Unlabeled LiDAR logs, in autonomous driving applications, are inherently a gold mine of dense 3D geometry hiding in plain sight - yet they are almost useless without human labels, highlighting a dominant cost barrier for autonomous-perception research. In this work we tackle this bottleneck by leveraging temporal-geometric consistency across LiDAR sweeps to lift and fuse cues from text and 2 Dvision foundation models directly into 3D, without any manual input. We introduce an unsupervised multimodal pseudo-labeling method relying on strong geometric priors learned from temporally accumulated LiDAR maps, alongside with a novel iterative update rule that enforces joint geometric-semantic consistency, and vice-versa detecting moving objects from inconsistencies. Our method simultaneously produces 3D semantic labels, 3D bounding boxes, and dense LiDAR scans, demonstrating robust generalization across three datasets. We experimentally validate that our method compares favorably to existing semantic segmentation and object detection pseudo-labeling methods, which often require additional manual supervision. We confirm that even a small fraction of our geometrically consistent, densified LiDAR improves depth prediction by 51.5 % and 22.0 % MAE in the 80-150 and 150-250 meters range, respectively.

Few Shot Segmentation aims to segment novel object classes given only a handful of labeled examples, enabling rapid adaptation with minimal supervision. Current literature crucially lacks a selection method that goes beyond visual similarity between the query and example images, leading to suboptimal predictions. We present MARS, a plug-and-play ranking system that leverages multimodal cues to filter and merge mask proposals robustly. Starting from a set of mask predictions for a single query image, we score, filter, and merge them to improve results. Proposals are evaluated using multimodal scores computed at local and global levels. Extensive experiments on COCO-20i, Pascal-5i, LVIS-92i, and FSS-1000 demonstrate that integrating all four scoring components is crucial for robust ranking, validating our contribution. As MARS can be effortlessly integrated with various mask proposal systems, we deploy it across a wide range of top-performing methods and achieve new state-of-the-art results on multiple existing benchmarks. Code is available on GitHub.1

The design of accurate Digital Twins (DTs) of electromagnetic environments strictly depends on the fidelity of the underlying environmental modeling. Evaluating the differences among diverse levels of modeling accuracy is key to determine the relevance of the model features towards both efficient and accurate DT simulations. In this paper, we propose two metrics, the Hausdorff ray tracing (HRT) and chamfer ray tracing (CRT) distances, to consistently compare the temporal, angular and power features between two ray tracing simulations performed on 3D scenarios featured by environmental changes. To evaluate the introduced metrics, we considered a high-fidelity digital twin model of an area of Milan, Italy and we enriched it with two different types of environmental changes: (i) the inclusion of parked vehicles meshes, and (ii) the segmentation of the buildings facade faces to separate the windows mesh components from the rest of the building. We performed grid-based and vehicular ray tracing simulations at 28 GHz carrier frequency on the obtained scenarios integrating the NVIDIA Sionna RT ray tracing simulator with the SUMO vehicular traffic simulator. Both the HRT and CRT metrics highlighted the areas of the scenarios where the simulated radio propagation features differ owing to the introduced mesh integrations, while the vehicular ray tracing simulations allowed to uncover the distance patterns arising along realistic vehicular trajectories.

Smart eyewear is emerging as an always-on platform capable of perceiving the environment and inferring intent through eye movements, making pupil tracking essential for personalized interaction. However, reliable tracking on wearable hardware remains challenging due to strict power limits and scarce annotated data. Event-based cameras offer a low-power, microsecond-latency solution, but labeled recordings still remain limited. We address this issue with a training framework that combines limited annotated real data with synthetic events and unlabeled real recordings, learning event-based pupil trackers with strong real-world generalization. We pair the U2Eyes tool with the v2e event camera simulator to generate realistic event streams, showing that networks trained on these events exhibit smaller sim-to-real gaps than networks trained on synthetic images. Moreover, our training procedure further bridges this gap, enabling our models to outperform networks trained exclusively on real data across all benchmarks, advancing toward more robust event-based eye tracking on wearable platforms.

Survival analysis is a statistical framework for modeling time-to-event data. It plays a pivotal role in medicine, reliability engineering, and social science research, where understanding event dynamics even with few data samples is critical. Recent advancements in machine learning, particularly those employing neural networks and decision trees, have introduced sophisticated algorithms for survival modeling. However, many of these methods rely on restrictive assumptions about the underlying event-time distribution, such as proportional hazard, time discretization, or accelerated failure time. In this study, we propose FPBoost, a survival model that combines a weighted sum of fully parametric hazard functions with gradient boosting. Distribution parameters are estimated with decision trees trained by maximizing the full survival likelihood. We show how FPBoost is a universal approximator of hazard functions, offering full event-time modeling flexibility while maintaining interpretability through the use of well-established parametric distributions. We evaluate concordance and calibration of FPBoost across multiple benchmark datasets, showcasing its robustness and versatility as a new tool for survival estimation.

Agriculture, fundamental for human sustenance, faces unprecedented challenges. The need for efficient, human-cooperative, and sustainable farming methods has never been greater. The core contributions of this work involve leveraging Active Vision (AV) techniques and ZeroShot Learning (ZSL) to improve the robot's ability to perceive and interact with agricultural environment in the context of fruit harvesting. The AV Pipeline implemented within ROS 2 integrates the Next-Best View (NBV) Planning for 3D environment reconstruction through a dynamic 3D Occupancy Map. Our system allows the robotics arm to dynamically plan and move to the most informative viewpoints and explore the environment, updating the 3D reconstruction using semantic information produced through ZSL models. Simulation and real-world experimental results demonstrate our system's effectiveness in complex visibility conditions, outperforming traditional and static predefined planning methods. ZSL segmentation models employed, such as YOLO World + EfficientViT SAM, exhibit high-speed performance and accurate segmentation, allowing flexibility when dealing with semantic information in unknown agricultural contexts without requiring any fine-tuning process.

Event-based cameras are becoming a popular solution for efficient, low-power eye tracking. Due to the sparse and asynchronous nature of event data, they require less processing power and offer latencies in the microsecond range. However, many existing solutions are limited to validation on powerful GPUs, with no deployment on real embedded devices. In this paper, we present EETnet, a convolutional neural network designed for eye tracking using purely event-based data, capable of running on microcontrollers with limited resources. Additionally, we outline a methodology to train, evaluate, and quantize the network using a public dataset. Finally, we propose two versions of the architecture: a classification model that detects the pupil on a grid superimposed on the original image, and a regression model that operates at the pixel level.

Event-based eye tracking represents an innovative approach to analyzing eye movements, leveraging event cameras’ high temporal resolution and asynchronous nature. This paper provides a comprehensive field review, focusing on datasets, algorithms, and challenges. It examines key datasets, highlighting their diversity and limitations, and categorizes algorithms into two core approaches: frame-based and spiking neural network (SNN)-based approaches. Frame-based methods leverage traditional techniques and deep learning, while SNNs represent an emerging field offering biologically inspired, energy-efficient solutions. However, the computational demands of emerging deep-learning methods raise questions about their feasibility for deployment on consumer-grade devices, particularly embedded systems like smart eyewear. This review provides a structured analysis of current advancements, datasets, and challenges, offering insights to guide the development of efficient and deployable event-based eye-tracking systems.

Scene view synthesis, which generates novel views from limited perspectives, is increasingly vital for applications like virtual reality, augmented reality, and robotics. Unlike object-based tasks, such as generating 360° views of a car, scene view synthesis handles entire environments where non-uniform observations pose unique challenges for stable rendering quality. To address this issue, we propose a novel approach: renderability field-guided gaussian splatting (RF-GS). This method quantifies input inhomogeneity through a renderability field, guiding pseudo-view sampling to enhanced visual consistency. To ensure the quality of wide-baseline pseudo-views, we train an image restoration model to map point projections to visiblelight styles. Additionally, our validated hybrid data optimization strategy effectively fuses information of pseudo-view angles and source view textures. Comparative experiments on simulated and real-world data show that our method outperforms existing approaches in rendering stability.

Real-time open-vocabulary scene understanding is essential for efficient 3D perception in applications such as vision-language navigation, embodied intelligence, and augmented reality. However, existing methods suffer from imprecise instance segmentation, static semantic updates, and limited handling of complex queries. To address these issues, we present OpenFusion++, a TSDF-based real-time 3D semantic-geometric reconstruction system. Our approach refines 3D point clouds by fusing confidence maps from foundational models, dynamically updates global semantic labels via an adaptive cache based on instance area, and employs a dual-path encoding framework that integrates object attributes with environmental context for precise query responses. Experiments on the ICL, Replica, ScanNet, and ScanNet++ datasets demonstrate that OpenFusion++ significantly outperforms the baseline in both semantic accuracy and query responsiveness.

The continuous evolution of side-channel analysis motivates a continuous investigation to deliver novel countermeasures. This work presents a hiding countermeasure leveraging a randomized Dynamic Frequency Scaling (DFS) actuator built on top of the clocking resources available in modern FPGAs. In contrast to state-of-the-art DFS-based solutions, our approach is meant to optimize security and performance metrics with a modest increase in power consumption. We experimentally validated our countermeasure on real hardware by comparing it against recently proposed hiding methods employing clock desynchronization. To strengthen our security assessment, we also considered a large variety of state-of-the-art side-channel attacks, including recent deep-learning ones. The experimental results confirm that none of the evaluated attack techniques can breach our protected target, and TLVA shows no information leakage with 10 million traces. The performance overhead is zero, while the power overhead is limited to 1.55×.

The whole is greater than the sum of its parts, even in 3D-text contrastive learning. We introduce SceneForge, a novel framework that enhances contrastive alignment between 3D point clouds and text through structured multi-object scene compositions. SceneForge leverages individual 3D shapes to construct multi-object scenes with explicit spatial relations, pairing them with coherent multi-object descriptions refined by a large language model. By augmenting contrastive training with these structured, compositional samples, SceneForge effectively addresses the scarcity of large-scale 3D-text datasets, significantly enriching data complexity and diversity. We systematically investigate critical design elements, such as the optimal number of objects per scene, the proportion of compositional samples in training batches, and scene construction strategies. Extensive experiments demonstrate that SceneForge delivers substantial performance gains across multiple tasks, including zero-shot classification on ModelNet, ScanObjNN, Objaverse-LVIS, and ScanNet, as well as few-shot part segmentation on ShapeNetPart. SceneForge’s compositional augmentations are model-agnostic, consistently improving performance across multiple encoder architectures. Moreover, SceneForge improves 3D visual question answering on ScanQA, generalizes robustly to retrieval scenarios with increasing scene complexity, and showcases spatial reasoning capabilities by adapting spatial configurations to align precisely with textual instructions.

Wireless communication systems can significantly benefit from the availability of spatially consistent representations of the wireless channel to efficiently perform a wide range of communication tasks. Towards this purpose, channel charting has been introduced as an effective unsupervised learning technique to achieve both locally and globally consistent radio maps. In this letter, we propose Chartwin, a case study on the integration of localization-oriented channel charting with dynamic Digital Network Twins (DNTs). Numerical results showcase the significant performance of semi-supervised channel charting in constructing a spatially consistent chart of the considered extended urban environment. The considered method results in ≈ 4.5 m localization error for the static DNT and ≈ 6 m in the dynamic DNT, fostering DNT-aided channel charting and localization.

