. Collaborative filtering is then realized by transforming each group through a decorrelating 4-D separable transform and then by shrinkage and inverse transformation. In this way, the collaborative filtering provides estimates for each volume stacked in the group, which are then returned and adaptively aggregated to their original positions in the video. The proposed filtering procedure addresses several video processing applications, such as denoising, deblocking, and enhancement of both grayscale and color data. Experimental results prove the effectiveness of our method in terms of both subjective and objective visual quality, and show that it outperforms the state of the art in video denoising.

The pervasiveness of mobile devices increases the risk of exposing sensitive information on the go. In this paper, we arise this concern by presenting an automatic attack against modern touchscreen keyboards. We demonstrate the attack against the Apple iPhone - 2010's most popular touchscreen device - although it can be adapted to other devices (e.g., Android) that employ similar key-magnifying keyboards. Our attack processes the stream of frames from a video camera (e.g., surveillance or portable camera) and recognizes keystrokes online, in a fraction of the time needed to perform the same task by direct observation or offline analysis of a recorded video, which can be unfeasible for large amount of data. Our attack detects, tracks, and rectifies the target touchscreen, thus following the device or camera's movements and eliminating possible perspective distortions and rotations In real-world settings, our attack can automatically recognize up to 97.07 percent of the keystrokes (91.03 on average), with 1.15 percent of errors (3.16 on average) at a speed ranging from 37 to 51 keystrokes per minute.

Classification systems designed to work in nonstationary conditions rely on the ability to track the monitored process by detecting possible changes and adapting their knowledge-base accordingly. Adaptive classifiers present in the literature are effective in handling abrupt concept drifts (i.e., sudden variations), but, unfortunately, they are not able to adapt to gradual concept drifts (i.e., smooth variations) as these are, in the best case, detected as a sequence of abrupt concept drifts. To address this issue we introduce a novel adaptive classifier that is able to track and adapt its knowledge base to gradual concept drifts (modeled as polynomial trends in the expectations of the conditional probability density functions of input samples), while maintaining its effectiveness in dealing with abrupt ones. Experimental results show that the proposed classifier provides high classification accuracy both on synthetically generated datasets and measurements from real sensors.

Design of applications working in nonstationary environments requires the ability to detect and anticipate possible behavioral changes affecting the system under investigation. In this direction, the literature provides several tests aiming at assessing the stationarity of a data generating process; of particular interest are nonparametric sequential change-point detection tests that do not require any a-priori information regarding both process and change. Moreover, such tests can be made automatic through an on-line inspection of sequences of data, hence making them particularly interesting to address real applications. Following this approach, we suggest a novel two-level hierarchical change-detection test designed to detect possible occurrences of changes by observing incoming measurements. This hierarchical solution significantly reduces the number of false positives at the expenses of a negligible increase of false negatives and detection delays. Experiments show the effectiveness of the proposed approach both on synthetic dataset and measurements from real applications.

In this paper we consider the restoration of images corrupted by both uniform motion blur and Poissonian noise. We formulate an image formation model that explicitly takes into account the length of the blur point-spread function and the noise level as functions of the exposure time. Further, we present an analysis of the achievable restoration performance by showing how the root mean squared error varies with respect to the exposure time. It turns out that the worst situations are represented by either too short or too long exposure times. In between there exists an optimal exposure time that maximizes the restoration performance, balancing the amount of blur and noise in the observation. We justify such result through a mathematical analysis of the signal-to-noise ratio in Fourier domain; this study is then validated by deblurring synthetic data as well as camera raw data.

Designing tests able to effectively detect changes in the stationarity of a process generating data is a challenging problem, in particular when the process is unknown, and the only information available has to be extracted from a set of observations. This work proposes a novel approach for detecting changes in a process generating data whose distribution is unknown. Peculiarity of the approach is the use of the Intersection of Confidence Intervals (ICI) rule to monitor the process evolution. A change detection test derived from this approach is also presented. Experimental results show that the proposed test outperforms state-of-the art solutions, both in terms of efficiency and effectiveness, in particular when a reduced test configuration set is available.

A classifier expected to work in a non-stationary environment has to: (i) detect changes in the process generating the data; (ii) suitably react to the change by adapting to the new working condition. Just-in-time adaptive classifiers, a classification structure addressing stationary and nonstationary conditions, have been presented to the computational intelligence community. Such classifiers require a temporal detection of a (possible) process deviation followed by an adaptive management of the knowledge base characterizing the classifier to cope with the process change. This paper improves just-in-time adaptive classifiers by integrating temporal information about the state of the process under monitoring. An index for the process deviation is defined which, coupled with an adaptive weighted k-NN classifier, shows to be particularly effective in dealing with smooth process drifts and ageing phenomena.

We present a novel calibration method for off-axis catadioptric cameras, i.e. standard perspective cameras placed in a generic position w.r.t. an axial-symmetric mirror of unknown shape. The proposed method estimates the intrinsic parameters of the natural perspective camera, the 3D shape of the mirror and its pose w.r.t. the camera. The peculiarity of our approach is that, unlike several other calibration methods, we do not require any cross section of the mirror to be visible in the image. Instead, we require that the catadioptric image contains at least the image of one generic space line. We then derive some constraints that, combined with the harmonic homology relating the apparent contours of the mirror, allow us to calibrate the off-axis camera. We provide experimental results both on synthetic and camera images that prove the validity of the technique.