Generative diffusion models have emerged as powerful tools to synthetically produce training data, offering potential solutions to data scarcity and reducing labelling costs for downstream supervised deep learning applications. However, existing approaches for synthetic dataset generation face significant limitations: previous methods like Knowledge Recycling rely on label-conditioned generation with models trained from scratch, limiting flexibility and requiring extensive computational resources, while simple class-based conditioning fails to capture the semantic diversity and intra-class variations found in real datasets. Additionally, effectively leveraging text-conditioned image generation for building classifier training sets requires addressing key issues: constructing informative textual prompts, adapting generative models to specific domains, and ensuring robust performance. This paper proposes the Text-Conditioned Knowledge Recycling (TCKR) pipeline to tackle these challenges. TCKR combines dynamic image captioning, parameter-efficient diffusion model fine-tuning, and Generative Knowledge Distillation techniques to create synthetic datasets tailored for image classification. The pipeline is rigorously evaluated on ten diverse image classification benchmarks. The results demonstrate that models trained solely on TCKR-generated data achieve classification accuracies on par with (and in several cases exceeding) models trained on real images. Furthermore, the evaluation reveals that these synthetic-data-trained models exhibit substantially enhanced privacy characteristics: their vulnerability to Membership Inference Attacks is significantly reduced, with the membership inference AUC lowered by 5.49 points on average compared to using real training data, demonstrating a substantial improvement in the performance-privacy trade-off. These findings indicate that high-fidelity synthetic data can effectively replace real data for training classifiers, yielding strong performance whilst simultaneously providing improved privacy protection as a valuable emergent property. The code and trained models are available in the accompanying open-source repository . • Introduces TCKR: text-conditioned diffusion + LoRA + distillation for synthetic data. • Proposes dynamic captioning (BLIP-2) to craft instance-specific prompts for images. • Synthetic-trained classifiers match or surpass real-trained ones on 10 benchmarks. • Synthetic training lowers MIA AUC by up to 8.14 vs models trained on real data. • Increasing synthetic size lifts accuracy but raises MIA risk.

As is well known, the Turing Test proposes the possibility of distinguishing the behavior of a machine from that of a human being through an experimental session. The Turing Test assesses whether a person asking questions to two different entities, can tell from their answers which of them is the human being and which is the machine. With the progress of Artificial Intelligence, the number of contexts in which the capacities of response of a machine will be indistinguishable from those of a human being is expected to increase rapidly. In order to configure a Turing Test in which it is possible to distinguish human behavior from machine behavior independently from the advances of Artificial Intelligence, at least in the short-medium term, it would be important to base it not on the differences between man and machine in terms of performance and dialogue capacity, but on some specific characteristic of the human mind that cannot be reproduced by the machine even in principle. We studied a new kind of test based on the hypothesis that such characteristic of the human mind exists and can be made experimentally evident. This peculiar characteristic is the emotional content of human cognition and, more specifically, its link with memory enhancement. To validate this hypothesis we recorded the EEG signals of 39 subjects that underwent a specific test and analyzed their signals with a neural network able to label similar signal patterns with similar binary codes. The results showed that, with a statistically significant difference, the test participants more easily recognized images associated in the past with an emotional reaction than those not associated with such a reaction. This distinction in our view is not accessible to a software system, even AI-based, and a Turing Test based on this feature of the mind may make distinguishable human versus machine responses.

Generative artificial intelligence has transformed the generation of synthetic data, providing innovative solutions to challenges like data scarcity and privacy, which are particularly critical in fields such as medicine. However, the effective use of this synthetic data to train high-performance models remains a significant challenge. This paper addresses this issue by introducing Knowledge Recycling (KR), a pipeline designed to optimise the generation and use of synthetic data for training downstream classifiers. At the heart of this pipeline is Generative Knowledge Distillation, the proposed technique that significantly improves the quality and usefulness of the information provided to classifiers through a synthetic dataset regeneration and soft labelling mechanism. The KR pipeline has been tested on a variety of datasets, with a focus on six highly heterogeneous medical image datasets, ranging from retinal images to organ scans. The results show a significant reduction in the performance gap between models trained on real and synthetic data, with models based on synthetic data outperforming those trained on real data in some cases. Furthermore, the resulting models show almost complete immunity to Membership Inference Attacks, manifesting privacy properties missing in models trained with conventional techniques. • Synthetic data can create more private and high-performance image classifiers. • Larger synthetic datasets improve classifier test accuracy. • The Knowledge Recycling pipeline excels on small medical datasets. • Tuning dataset size, generator std and recycling rate enhances classifier performance. • Knowledge Recycling makes models highly resistant to Membership Inference Attacks.

Federated learning has emerged as a paradigm for collaborative learning, enabling the development of robust models without the need to centralise sensitive data. However, conventional federated learning techniques have privacy and security vulnerabilities due to the exposure of models, parameters or updates, which can be exploited as an attack surface. This paper presents Federated Knowledge Recycling (FedKR), a cross-silo federated learning approach that uses locally generated synthetic data to facilitate collaboration between institutions. FedKR combines advanced data generation techniques with a dynamic aggregation process to provide greater security against privacy attacks than existing methods, significantly reducing the attack surface. Experimental results on generic and medical datasets show that FedKR achieves competitive performance, with an average improvement in accuracy of 4.24% compared to training models from local data, demonstrating particular effectiveness in data scarcity scenarios. • FedKR is a synthetic data-based federated learning technique for enhanced privacy. • Synthetic data is exchanged and aggregated locally instead of models or metadata. • Robust against membership inference, model inversion and gradient leakage attacks. • Mitigates various privacy attacks whilst maintaining competitive performance. • Particularly effective in scenarios with limited data availability.

In the last decades, machine learning techniques have been extensively used in place of classical template attacks to implement profiled side-channel analysis. This manuscript focuses on the application of machine learning to counteract Dynamic Frequency Scaling defenses. While state-of-the-art attacks have shown promising results against desynchronization countermeasures, a robust attack strategy has yet to be realized. Motivated by the simplicity and effectiveness of template attacks for devices lacking desynchronization countermeasures, this work presents a Deep Learning-assisted Template Attack (DLaTA) methodology specifically designed to target highly desynchronized traces through Dynamic Frequency Scaling. A deep learning-based pre-processing step recovers information obscured by desynchronization, followed by a template attack for key extraction. Specifically, we developed a three-stage deep learning pipeline to resynchronize traces to a uniform reference clock frequency. The experimental results on the AES cryptosystem executed on a RISC-V System-on-Chip reported a Guessing Entropy equal to 1 and a Guessing Distance greater than 0.25. Results demonstrate the method's ability to successfully retrieve secret keys even in the presence of high desynchronization. As an additional contribution, we publicly release ourDFS_DESYNCHdatabase11https://github.com/hardware-fab/DLaTAcontaining the first set of real-world highly desynchronized power traces from the execution of a software AES cryptosystem.

Ageing is associated with cognitive and functional decline, which can hamper daily activities and independent living. Chronic diseases may intensify this process. Early detection of unhealthy decline is key but hindered by similarity to normal ageing. This study presents an approach for early screening of healthy ageing, using an instrumented ink pen to ecologically assess handwriting performance in three age groups: 40-59, 60-69 and 70+ years old. Raw handwriting data from 60 healthy subjects were used to extract fourteen indicators related to gesture and tremor. The indicators were then used to discriminate between subjects of different age groups in three binary classification tasks, using machine learning algorithms. This approach produced remarkable results, particularly in identifying subjects at the very beginning of the ageing process (Group 2) from elderly subjects (Group 3), achieving an accuracy of 97.5%, an F1 score of 97.44% and a ROC-AUC of 95%. Analysis of the Shapley values revealed age-dependent sensitivity of handwriting and tremor-related indicators. The proposed method represents a promising solution for early detection of abnormal signs of ageing, designed for remote, non-invasive, unsupervised home monitoring to improve the care of older adults.

The increasing adoption of digital communication technologies has led to the accumulation of vast amounts of unstructured data, providing new opportunities to analyze organizational dynamics. In this study, we present a comprehensive framework for capturing, processing, and analyzing large-scale digital communication data to uncover hidden informal communication patterns within a real-world corporate environment. Utilizing a unique dataset of 40.425.247 emails within a banking company, spanning a two-year period, we construct a social network highlighting the email interactions. The graph is compared with organizational data, including business unit affiliations, enabling a detailed analysis of the relationship between formal organizational structures and informal social networks. By applying state-of-the-art community detection algorithms, we extract informal communities and compare them to the formal structure using various external validation metrics. Our results reveal that while informal communities may overlap with the formal structure to some extent, they often diverge in ways that may reveal organizational dynamics, communication flow, and inter-departmental collaboration that are not apparent from the formal structure alone. This study aims to pave the way for future research by exploring how human resource management can leverage insights derived from digital communication interactions to foster positive organizational behavior and enhance management practices.

The AI-SPRINT project, launched in 2021 and funded by the European Commission, focuses on the development and implementation of AI applications across the computing continuum. This continuum ensures the coherent integration of computational resources and services from centralized data centers to edge devices, facilitating efficient and adaptive computation and application delivery. AI-SPRINT has achieved significant scientific advances, including streamlined processes, improved efficiency, and the ability to operate in real time, as evidenced by three practical use cases. This paper provides an in-depth examination of these applications – Personalized Healthcare, Maintenance and Inspection, and Farming 4.0 – highlighting their practical implementation and the objectives achieved with the integration of AI-SPRINT technologies. We analyze how the proposed toolchain effectively addresses a range of challenges and refines processes, discussing its relevance and impact in multiple domains. After a comprehensive overview of the main AI-SPRINT tools used in these scenarios, the paper summarizes of the findings and key lessons learned.

Side-channel attacks allow extracting secret infor-mation from the execution of cryptographic primitives by cor-relating the partially known computed data and the measured side-channel signal. However, to set up a successful side-channel attack, the attacker has to perform i) the challenging task of locating the time instant in which the target cryptographic primitive is executed inside a side-channel trace and then ii) the time-alignment of the measured data on that time instant. This paper presents a novel deep-learning technique to locate the time instant in which the target computed cryptographic operations are executed in the side-channel trace. In contrast to state-of-the-art solutions, the proposed methodology works even in the presence of trace deformations obtained through random delay insertion techniques. We validated our proposal through a successful attack against a variety of unprotected and protected cryptographic primitives that have been executed on an FPGA-implemented system-on-chip featuring a RISC- V CPU.

Recent advances in generative artificial intelligence have enabled the creation of high-quality synthetic data that closely mimics real-world data.This paper explores the adaptation of the Stable Diffusion 2.0 model for generating synthetic datasets, using Transfer Learning, Fine-Tuning and generation parameter optimisation techniques to improve the utility of the dataset for downstream classification tasks.We present a class-conditional version of the model that exploits a Class-Encoder and optimisation of key generation parameters.Our methodology led to synthetic datasets that, in a third of cases, produced models that outperformed those trained on real datasets. * This paper is supported by the FAIR (Future Artificial Intelligence Research) project, funded by the NextGenerationEU program within the PNRR-PE-AI scheme (M4C2, investment 1.3, line on Artificial Intelligence).1 The authors have contributed in equal measure.

Accurate classification of medical images is essential for modern diagnostics.Deep learning advancements led clinicians to increasingly use sophisticated models to make faster and more accurate decisions, sometimes replacing human judgment.However, model development is costly and repetitive.Neural Architecture Search (NAS) provides solutions by automating the design of deep learning architectures.This paper presents ZO-DARTS+, a differentiable NAS algorithm that improves search efficiency through a novel method of generating sparse probabilities by bilevel optimization.Experiments on five public medical datasets show that ZO-DARTS+ matches the accuracy of state-of-the-art solutions while reducing search times by up to three times.

This paper presents an innovative approach to gaze tracking in the context of smart eyewear, utilizing a fully event-based algorithm. Traditional gaze-tracking methods often rely on grayscale or infrared imaging, which can be computationally intensive and raise privacy concerns. Our research addresses these issues by developing an algorithm that exclusively uses data from event-based sensors, optimizing for the limited computational capabilities of smart eyewear. The system uses simple geometrical operations, enabling efficient real-time processing. Experimental results demonstrate the feasibility of this approach, offering a promising solution for gaze tracking in compact, computationally constrained devices. Despite certain limitations in accuracy due to optimization for efficiency, the research underscores the practicality of this approach for practical, privacy-conscious applications in smart eyewear technology.

Semantic segmentation is a key prerequisite to robust image understanding for applications in Artificial Intelligence and Robotics. Few Shot Segmentation, in particular, concerns the extension and optimization of traditional segmentation methods in challenging conditions where limited training examples are available. A predominant approach in Few Shot Segmentation is to rely on a single backbone for visual feature extraction. Choosing which backbone to leverage is a deciding factor contributing to the overall performance. In this work, we interrogate on whether fusing features from different backbones can improve the ability of Few Shot Segmentation models to capture richer visual features. To tackle this question, we propose and compare two ensembling techniques—Independent Voting and Feature Fusion. Among the available Few Shot Segmentation methods, we implement the proposed ensembling techniques on PANet. The module dedicated to predicting segmentation masks from the backbone embeddings in PANet avoids trainable parameters, creating a controlled ‘in vitro’ setting for isolating the impact of different ensembling strategies. Leveraging the complementary strengths of different backbones, our approach outperforms the original single-backbone PANet across standard benchmarks even in challenging one-shot learning scenarios. Specifically, it achieved a performance improvement of +7.37% on PASCAL-5iand of +10.68% on COCO-20iin the top-performing scenario where three backbones are combined. These results, together with the qualitative inspection of the predicted subject masks, suggest that relying on multiple backbones in PANet leads to a more comprehensive feature representation, thus expediting the successful application of Few Shot Segmentation methods in challenging, data-scarce environments.

In this work, we leverage self-supervised pre-trained Vision Transformers (ViTs) and their self-attention maps to build a novel anomaly detection algorithm. The method, SAMSAD (Self-Attention MapS for Anomaly Detection), leverages the ability of pre-training procedures to obtain implicit images’ semantic segmentation masks. From this organization of features, a three-step procedure is proposed based on fine-tuning, clustering, and anomaly scoring. The algorithm achieves state-of-the-art performances of anomaly detection and segmentation in industrial product images, scaling to higher-resolution images with more effectiveness than its competitors.

Recent advancements in deep generative models, particularly with the application of CLIP (Contrastive Language–Image Pre-training) to Denoising Diffusion Probabilistic Models (DDPMs), have demonstrated remarkable effectiveness in text-to-image generation. The well-structured embedding space of CLIP has also been extended to image-to-shape generation with DDPMs, yielding notable results. Despite these successes, some fundamental questions arise: Does CLIP ensure the best results in shape generation from images? Can we leverage conditioning to bring explicit 3D knowledge into the generative process and obtain better quality? This study introduces CISP (Contrastive Image-Shape Pre-training), designed to enhance 3D shape synthesis guided by 2D images. CISP aims to enrich the CLIP framework by aligning 2D images with 3D shapes in a shared embedding space, specifically capturing 3D characteristics potentially overlooked by CLIP’s text-image focus. Our comprehensive analysis assesses CISP’s guidance performance against CLIP-guided models, focusing on generation quality, diversity, and coherence of the produced shapes with the conditioning image. We find that, while matching CLIP in generation quality and diversity, CISP substantially improves coherence with input images, underscoring the value of incorporating 3D knowledge into generative models. These findings suggest a promising direction for advancing the synthesis of 3D visual content by integrating multimodal systems with 3D representations.

Loop Closure Detection (LCD) is an essential task in robotics and computer vision, serving as a fundamental component for various applications across diverse domains. These applications encompass object recognition, image retrieval, and video analysis. LCD consists in identifying whether a robot has returned to a previously visited location, referred to as a loop, and then estimating the related roto-translation with respect to the analyzed location. Despite the numerous advantages of radar sensors, such as their ability to operate under diverse weather conditions and provide a wider range of view compared to other commonly used sensors (e.g., cameras or LiDARs), integrating radar data remains an arduous task due to intrinsic noise and distortion. To address this challenge, this research introduces RadarLCD, a novel supervised deep learning pipeline specifically designed for Loop Closure Detection using the FMCW Radar (Frequency Modulated Continuous Wave) sensor. RadarLCD, a learning-based LCD methodology explicitly designed for radar systems, makes a significant contribution by leveraging the pre-trained HERO (Hybrid Estimation Radar Odometry) model. Being originally developed for radar odometry, HERO’s features are used to select key points crucial for LCD tasks. The methodology undergoes evaluation across a variety of FMCW Radar dataset scenes, and it is compared to state-of-the-art systems such as Scan Context for Place Recognition and ICP for Loop Closure. The results demonstrate that RadarLCD surpasses the alternatives in multiple aspects of Loop Closure Detection.

Accurate 3D road surfaces are important for the development of detailed and realistic scenarios to validate autonomous driving algorithms. In these scenarios, simulations can be conducted, for instance, to evaluate the response of a safety system under dangerous conditions. In this paper, we propose an approach designed to automatically generate 3D road surfaces from data collected by a vehicle equipped with various sensors, including a LiDAR. These road surfaces are meant to be both accurate and realistic for driving simulations. The proposed approach, after deriving the clothoidal representation of the surface borders, pursues the idea of extracting and interpolating a set of smooth 3D cross-section profiles. The resulting surface provides a 3D representation in analytical form, allowing detailed rendering at the desired resolution. We experimentally evaluate the proposed approach in a real-world scenario to assess its performance in terms of accuracy, scalability, and computing time.

Radar odometry estimation has emerged as a critical technique in the field of autonomous navigation, providing robust and reliable motion estimation under various environmental conditions. Despite its potential, the complex nature of radar signals and the inherent challenges associated with processing these signals have limited the widespread adoption of this technology. This paper aims to address these challenges and simultaneously present an understanding about the current advancements in radar odometry estimation. First, we propose novel improvements to an existing state-of-the-art method, which are designed to enhance accuracy and reliability in diverse scenarios. Our pipeline consists of filtering, motion compensation, oriented surface points computation, smoothing, one-to-many radar scan registration, and pose refinement. In particular, we enforce local understanding of a scene by including additional information through smoothing (Gaussian kernels) and alignment (ICP), introduced by us in the existing pipeline. Then, we present an in-depth investigation of the contribution of each improvement to the localization accuracy. Lastly, we benchmark our system and state-of-the-art methods on all sequences of well-known datasets for radar understanding, i.e., the Oxford Radar RobotCar, MulRan, and Boreas datasets. In particular, Boreas includes scenarios with challenging weather conditions, such as snow or overcast, and, to our knowledge, it has never been used for evaluation or benchmarking in the literature. The effectiveness of the proposed improvements is proven by an increased translation and rotation accuracy on the majority of scenarios considered.

This paper presents a novel approach to generating behavior trees for robots using lightweight large language models (LLMs) with a maximum of 7 billion parameters. The study demonstrates that it is possible to achieve satisfying results with compact LLMs when fine-tuned on a specific dataset. The key contributions of this research include the creation of a fine-tuning dataset based on existing behavior trees using GPT-3.5 and a comprehensive comparison of multiple LLMs (namely llama2, llama-chat, and code-llama) across nine distinct tasks. To be thorough, we evaluated the generated behavior trees using static syntactical analysis, a validation system, a simulated environment, and a real robot. Furthermore, this work opens the possibility of deploying such solutions directly on the robot, enhancing its practical applicability. Findings from this study demonstrate the potential of LLMs with a limited number of parameters in generating effective and efficient robot behaviors.

In the field of autonomous vehicles, the detection of road line markings is a crucial yet versatile component. It provides real-time guidance for navigation and low-level vehicle control, while it also enables the generation of lane-level HD maps. These maps require high precision to provide low-level details to all future map users. At the same time, control-oriented detection pipelines require increased inference frequency and high robustness to be deployed on a safety-critical system. With this work, we present OptimusLine, a versatile line detection pipeline tackling with ease both scenarios. Built around a frame-by-frame transformer-based neural model operating in image segmentation, we show that OptimusLine achieves state-of-the-art performance and analyze its computational impact. To provide robustness to perturbations when deployed on an actual vehicle, OptimusLine introduces a scheme exploiting temporal links between consecutive frames. Enforcing temporal consistency on each new line prediction, OptimusLine can generate more robust line descriptions and produce an estimate of its prediction uncertainty.

This work presents the design and validation of a distributed multi-sensor object tracking algorithm designed to integrate heterogeneous sensory data from multiple static acquisition stations. The primary challenge addressed is the accurate tracking of targets in complex urban environments, where occlusions and the dynamic nature of traffic frequently hinder detection and tracking efforts. This challenge is particularly relevant in multimodal exchange areas, where vehicular traffic merges with heavy pedestrian and bicycle flow. We also address the scenario of delayed detection, which can easily occur when data from multiple stations are combined or when intensive data processing is performed. Our algorithm ensures high coverage and accuracy by maintaining dual Extended Kalman Filter states for each object, thus allowing for the assimilation of delayed detections and preserving optimal filter estimates at all times. The results of the proposed pipeline, tested using a digital twin of the Milano Bovisa Campus, demonstrate its efficacy, achieving high tracking precision across various scenarios and sensor combinations. Moreover, the results highlight the advantages of a distributed multi-sensor acquisition system compared to a single central station.

Digital Twins (DTs) for physical wireless environments have been recently proposed as accurate virtual representations of the propagation environment that can enable multi-layer decisions at the physical communication equipment. At high-frequency bands, DTs can help to overcome the challenges emerging in high mobility conditions featuring vehicular environments. In this paper, we propose a novel data-driven workflow for the creation of the DT of a Vehicle-to-Everything (V2X) communication scenario and a multi-modal simulation framework for the generation of realistic sensor data and accurate mmWave/sub-THz wireless channels. The proposed method leverages an automotive simulation and testing framework and an accurate ray-tracing channel simulator. Simulations over an urban scenario show the achievable realistic sensor and channel modelling both at the infrastructure and at ego-vehicles. We showcase the proposed framework on the DT-aided blockage handover task for V2X link restoration, leveraging the framework’s dynamic channel generation capabilities for realistic vehicular blockage simulation.

Survival analysis is a subfield of statistics concerned with modeling the occurrence time of a particular event of interest for a population. Survival analysis found widespread applications in healthcare, engineering, and social sciences. However, real-world applications involve survival datasets that are distributed, incomplete, censored, and confidential. In this context, federated learning can tremendously improve the performance of survival analysis applications. Federated learning provides a set of privacy-preserving techniques to jointly train machine learning models on multiple datasets without compromising user privacy, leading to a better generalization performance. However, despite the widespread development of federated learning in recent AI research, few studies focus on federated survival analysis. In this work, we present a novel federated algorithm for survival analysis based on one of the most successful survival models, the random survival forest. We call the proposed method Federated Survival Forest (FedSurF). With a single communication round, FedSurF obtains a discriminative power comparable to deep-learning-based federated models trained over hundreds of federated iterations. Moreover, FedSurF retains all the advantages of random forests, namely low computational cost and natural handling of missing values and incomplete datasets. These advantages are especially desirable in real-world federated environments with multiple small datasets stored on devices with low computational capabilities. Numerical experiments compare FedSurF with state-of-the-art survival models in federated networks, showing how FedSurF outperforms deep-learning-based federated algorithms in realistic environments with non-identically distributed data.

Multiple applications require a detailed representation of the world, especially in urban scenarios, including localization, mapping, and autonomous driving. Various solutions are available to achieve the 3D reconstruction of entire urban maps, starting from point clouds, in the form of surface meshes. Nevertheless, such systems are not able to obtain precise reconstructions, which only show coarse-grained detail. To tackle this issue, while exploiting existing deep learning methods, we propose a complete pipeline for object-level 3D reconstruction, with the goal of increasing also the expressiveness of data by replacing objects' point clouds with surface meshes. While focusing only on vehicles, the method is easily extendable to other elements of the scene. We also propose a systemic approach to studying existing deep learning works on single tasks to be used in the developed pipeline. The proposed system consists of multiple steps, including: point cloud registration, semantic segmentation, clustering, object detection, point cloud completion, point cloud rendering, and 3D reconstruction. We evaluate our pipeline on sequences of the SemanticKITTI dataset, including also quantitative and qualitative analyses, which demonstrate the validity of the achieved results.

Synthetic Aperture Radar (SAR) is a cutting-edge remote sensing technology that offers a unique perspective on the Earth's surface through its advanced microwave imaging, providing valuable insights into various aspects of the environment. However, SAR images are impacted by the speckle phenomenon, which acts as noise, hindering accurate interpretation of the scene and presenting a major challenge for SAR image analysis and understanding. Deep Learning has emerged as a powerful solution for despeckling SAR images, but acquiring large amounts of labeled data for training is a significant obstacle as obtaining ground truth in the SAR domain is not feasible. The proposed method overcomes this limitation with a novel unsupervised approach to single-look SAR image despeckling. Our method, CycleSAR, leverages the power of cycle-consistent generative adversarial networks (CycleGANs) to formulate the despeckling problem as an unpaired image-to-image translation, by effectively bypassing the need for ground truth data. Additionally, our method not only effectively reduces the speckle in single-look SAR images, but also enables, by construction, the simultaneous learning of a generative model to generate realistic speckled realizations of multi-look SAR images. The addition of a conditional variational autoencoder (CVAE) further enhances the method, enabling the one-to-many generation of speckled images and leading to an overall improvement in despeckling performance. The experimental results demonstrate the remarkable capabilities of the proposed method, CycleSAR, in providing high-quality despeckling and realistic speckle realizations. CycleSAR stands apart from existing state-of-the-art methods as it does not rely on data simulation or make any assumptions about the speckle distribution.

The use of Neural Architecture Search (NAS) techniques to automate the design of neural networks has become increasingly popular in recent years. The proliferation of devices with different hardware characteristics using such neural networks, as well as the need to reduce the power consumption for their search, has led to the realisation of Once-For-All (OFA), an algorithm engineered for low CO2 consumption and characterised by the ability to generate easily adaptable models through a single learning process. In order to improve this paradigm and develop high-performance yet eco-friendly NAS techniques, this paper presents OFAv2, the extension of OFA aimed at improving its performance while maintaining the same ecological advantage. The algorithm is improved from an architectural point of view by including early exits, parallel blocks and dense skip connections. The training process is extended by two new steps called Elastic Level and Elastic Exit. A new Knowledge Distillation technique is presented to handle multi-output networks, and finally a new strategy for dynamic teacher network selection is proposed. These modifications allow OFAv2 to improve its accuracy performance on the Tiny ImageNet dataset by up to 12.07% compared to the original version of OFA, while maintaining the algorithm flexibility and advantages.

Autonomous driving and 3D mapping are a few applications associated with real-time six-degrees-of-freedom pose estimation of ground vehicles, especially in outdoor (e.g., urban) environments. During the past decades, many systems have been proposed, with the majority working on data coming from only one sensor, while also struggling to keep accuracy and performance balanced. In this paper, we present D3VIL-SLAM, which extends an existing LiDAR-based SLAM system, ART-SLAM, to include inertial and visual information. The front-end comprises three branches that perform short-term data association, i.e., tracking, by exploiting laser, visual, and inertial data, respectively. All motion estimates and loop constraints derived from both LiDAR scans and images are used to build a robust g2o pose graph, which is later optimized to best satisfy all motion constraints. We compare the accuracy of our system with state-of-the-art SLAM methods, showing that D3VIL-SLAM is more accurate and produces highly detailed 3D maps while retaining real-time performance. Lastly, we perform a brief ablation study with different limitations (e.g., only images are allowed). All experimental campaigns are done by evaluating the estimated trajectory displacement using the KITTI dataset.

The Rey Osterrieth complex figure (ROCF) is one of the most used neuropsychological tests for the assessment of mild cognitive impairment (MCI) and dementia. In the copy test, the patient has to draw a replica of a 18-pattern image and the outcome is a score based on the accuracy of the overall drawing. The standard scoring system however have limitations related to its subjective nature and its inability to evaluate other cognitive domains than constructional abilities. Previous works addressed those problems by proposing tablet-based automated evaluation systems. Even promising, such methods are still far away from clinical validation and translation. In this work, we developed a decision support system (DSS) for the evaluation of the ROCF copy test in the common practice using retrospective information from previously performed drawings. The goal of our system was to support the professionals providing a qualitative judgement for each of the 18 patterns, estimating the most probable diagnosis for the patient, and identifying the main signs associated to the obtained diagnosis. A total of 250 human evaluated ROCF copies were scanned from 57 healthy subjects, 131 individuals with MCI, and 62 individuals with dementia. The images were pre-processed and analysed using both computer vision and deep learning techniques to assign a qualitative label to the 18 patterns. Then, the 18 labels were used as features in 3 binary (healthy VS MCI, healthy VS dementia, MCI VS dementia) and a 3-class classifications with model explanation (SHAP). Very good to excellent performance were obtained in all the diagnosis classification tasks. Indeed, an accuracy of about 85%, 91%, and 83% was obtained in discriminating healthy subjects from MCI, healthy subjects from dementia and MCI from dementia respectively. An accuracy of 73% was achieved in the 3-class classification. The model explanation showed which patterns are responsible for each prediction and how the importance of some patterns changes according to the severity of the cognitive decline. The proposed DSS enriches the standard evaluation and interpretation of the ROCF copy test. Being trained with retrospective knowledge, the performance of the DSS can be further enhanced by extending the dataset with existing ROCF copies.

Event cameras are novel bio-inspired sensors, which asynchronously capture pixel-level intensity changes in the form of “events”. Due to their sensing mechanism, event cameras have little to no motion blur, a very high temporal resolution and require significantly less power and memory than traditional frame-based cameras. These characteristics make them a perfect fit to several real-world applications such as egocentric action recognition on wearable devices, where fast camera motion and limited power challenge traditional vision sensors. However, the ever-growing field of event-based vision has, to date, overlooked the potential of event cameras in such applications. In this paper, we show that event data is a very valuable modality for egocentric action recognition. To do so, we introduce N-EPIC-Kitchens, the first event-based camera extension of the large-scale EPIC-Kitchens dataset. In this context, we propose two strategies: (i) directly processing event-camera data with traditional video-processing architectures (E2(GO)) and (ii) using event-data to distill optical flow information (E2(GO)MO). On our proposed benchmark, we show that event data provides a comparable performance to RGB and optical flow, yet without any additional flow computation at deploy time, and an improved performance of up to 4% with respect to RGB only information. The N-EPIC-Kitchens dataset is available at https://github.com/EgocentricVision/N-EPIC-Kitchens.

In this work, we investigate a methodology to perform anomaly detection and localization in images. The method leverages both sparse representation learning and the adoption of a pre-trained neural network for classification purposes. The objective is to assess the effectiveness of the K-SVD sparse dictionary learning algorithm and understand the role of neural network activation maps as data descriptors. We extract meaningful representation features and build a sparse dictionary of the most expressive ones. The dictionary is built only over features coming from images without anomalies. Thus, images containing anomalies will either have a non-sparse representation as linear combinations of the dictionary elements or a high reconstruction error. We show that the proposed pipeline achieves state-of-the-art performance in terms of AUC-ROC score over benchmarks such as MVTec Anomaly Detection, Rd-MVTec Anomaly Detection, Magnetic Tiles Defect, BeanTech Anomaly Detection, Kolektor Surface Defect datasets.

Automating the research for the best neural network model is a task that has gained more and more relevance in the last few years. In this context, Neural Architecture Search (NAS) represents the most effective technique whose results rival the state of the art hand-crafted architectures. However, this approach requires a lot of computational capabilities as well as research time, which make prohibitive its usage in many real-world scenarios. With its sequential model-based optimization strategy, Progressive Neural Architecture Search (PNAS) represents a possible step forward to face this resources issue. Despite the quality of the found network architectures, this technique is still limited in research time. A significant step in this direction has been done by Pareto-Optimal Progressive Neural Architecture Search (POPNAS), which expand PNAS with a time predictor to enable a trade-off between search time and accuracy, considering a multi-objective optimization problem. This paper proposes a new version of the Pareto-Optimal Progressive Neural Architecture Search, called POPNASv2. Our approach enhances its first version and improves its performance. We expanded the search space by adding new operators and improved the quality of both predictors to build more accurate Pareto fronts. Moreover, we introduced cell equivalence checks and enriched the search strategy with an adaptive greedy exploration step. Our efforts allow POPNASv2 to achieve PNAS-like performance with an average 4x factor search time speed-up. Code: https://doi.org/10.5281/zenodo.6574040

Deep Learning systems, able to achieve significant breakthroughs in many fields, including computer vision and speech recognition, are not inherently secure. Adversarial attacks on computer vision models can craft slightly perturbed inputs that exploit the models' multi-dimensional boundary shape to dramatically reduce their performance without compromising the perception that human beings have of such input. In this work, we present Trust-No-Pixel, a novel plug-and-play strategy to harden neural network image classifiers from adversarial attacks, based on a massive inpainting strategy. The inpainting technique of our defense performs a total erase of the input image and its reconstruction from scratch. Our experiments show Trust-No-Pixel improved accuracy against the more challenging type of such attacks, namely the white box adversarial attacks. Moreover, an exhaustive comparison of our technique against state-of-the-art approaches taken from academic literature confirmed the solid defense performances of Trust-No-Pixel under a wide variety of scenarios, including different attacks and attacked network architectures.

Cameras are among the most used sensors in Advanced Driver Assistance Systems (ADAS) and autonomous vehicles for their low cost and rich stream of information. Nevertheless, they require accurate extrinsic calibration to refer external features, e.g., obstacles and road line markings, to the vehicle reference frame. In this paper, we present a real-time online calibration procedure designed to adjust the camera’s pitch and height estimates by enforcing road line markings parallelism. Differently from most of the approaches in the literature, our is not limited to straight line markings as, under the assumption of local width constancy, parallelism is enforced also in case of high curvature line markings. Furthermore, to take into account the vehicle dynamics, e.g., accelerations and braking, our estimation procedure is framed in the context of an inverted pendulum dynamical system for which a robust filter is proposed. Finally, we experimentally assess the performance of the overall approach both in simulated and real scenarios.

Simultaneous localization and mapping (SLAM) is one fundamental topic in robotics due to its applications in autonomous driving. Over the last decades, many systems have been proposed, working on data coming from different sensors, such as cameras or LiDARs. Although excellent results were reached, the majority of these methods exploit the data as is, without extracting additional information or considering multiple sensors simultaneously. In this paper, we present MCS-SLAM, a Graph SLAM system that performs sensor fusion by exploiting multi-cues extracted from sensor data: color/intensity, depth/range and normal information. For each sensor, motion estimation is achieved through minimization of the pixel-wise difference between two multi-cue images. All estimates are then collectively optimized to achieve a coherent transformation. Point clouds received as input are also used to perform loop detection and closure. We compare the performance of the proposed system with state-of-the-art point cloud-based methods, LeGO-LOAM-BOR, LIO-SAM, HDL and ART-SLAM, and show that the proposed algorithm achieves less accuracy than the state-of-the-art, while needing much less computational time. The comparison is made by evaluating the estimated trajectory displacement, using the KITTI dataset.

Lane-level HD maps are crucial for trajectory planning and control in current autonomous vehicles. For this reason, appropriate line models should be adopted to define them. Whereas mapping algorithms often rely on inaccurate representations, clothoid curves possess peculiar smoothness properties that make them desirable representations of road lines in control algorithms. We propose a multi-stage pipeline for the generation of lane-level HD maps from monocular vision relying on clothoidal spline models. We obtain measurements of the line positions using a line detection algorithm, and we exploit a graph-based optimization framework to reach an optimal fitting. An iterative greedy procedure reduces the model complexity removing unnecessary clothoids. We validate our system on a real-world dataset, which we make publicly available for further research at https://airlab.deib.polimi.it/datasets-and-tools/.

MIMO systems in the context of 6G Vehicle-to-Everything (V2X) will require an accurate channel knowledge to enable efficient communication. Standard channel estimation techniques, such as Unconstrained Maximum Likelihood (UML), are extremely noisy in massive MIMO settings, while structured approaches, e.g., compressed sensing, are sensitive to hardware impairments. We propose a novel multi-vehicular algebraic channel estimation method for 6G V2X based on unsupervised learning which exploits recurrent vehicle passages in typical urban settings. Multiple training sequences from different vehicle passages are clustered via K-medoids algorithm based on their algebraic similarity to retrieve the MIMO channel eigenmodes, which can be used to improve the channel estimates. Numerical results show the presence of an optimal number of clusters and remarkable benefits of the proposed method in terms of Mean Squared Error (MSE) compared to standard U-ML solution (15 dB less).

BACKGROUND: Histone Mark Modifications (HMs) are crucial actors in gene regulation, as they actively remodel chromatin to modulate transcriptional activity: aberrant combinatorial patterns of HMs have been connected with several diseases, including cancer. HMs are, however, reversible modifications: understanding their role in disease would allow the design of 'epigenetic drugs' for specific, non-invasive treatments. Standard statistical techniques were not entirely successful in extracting representative features from raw HM signals over gene locations. On the other hand, deep learning approaches allow for effective automatic feature extraction, but at the expense of model interpretation. RESULTS: Here, we propose ShallowChrome, a novel computational pipeline to model transcriptional regulation via HMs in both an accurate and interpretable way. We attain state-of-the-art results on the binary classification of gene transcriptional states over 56 cell-types from the REMC database, largely outperforming recent deep learning approaches. We interpret our models by extracting insightful gene-specific regulative patterns, and we analyse them for the specific case of the PAX5 gene over three differentiated blood cell lines. Finally, we compare the patterns we obtained with the characteristic emission patterns of ChromHMM, and show that ShallowChrome is able to coherently rank groups of chromatin states w.r.t. their transcriptional activity. CONCLUSIONS: In this work we demonstrate that it is possible to model HM-modulated gene expression regulation in a highly accurate, yet interpretable way. Our feature extraction algorithm leverages on data downstream the identification of enriched regions to retrieve gene-wise, statistically significant and dynamically located features for each HM. These features are highly predictive of gene transcriptional state, and allow for accurate modeling by computationally efficient logistic regression models. These models allow a direct inspection and a rigorous interpretation, helping to formulate quantifiable hypotheses.

Interferometric SAR (InSAR) algorithms exploit synthetic aperture radar (SAR) images to estimate ground displacements, which are updated at each new satellite acquisition, over wide areas. The analysis of the resulting time series finds its application, among others, in monitoring tasks regarding seismic faults, subsidence, landslides, and urban structures, for which an accurate and timely response is required. Typical analyses consist of identifying among the numerous time series the ones that exhibit an anomalous displacement, thus deserving to be further investigated. In practice, this is realized by selecting the time series that is characterized by trend changes w.r.t. the historical behavior. In this work, we propose a deep learning approach for change point detection in the InSAR time series. The designed architecture combines long short-term memory (LSTM) cells, to model the temporal correlation among samples in the input time series, and time-gated LSTM (TGLSTM) cells, to consider the sampling rate as additional information during learning. We further propose a solution to the lack of ground truth by developing a suitable pipeline for realistic data simulation. The method has been developed and validated through a large suite of experiments. Both quantitative and qualitative analyses have been conducted to demonstrate the detection capabilities of the learned model and how it is a valid alternative to the statistical reference algorithm. We further applied the developed method in a real continuous monitoring project to analyze the InSAR time series over the Tuscany region in Italy, proving its effectiveness in the real domain.

The adoption of Artificial intelligence (AI) technologies is steadily increasing. However, to become fully pervasive, AI needs resources at the edge of the network. The cloud can provide the processing power needed for big data, but edge computing is close to where data are produced and therefore crucial to their timely, flexible, and secure management. In this paper, we introduce the AI-SPRINT project, which will provide solutions to seamlessly design, partition, and run AI applications in computing continuum environments. AI-SPRINT will offer novel tools for AI applications development, secure execution, easy deployment, as well as runtime management and optimization: AI-SPRINT design tools will allow trading-off application performance (in terms of end-to-end latency or throughput), energy efficiency, and AI models accuracy while providing security and privacy guarantees. The runtime environment will support live data protection, architecture enhancement, agile delivery, runtime optimization, and continuous adaptation.

In modern manufacturing, divergent market dynamics impel companies to move toward a zero-defect production by reducing the risk of errors and defects down to zero. Paint-coating of metal surfaces is one of such process steps and most prominent as consumers will be animated to buy based on their first impression. Despite significant advances in automation and precision engineering of paint-coating, the presence of process contaminants as residual of different stages of production may compromise the process. In this contribution, we focus on the paint-coating of washing machine cabinets as a representative. Within the last decade, hyperspectral imaging technology has shown promising potentials in a variety of applications that aim at detecting objects and discriminating materials. In this work, we present a hyperspectral acquisition and analysis system that verifies the feasibility of detection and discrimination of process contaminants smeared on the washing machine cabinet based on spectral information. The acquisition system, aided by a robot arm, collects hyperspectral images based on two scenarios: contaminants on flat steel sheets and contaminants on washing machine chassis. This dataset, which is published publicly, is calibrated, analysed, and segmented through the proposed analysis models. The results for both flat base and structured washing machine surfaces indicate the great capacity of this technology for being integrated into the pre-treatment stage before painting metal parts.

BACKGROUND: In-depth analysis of regulation networks of genes aberrantly expressed in cancer is essential for better understanding tumors and identifying key genes that could be therapeutically targeted. RESULTS: We developed a quantitative analysis approach to investigate the main biological relationships among different regulatory elements and target genes; we applied it to Ovarian Serous Cystadenocarcinoma and 177 target genes belonging to three main pathways (DNA REPAIR, STEM CELLS and GLUCOSE METABOLISM) relevant for this tumor. Combining data from ENCODE and TCGA datasets, we built a predictive linear model for the regulation of each target gene, assessing the relationships between its expression, promoter methylation, expression of genes in the same or in the other pathways and of putative transcription factors. We proved the reliability and significance of our approach in a similar tumor type (basal-like Breast cancer) and using a different existing algorithm (ARACNe), and we obtained experimental confirmations on potentially interesting results. CONCLUSIONS: The analysis of the proposed models allowed disclosing the relations between a gene and its related biological processes, the interconnections between the different gene sets, and the evaluation of the relevant regulatory elements at single gene level. This led to the identification of already known regulators and/or gene correlations and to unveil a set of still unknown and potentially interesting biological relationships for their pharmacological and clinical use.

The availability of advanced hybrid system identification techniques is fundamental to extract knowledge in form of models from data streams. Starting from the current state of the art, we propose an approach based on a specialized architecture, conceived to address the peculiar integration of nonlinear dynamics and finite state switching behavior of hybrid systems. Following the Mixtures of Experts concept, we combine a set of Neural Network ARX (NNARX) models with a Gated Recurrent Units network with softmax output. The former are exploited to map specific nonlinear dynamical models representing the behavior of the system in each discrete mode of operation. The latter, operating as a neural switching machine, infers the unobserved active mode and learns the state-transition logic, conditioned on input-output data sequences. Besides, we integrate a LASSO based input features and model selection mechanism, aimed to extract the most informative lags over the sequences for each NNARX and calibrate the modes to be employed. The overall system is trained end-to-end. Experiments have been performed on a benchmark hybrid automata with nonlinear dynamics and transitions, showing the capability to achieve improved performances than conventional architectures.

Probabilistic electricity price forecast (EPF) systems represent a fundamental tool to achieve robust production scheduling and day-ahead bidding strategies. However, most EPF methods, including recently proposed deep learning based techniques, are still targeting point predictions, following the common Gaussian assumption. In this work, we propose a novel probabilistic EPF approach based on the integration of a Gaussian Mixture layer, parametrized by a Recurrent Neural Network with Gated Recurrent Units, including an L1-norm based feature selection mechanisms. The network is conceived to approximate general conditional price distributions through learning. Moreover, we developed a multi-hours prediction approach exploiting correlations and patters both in hourly and cross-hour contexts. Experiments have been performed on the Italian market dataset, showing the capability of the proposed method to achieve accurate out-of-sample predictions while providing explicit uncertainty indications supporting enhanced decision making.

Neural networks are being broadly explored for the identification of Industrial Cyber Physical Systems (ICPS) models from data sequences. However, learned representations typically lack explainability, representing a major challenge of deep learning. Interpreting the information structured across the synaptic links is particularly challenging for recurrent neural networks (RNN), encoding input features and observed system dynamics within a continuous latent space. In this work, we target the representations built within RNNs while learning behavioral models of a class of discrete dynamical systems. To this end, we propose a method to extract the symbolic knowledge structured by the continuous state, based on Gaussian Mixture Model clustering, handling latent activations characterized by partially overlapping and non-isotropic distributions. Experiments are performed on a pilot remanufacturing plant, by learning the model of a conveyor controller from process data. We show the capability of the proposed method to extract the hidden finite state machine from the trained RNN, providing a human interpretable representation of the input conditioned computations performed through the continuous latent space.

Image inpainting aims at synthesizing the missing content of damaged or corrupted images to produce visually realistic restorations; typical applications are in image restoration, automatic scene editing, super-resolution, and dynamic object removal. In this paper, we propose Semantic and Edge Conditioned Inpainting Generative Adversarial Network (SECI-GAN), an architecture that jointly exploits the high-level cues extracted by semantic segmentation and the fine-grained details captured by edge extraction to condition the image inpainting process. SECI-GAN is designed with a particular focus on recovering big regions belonging to the same object (e.g. cars or pedestrians) in the context of dynamic object removal from complex street views. To demonstrate the effectiveness of SECI-GAN, we evaluate our results on the Cityscapes dataset, showing that SECI-GAN is better than competing state-of-the-art models at recovering the structure and the content of the missing parts while producing consistent predictions.

Mesh refinement is a fundamental step for accurate Multi- View Stereo. It modifies the geometry of an initial manifold mesh to minimize the photometric error induced in a set of camera pairs. This initial mesh is usually the output of volumetric 3D reconstruction based on min-cut over Delaunay Triangulations. Such methods produce a significant amount of non-manifold vertices, therefore they require a vertex split step to explicitly repair them. In this paper, we extend this method to preemptively fix the non-manifold vertices by reasoning directly on the Delaunay Triangulation and avoid most vertex splits. The main contribution of this paper addresses the problem of choosing the camera pairs adopted by the refinement process. We treat the problem as a mesh labeling process, where each label corresponds to a camera pair. Differently from the state-of-the-art methods, which use each camera pair to refine all the visible parts of the mesh, we choose, for each facet, the best pair that enforces both the overall visibility and coverage. The refinement step is applied for each facet using only the camera pair selected. This facetwise refinement helps the process to be applied in the most evenly way possible.

The ability of autonomous vehicles to maintain an accurate trajectory within their road lane is crucial for safe operation. This requires detecting the road lines and estimating the car relative pose within its lane. Lateral lines are usually retrieved from camera images. Still, most of the works on line detection are limited to image mask retrieval and do not provide a usable representation in world coordinates. What we propose in this paper is a complete perception pipeline based on monocular vision and able to retrieve all the information required by a vehicle lateral control system: road lines equation, centerline, vehicle heading and lateral displacement. We evaluate our system by acquiring data with accurate geometric ground truth. To act as a benchmark for further research, we make this new dataset publicly available at http://airlab.deib.polimi.it/datasets/.

One of the most successful approaches in Multi-View Stereo estimates a depth map and a normal map for each view via PatchMatch-based optimization and fuses them into a consistent 3D points cloud. This approach relies on photo-consistency to evaluate the goodness of a depth estimate. It generally produces very accurate results; however, the reconstructed model often lacks completeness, especially in correspondence of broad untextured areas where the photo-consistency metrics are unreliable. Assuming the untextured areas piecewise planar, in this paper we generate novel PatchMatch hypotheses so to expand reliable depth estimates in neighboring untextured regions. At the same time, we modify the photo-consistency measure such to favor standard or novel PatchMatch depth hypotheses depending on the textureness of the considered area. We also propose a depth refinement step to filter wrong estimates and to fill the gaps on both the depth maps and normal maps while preserving the discontinuities. The effectiveness of our new methods has been tested against several state of the art algorithms in the publicly available ETH3D dataset containing a wide variety of high and low-resolution images.

Dense 3D visual mapping estimates as many as possible pixel depths, for each image. This results in very dense point clouds that often contain redundant and noisy information, especially for surfaces that are roughly planar, for instance, the ground or the walls in the scene. In this paper we leverage on semantic image segmentation to discriminate which regions of the scene require simplification and which should be kept at high level of details. We propose four different point cloud simplification methods which decimate the perceived point cloud by relying on class-specific local and global statistics still maintaining more points in the proximity of class boundaries to preserve the infra-class edges and discontinuities. 3D dense model is obtained by fusing the point clouds in a 3D Delaunay Triangulation to deal with variable point cloud density. In the experimental evaluation we have shown that, by leveraging on semantics, it is possible to simplify the model and diminish the noise affecting the point clouds.

The availability of accurate day-ahead price forecasts is crucial to achieve an effective participation to electricity markets. Starting from available state of the art, we propose a forecast technique exploiting a nonlinear auto regressive model with exogenous input, including a feature selection mechanism based on the Least Absolute Shrinkage and Selection Operator (LASSO). The rationale behind such a choice is twofold. On the one hand, we aim to target potential increase of forecast accuracy by learning complex non-linear mappings. On the other hand, we want to increase the interpretability of the resulting model and minimize the effort needed to properly set up the forecaster. A framework such as the LASSO, capable to self-extract features from spot price multi-variate time series, might represent a very useful tool for industrial practitioners. Experiments have been performed on Italian market dataset, demonstrating that the proposed method can extract useful features and achieve robust performance. Moreover, we show how the proposed method can support interpretation of forecaster structure and it can reveal interesting correlations within the regression set.

Modern deep neural networks are being widely exploited to solve challenging learning tasks, including nonlinear system identification. Bayesian system identification intrinsically encapsulate uncertainty in model parameters and provides forecasting distribution enabling enhanced analysis, simulation and control system design. Nevertheless, the application of the full Bayesian approach to articulated models as deep neural networks results quite challenging in practice. In this work we propose an identification technique for nonlinear dynamic systems exploiting a deep recurrent neural network with Long-Short Term Memory (LSTM) units retaining a Bayesian framework. To such an aim, we stacked the recurrent neural network with a probabilistic layer, decomposing the nonlinear dynamic model into a combination of flexible functions. Hence, deterministic and stochastic layers are trained jointly, forcing the learning algorithm to transform the input data sequences into a deterministic feature space encoded by the LSTM, useful for predictions. Besides, we deployed a scalable technique based on Variational Inference to deal with the exact inference intractability. We show the effectiveness of the proposed approach by the application to a widely exploited open benchmark for nonlinear system identification.

The results of scientific experiments performed by different groups are rarely directly comparable. Efforts such as the European Robotics League offer to the community, in the form of competitions, well documented and stable benchmarks to assess the performance of existing systems. However, benchmarks can be equally useful at design time: the Plug&Bench Benchmark Meta-model provides robot designers with a valuable addition to their toolkit. Additionally, it enables -with benchmark composition-to predict system performance given the benchmark results of individual components.

Event-based cameras are neuromorphic sensors capable of efficiently encoding visual information in the form of sparse sequences of events. Being biologically inspired, they are commonly used to exploit some of the computational and power consumption benefits of biological vision. In this paper we focus on a specific feature of vision: visual attention. We propose two attentive models for event based vision: an algorithm that tracks events activity within the field of view to locate regions of interest and a fully-differentiable attention procedure based on DRAW neural model. We highlight the strengths and weaknesses of the proposed methods on four datasets, the Shifted N-MNIST, Shifted MNIST-DVS, CIFAR10-DVS and N-Caltech101 collections, using the Phased LSTM recognition network as a baseline reference model obtaining improvements in terms of both translation and scale invariance.

One-dimensional intervals incremental inverted index (Di4) is a multi-resolution, single-dimension indexing framework for efficient, scalable, and extensible computation of genomic interval expressions. The framework has a tri-layer architecture: the semantic layer provides orthogonal and generic means (including the support of user-defined function) of sense-making and higher-lever reasoning from region-based datasets; the logical layer provides building blocks for region calculus and topological relations between intervals; the physical layer abstracts from persistence technology and makes the model adaptable to variety of persistence technologies, spanning from small-scale (e.g., B+tree) to large-scale (e.g., LevelDB). The extensibility of Di4 to application scenarios is shown with an example of comparative evaluation of ChIP-seq and DNase-Seq replicates. Performance of Di4 is benchmarked for small and large scale scenarios under common bioinformatics application scenarios. Di4 is freely available from https://genometric.github.io/Di4.

Mesh labeling is the key problem of classifying the facets of a 3D mesh with a label among a set of possible ones. State-of-the-art methods model mesh labeling as a Markov Random Field over the facets. These algorithms map image segmentations to the mesh by minimizing an energy function that comprises a data term, a smoothness terms, and class-specific priors. The latter favor a labeling with respect to another depending on the orientation of the facet normals. In this paper we propose a novel energy term that acts as a prior, but does not require any prior knowledge about the scene nor scene-specific relationship among classes. It bootstraps from a coarse mapping of the 2D segmentations on the mesh, and it favors the facets to be labeled according to the statistics of the mesh normals in their neighborhood. We tested our approach against five different datasets and, even if we do not inject prior knowledge, our method adapts to the data and overcomes the state-of-the-art.

In the wide context of biological processes regulating gene expression, transcriptional regulation driven by epigenetic activity is among the most effective and intriguing ones. Understanding the complex language of histone modifications and transcription factor bindings is an appealing yet hard task, given the large number of involved features and the specificity of their combinatorial behavior across genes. Genome-wide regression models for predicting mRNA abundance quantifications from epigenetic activity are interesting in an exploratory framework, but their effectiveness is limited as the relative predictive power of epigenetic features is hard to discern at such level of resolution. On the other hand, an investigative analysis cannot rely on prior biological knowledge to perform sensible grouping of genes and locally study epigenetic regulative processes. In this context, we shaped the “gene stratification problem” as a form of epigenetic feature-based hyperplanes clustering, and proposed a genetic algorithm to approach this task, aiming at performing datadriven partitioning of the whole set of protein coding genes of an organism based on the characteristic relation between their expression and the associated epigenetic activity. We observed how, not only the hyperplanes described by the resulting partitions significantly differ from each other, but also how different epigenetic features are of diverse importance in predicting gene expression within each partition. This demonstrates the validity and biological interest of the proposed computational method and the obtained results.

This paper presents a novel method for the reconstruction of 3D edges in multi-view stereo scenarios. Previous research in the field typically relied on video sequences and limited the reconstruction process to either straight linesegments, or edge-points, i.e., 3D points that correspond to image edges. We instead propose a system, denoted as EdgeGraph3D, able to recover both straight and curved 3D edges from an unordered image sequence. A second contribution of this work is a graph-based representation for 2D edges that allows the identification of the most structurally significant edges detected in an image. We integrate Edge-Graph3D in a multi-view stereo reconstruction pipeline and analyze the benefits provided by 3D edges to the accuracy of the recovered surfaces. We evaluate the effectiveness of our approach on multiple datasets from two different collections in the multi-view stereo literature. Experimental results demonstrate the ability of EdgeGraph3D to work in presence of strong illumination changes and reflections, which are usually detrimental to the effectiveness of classical photometric reconstruction systems.

Bioinformatics (2015) https://doi.org/10.1093/bioinformatics/btv293 The authors of the above paper wish to inform readers that the source code for the project has been migrated from CodePlex to Github. The correct link to access the source code is: https://github.com/Genometric/MSPC. The paper has now been corrected online.

In the era of autonomous driving, urban mapping represents a core step to let vehicles interact with the urban context. Successful mapping algorithms have been proposed in the last decade building the map leveraging on data from a single sensor. The focus of the system presented in this paper is twofold: the joint estimation of a 3D map from lidar data and images, based on a 3D mesh, and its texturing. Indeed, even if most surveying vehicles for mapping are endowed by cameras and lidar, existing mapping algorithms usually rely on either images or lidar data; moreover both image-based and lidar-based systems often represent the map as a point cloud, while a continuous textured mesh representation would be useful for visualization and navigation purposes. In the proposed framework, we join the accuracy of the 3D lidar data, and the dense information and appearance carried by the images, in estimating a visibility consistent map upon the lidar measurements, and refining it photometrically through the acquired images. We evaluate the proposed framework against the KITTI dataset and we show the performance improvement with respect to two state of the art urban mapping algorithms, and two widely used surface reconstruction algorithms in Computer Graphics.

Enriched region (ER) identification is a fundamental step in several next-generation sequencing (NGS) experiment types. Yet, although NGS experimental protocols recommend producing replicate samples for each evaluated condition and their consistency is usually assessed, typically pipelines for ER identification do not consider available NGS replicates. This may alter genome-wide descriptions of ERs, hinder significance of subsequent analyses on detected ERs and eventually preclude biological discoveries that evidence in replicate could support. MuSERA is a broadly useful stand-alone tool for both interactive and batch analysis of combined evidence from ERs in multiple ChIP-seq or DNase-seq replicates. Besides rigorously combining sample replicates to increase statistical significance of detected ERs, it also provides quantitative evaluations and graphical features to assess the biological relevance of each determined ER set within its genomic context; they include genomic annotation of determined ERs, nearest ER distance distribution, global correlation assessment of ERs and an integrated genome browser. We review MuSERA rationale and implementation, and illustrate how sets of significant ERs are expanded by applying MuSERA on replicates for several types of NGS data, including ChIP-seq of transcription factors or histone marks and DNase-seq hypersensitive sites. We show that MuSERA can determine a new, enhanced set of ERs for each sample by locally combining evidence on replicates, and prove how the easy-to-use interactive graphical displays and quantitative evaluations that MuSERA provides effectively support thorough inspection of obtained results and evaluation of their biological content, facilitating their understanding and biological interpretations. MuSERA is freely available at http://www.bioinformatics.deib.polimi.it/MuSERA/.

BACKGROUND: With the wide-spreading of public repositories of NGS processed data, the availability of user-friendly and effective tools for data exploration, analysis and visualization is becoming very relevant. These tools enable interactive analytics, an exploratory approach for the seamless "sense-making" of data through on-the-fly integration of analysis and visualization phases, suggested not only for evaluating processing results, but also for designing and adapting NGS data analysis pipelines. RESULTS: This paper presents abstractions for supporting the early analysis of NGS processed data and their implementation in an associated tool, named GenoMetric Space Explorer (GeMSE). This tool serves the needs of the GenoMetric Query Language, an innovative cloud-based system for computing complex queries over heterogeneous processed data. It can also be used starting from any text files in standard BED, BroadPeak, NarrowPeak, GTF, or general tab-delimited format, containing numerical features of genomic regions; metadata can be provided as text files in tab-delimited attribute-value format. GeMSE allows interactive analytics, consisting of on-the-fly cycling among steps of data exploration, analysis and visualization that help biologists and bioinformaticians in making sense of heterogeneous genomic datasets. By means of an explorative interaction support, users can trace past activities and quickly recover their results, seamlessly going backward and forward in the analysis steps and comparative visualizations of heatmaps. CONCLUSIONS: GeMSE effective application and practical usefulness is demonstrated through significant use cases of biological interest. GeMSE is available at http://www.bioinformatics.deib.polimi.it/GeMSE/ , and its source code is available at https://github.com/Genometric/GeMSE under GPLv3 open-source license.

Detecting moving objects in dynamic scenes from sequences of lidar scans is an important task in object tracking, mapping, localization, and navigation. Many works focus on changes detection in previously observed scenes, while a very limited amount of literature addresses moving objects detection. The state-of-the-art method exploits Dempster-Shafer Theory to evaluate the occupancy of a lidar scan and to discriminate points belonging to the static scene from moving ones. In this paper we improve both speed and accuracy of this method by discretizing the occupancy representation, and by removing false positives through visual cues. Many false positives lying on the ground plane are also removed thanks to a novel ground plane removal algorithm. Efficiency is improved through an octree indexing strategy. Experimental evaluation against the KITTI public dataset shows the effectiveness of our approach, both qualitatively and quantitatively with respect to the state-of-the-art.

In this paper we propose a new approach to incrementally initialize a manifold surface for automatic 3D reconstruction from images. More precisely we focus on the automatic initialization of a 3D mesh as close as possible to the final solution; indeed many approaches require a good initial solution for further refinement via multi-view stereo techniques. Our novel algorithm automatically estimates an initial manifold mesh for surface evolving multi-view stereo algorithms, where the manifold property needs to be enforced. It bootstraps from 3D points extracted via Structure from Motion, then iterates between a state-of-the-art manifold reconstruction step and a novel mesh sweeping algorithm that looks for new 3D points in the neighborhood of the reconstructed manifold to be added in the manifold reconstruction. The experimental results show quantitatively that the mesh sweeping improves the resolution and the accuracy of the manifold reconstruction, allowing a better convergence of state-of-the-art surface evolution multi-view stereo algorithms.

Urban reconstruction from a video captured by a surveying vehicle constitutes a core module of automated mapping. When computational power represents a limited resource and, a detailed map is not the primary goal, the reconstruction can be performed incrementally, from a monocular video, carving a 3D Delaunay triangulation of sparse points; this allows online incremental mapping for tasks such as traversability analysis or obstacle avoidance. To exploit the sharp edges of urban landscape, we propose to use a Delaunay triangulation of Edge-Points, which are the 3D points corresponding to image edges. These points constrain the edges of the 3D Delaunay triangulation to real-world edges. Besides the use of the Edge-Points, a second contribution of this paper is the Inverse Cone Heuristic that preemptively avoids the creation of artifacts in the reconstructed manifold surface. We force the reconstruction of a manifold surface since it makes it possible to apply computer graphics or photometric refinement algorithms to the output mesh. We evaluated our approach on four real sequences of the public available KITTI dataset by comparing the incremental reconstruction against Velodyne measurements.

Despite the recent attitude toward making modern urban cities accessible, many barriers and obstacles still challenge impaired people mobility every day. Having a complete and accurate map of both accessible paths and barriers could support people mobility and could drive urban planning renewing efforts. However, mapping accessible paths and barriers timely, and in an accurate way, requires a significant effort and it is often left to volunteers and no-profit associations. The Maps for Easy Paths (MEP) project represents an attempt to build accessibility maps (i.e., maps with accessible routes) fusing, in an automatic fashion, data implicitly collected through mobile devices carried by interested people when they move around the city. In particular, the paper focuses on the design of the MEP-Traces app used for the collection of relevant data and the MEP-Fusion engine for the reconstruction of accessible routes out of inertial readings and GPS measurements. Experimental results in reconstructing accessible routes in urban scenarios show the effectiveness of the approach in dealing with typical problems related to poor inertial sensors quality.

MOTIVATION: Chromatin Immunoprecipitation followed by sequencing (ChIP-seq) detects genome-wide DNA-protein interactions and chromatin modifications, returning enriched regions (ERs), usually associated with a significance score. Moderately significant interactions can correspond to true, weak interactions, or to false positives; replicates of a ChIP-seq experiment can provide co-localised evidence to decide between the two cases. We designed a general methodological framework to rigorously combine the evidence of ERs in ChIP-seq replicates, with the option to set a significance threshold on the repeated evidence and a minimum number of samples bearing this evidence. RESULTS: We applied our method to Myc transcription factor ChIP-seq datasets in K562 cells available in the ENCODE project. Using replicates, we could extend up to 3 times the ER number with respect to single-sample analysis with equivalent significance threshold. We validated the 'rescued' ERs by checking for the overlap with open chromatin regions and for the enrichment of the motif that Myc binds with strongest affinity; we compared our results with alternative methods (IDR and jMOSAiCS), obtaining more validated peaks than the former and less peaks than latter, but with a better validation. AVAILABILITY AND IMPLEMENTATION: An implementation of the proposed method and its source code under GPLv3 license are freely available at http://www.bioinformatics.deib.polimi.it/MSPC/ and http://mspc.codeplex.com/, respectively. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary Material are available at Bioinformatics online.

The design and development of the pose tracking system for an autonomous mobile robot and the time consuming calibration of its intrinsic sensor parameters (e.g., displacement, misalignment and iron distortions of an inertial measurement unit) are one of the preliminary requirements of any project involving a mobile robot platform. This paper introduces ROAMFREE, a turn-on-and-go multiple sensors pose tracking and self-calibration framework adaptable to different mobile robot platforms (e.g., Ackerman steering vehicles, quadrotor aerial vehicles, omnidirectional mobile robots). We formulate the sensor fusion problem as a Gauss-Newton optimization on an hyper-graph where nodes represent poses and calibration parameters while edges represent nonlinear measurement constraints. This formulation allows us to solve both online pose tracking and offline sensor self-calibration problems.

Thanks to the large diffusion of small wearable devices there are several systems designed for indoor localization. Among the pro- posed solutions, RF-based systems have been deeply investigated due to their flexibility and limited costs. When these systems are employed as assistive tools, they s

The geometric framework based on Stochastic Relaxation allows to describe from a common perspective different model-based optimization algorithms that make use of statistical models to guide the search for the optimum. In this paper Stochastic Relaxation is used to provide theoretical results on Estimation of Distribution Algorithms (EDAs). By the use of Stochastic Relaxation we show how the estimation of the fitness model by least squares linear regression corresponds to the estimation of the natural gradient. This equivalence allows to simultaneously perform model selection and robust estimation of the natural gradient. Finally, we interpet Linear Programming relaxation as an example of Stochastic Relaxation, with respect to the regular gradient.

The design and the development of the position and orientation tracking system of a mobile robot, and the calibration of its sensor parameters (e.g., displacement, misalignment and iron distortions), are challenging and time consuming tasks in every autonomous robotics project. The ROAMFREE framework delivers turn-on-and-go multi-sensors pose tracking and self-calibration modules and it is designed to be flexible and to adapt to every kind of mobile robotic platform. In ROAMFREE, the sensor data fusion problem is formulated as a hyper-graph optimization where nodes represent poses and calibration parameters and edges the non-linear measurement constraints. This formulation allows us to solve both the on-line pose tracking and the off-line sensor self-calibration problems. In this paper, we introduce the approach and we discuss a real platform case study, along with experimental results.

Persons suffering from motor disorders have limited possibilities for communicating and normally require assistive technologies to fulfill this primary need. Promising means of providing basic communication abilities to subjects affected by severe motor impairments include brain-computer interfaces (BCIs), that is, systems that directly translate brain signals into device commands, bypassing any muscle or nerve mediation. To date, the use of BCIs for effective verbal communication is yet an open issue, primarily due to the low rates of information transfer that can be achieved with this technology. Still, performance of BCI spelling applications could be considerably improved by a smart user interface design and by the adoption of natural language processing (NLP) techniques for text prediction. The objective of this work is to suggest an approach and a user interface for BCI spelling applications combining state-of-the-art BCI and NLP techniques to maximize the overall communication rate of the system. The BCI paradigm adopted is motor imagery, that is, when the subject imagines moving a certain part of the body, he/she produces modifications to specific brain rhythms that are detected in real-time through an electroencephalogram and translated into commands for a spelling application. By maximizing the overall communication rate, our approach is twofold: on one hand, we maximize the information transfer rate from the control signal, on the other hand, we optimize the way this information is employed for the purpose of verbal communication. The achieved results are satisfactory and comparable with the latest works reported in literature on motor-imagery BCI spellers. For the three subjects tested, we obtained a spelling rate of respectively 3 char/min, 2.7 char/min, and 2 char/min.

Stochastic relaxation aims at finding the minimum of a fitness function by identifying a proper sequence of distributions, in a given model, that minimize the expected value of the fitness function. Different algorithms fit this framework, and they differ according to the policy they implement to identify the next distribution in the model. In this paper we present two algorithms, in the stochastic relaxation framework, for the optimization of real-valued functions defined over binary variables: Stochastic Gradient Descent (SGD) and Stochastic Natural Gradient Descent (SNDG). These algorithms use a stochastic model to sample from as it happens for Estimation of Distribution Algorithms (EDAs), but the estimation of the model from the population is substituted by the direct update of model parameter through stochastic gradient descent. The two algorithms, SGD and SNDG, both use statistical models in the exponential family, but they differ in the use of the natural gradient, first proposed in the literature by Amari, in the context of Information Geometry. Due to the properties of the exponential family, both gradient and natural gradient can be evaluated in terms of covariances between the fitness function and the sufficient statistics of the exponential family. As the computation of the exact gradient is unfeasible, we approximate the gradient by evaluating empirical covariances. We test the performance of our algorithm over different standard benchmarks, and we compare the results with other well-known meta-heuristics in the framework of EDAs.

Estimation of Distribution Algorithms evolve populations of candidate solutions to an optimization problem by introducing a statistical model, and by replacing classical variation operators of Genetic Algorithms with statistical operators, such as estimation and sampling. The choice of the model plays a key role in the evolutionary process, indeed it strongly affects the convergence to the global optimum. From this point of view, in a black-box context, especially when the interactions among variables in the objective function are sparse, it becomes fundamental for an EDA to choose the right model, able to encode such correlations. In this paper we focus on EDAs based on undirected graphical models, such as Markov Networks. To learn the topology of the graph we apply a sparse method based on ℓ1-regularized logistic regression, which has been demonstrated to be efficient in the high-dimensional case, i.e., when the number of observations is much smaller than the sample space. We propose a new algorithm within the DEUM framework, called DEUM ℓ1, able to learn the interactions structure of the problem without the need of prior knowledge, and we compare its performance with other popular EDAs, over a set of well known benchmarks.

This paper presents Evolutionary Optimization Tool (Evoptool), an optimization toolkit that implements a set of meta-heuristics based on the Evolutionary Computation paradigm. Evoptool provides a common platform for the development and test of new algorithms, in order to facilitate the performance comparison activity. The toolkit offers a wide set of benchmark problems, from classical toy examples to complex tasks, and a collection of implementations of algorithms from the Genetic Algorithms and Estimation of Distribution Algorithms paradigms. Evoptool is flexible and easy to extend, also with algorithms based on other approaches that go beyond Evolutionary Computation.

We describe an unsupervised host-based intrusion detection system based on system call arguments and sequences. We define a set of anomaly detection models for the individual parameters of the call. We then describe a clustering process that helps to better fit models to system call arguments and creates interrelations among different arguments of a system call. Finally, we add a behavioral Markov model in order to capture time correlations and abnormal behaviors. The whole system needs no prior knowledge input; it has a good signal-to-noise ratio, and it is also able to correctly contextualize alarms, giving the user more information to understand whether a true or false positive happened, and to detect global variations over the entire execution flow, as opposed to punctual ones over individual instances.

We describe a system for the evaluation of the sleep macrostructure on the basis of Emfit sensor foils placed into bed mattress and of advanced signal processing. The signals on which the analysis is based are heart-beat interval (HBI) and movement activity obtained from the bed sensor, the relevant features and parameters obtained through a time-variant autoregressive model (TVAM) used as feature extractor, and the classification obtained through a hidden Markov model (HMM). Parameters coming from the joint probability of the HBI features were used as input to a HMM, while movement features are used for wake period detection. A total of 18 recordings from healthy subjects, including also reference polysomnography, were used for the validation of the system. When compared to wake-nonrapid-eye-movement (NREM)-REM classification provided by experts, the described system achieved a total accuracy of 79+/-9% and a kappa index of 0.43+/-0.17 with only two HBI features and one movement parameter, and a total accuracy of 79+/-10% and a kappa index of 0.44+/-0.19 with three HBI features and one movement parameter. These results suggest that the combination of HBI and movement features could be a suitable alternative for sleep staging with the advantage of low cost and simplicity.

Recent works have shown that it is possible to solve the Simultaneous Localization And Mapping problem using an Extended Kalman Filter and a single perspective camera. The principal drawback of these works is an inaccurate modeling of measurement uncertainties, which therefore causes inconsistencies in the filter estimations. A possible solution to proper uncertainty modeling is the Unified Inverse Depth parametrization. In this paper we propose the Inverse Scaling parametrization that still allows an un-delayed initialization of features, while reducing the number of needed parameters and simplifying the measurement model. This novel approach allows a better uncertainty modeling of both low and high parallax features and reduces the likelihood of inconsistencies. Experiments in simulation demonstrate that the use of the Inverse Scaling solution improves the performance of the monocular EKF SLAM filter when compared with the Unified Inverse Depth approach; experiment on real data confirm the applicability of the idea.

In the last twenty years the understanding of the brain function and the advent of powerful low-cost computer equipment allowed the birth and the development of the BCI (Brain-Computer Interface), a device that interprets brain activity to issue commands. P300 is a positive peak at about 300 ms from a stimulus, and has been used as a base for a BCI in many studies. The aim of this research consists in recognizing and classifying P300 signals by using wavelet transforms. This study analyzes both the kind of wavelets and which coefficients are more suited for a 100% correct decisions using as few repetitions of stimuli as possible. The classifier performs a quadratic discriminant analysis. The method is tested on the “BCI Competition 2003” data set IIb with excellent results.

The recent literature has shown that it is possible to solve the monocular Simultaneous Localization And Mapping using both undelayed features ini- tialization and an Extedend Kalman Filter. The key concept, to achieve this result, was the introduction of a new parametrization called Unified Inverse Depth that produces measurements equations with a high degree of linear- ity and allows an efficient and accurate modeling of uncertainties. In this paper we present a monocular EKF SLAM filter based on an alternative parametrization, i.e., the Inverse Scaling Parametrization, characterized by a reduced number of parameters, a more linear measurement model, and a bet- ter modeling of features uncertainty for both low and high parallax features. Experiments in simulation demonstrate that the use of the Inverse Scaling solution improves the monocular EKF SLAM filter when compared with the Unified Inverse Depth approach, while experiments on real data show the system working in practice as well.

We present a method to diagnose lung cancer by the analysis of breath using an electronic nose. This device can react to a gas substance by providing signals that can be analyzed to classify the input. It is composed of a sensor array (6 MOS sensors, in our case) and a pattern classification process based on machine learning techniques. During the first phase of our research, we have evaluated the possibility and accuracy of lung cancer diagnosis by classifying the olfactory signal associated to exhalations of subjects. The second part of the research, still in progress, is aimed at assessing the possibility of discriminating also the different types and stages of the disease. At the end of the first phase, results have been very satisfactory and promising: we achieved an average accuracy of 92.6%, sensitivity of 95.3% and specificity of 90.5%. In particular we analyzed the breath of 101 individuals, of which 58 control subjects, and 43 suffer from different types of lung cancer (primary and not) at different stages. In order to find the components able to discriminate between the two classes ‘healthy’ and ‘sick’ at best, and to reduce the dimensionality of the problem, we have extracted the most significant features and projected them into a lower dimensional space using Non Parametric Linear Discriminant Analysis. Finally, we have used these features as input to several supervised pattern classification algorithms, based on different k-nearest neighbors (k-NN) approaches (classic, modified and Fuzzy k-NN), linear and quadratic discriminant classifiers and on a feed-forward artificial neural network (ANN). The observed results have all been validated using cross-validation. These results pushed us to begin the second phase of the project to investigate the possibility of early lung cancer diagnosis: we are involving a larger number of subjects, partioned in different classes according to the type and stage of the disease. The research demonstrates that the electronic nose is a promising alternative to current lung cancer diagnostic techniques: the obtained predictive errors are lower than those achieved by present diagnostic methods, and the cost of the analysis, both in money, time and resources, is lower. The introduction of this technology will lead to very important social and business effects: its low price and small dimensions allow a large scale distribution, giving the opportunity to perform non invasive, cheap, quick, and massive early diagnosis and screening.

A Brain-Computer Interface (BCI) is an interface that directly analyzes brain activity to transform user intentions into commands. Many known techniques use the P300 event-related potential by extracting relevant features from the EEG signal and feeding those features into a classifier. In these approaches, feature extraction becomes the key point, and doing it by hand can be at the same time cumbersome and suboptimal. In this paper we face the issue of feature extraction by using a genetic algorithm able to retrieve the relevant aspects of the signal to be classified in an automatic fashion. We have applied this algorithm to publicly available data sets (a BCI competition) and data collected in our lab, obtaining with a simple logistic classifier results comparable to the best algorithms in the literature. In addition, the features extracted by the algorithm can be interpreted in terms of signal characteristics that are contributing to the success of classification, giving new insights for brain activity investigation.

We present FIXCS (fuzzy implementation of XCS), a learning classifier system that extend the accuracy-based extended classifier system (XCS) by allowing to match real-valued input by fuzzy sets, and to produce a fuzzy output, then translated into real values. This work gives XCS the ability to face real-valued problems with a fuzzy model that approximates a real valued function better than the original, interval-based model. First results show that, as expected, the learning time is longer, but the obtained fuzzy system is more robust than the interval-based one.

Self localization and mapping with vision is still an open research field. Since redundancy in the sensing suite is too expensive for consumer-level robots, we base on vision as the main sensing system for SLAM. We approach the problem with 3D data from a trinocular vision system. Past experience shows that problems arise as a consequence of inaccurate modeling of uncertainties; interestingly enough, we found that accuracy in modeling the robot pose uncertainty is much less relevant than for the uncertainty on the sensed data. To overcome the severe limitation of linear and Gaussian approximations, we applied a particle-based description of the inherently non-normal probability density distribution of the sensed data; the aim is to increase the success rate of data association, which we see as the most important problem. The increase in correct data associations reduces the uncertainty in the model and, consequently, in the robot pose, respectively estimated with a hierarchical map decomposition and a six degree of freedom extended Kalman filter. In this paper, we present approaches for particle-based sensor modeling and data association, with a comparative experimental evaluation on real 3D vision data.

We present a method to recognize the presence of lung cancer in individuals by classifying the olfactory signal acquired through an electronic nose based on an array of MOS sensors. We analyzed the breath of 101 persons, of which 58 as control and 43 suffering from different types of lung cancer (primary and not) at different stages. In order to find the components able to discriminate between the two classes 'healthy' and 'sick' as best as possible and to reduce the dimensionality of the problem, we extracted the most significative features and projected them into a lower dimensional space, using Non Parametric Linear Discriminant Analysis. Finally, we used these features as input to several supervised pattern classification techniques, based on different k-nearest neighbors (k-NN) approaches (classic, modified and Fuzzy k-NN), linear and quadratic discriminant classifiers and on a feedforward artificial neural network (ANN). The observed results, all validated using cross-validation, have been satisfactory, achieving an accuracy of 92.6%, a sensitivity of 95.3% and a specificity of 90.5%. These results put the electronic nose as a valid implementation of lung cancer diagnostic technique, being able to obtain excellent results with a non invasive, small, low cost and very fast instrument.

In this paper we propose a Bayesian framework for XCS [9], called BXCS. Following [4], we use probability distributions to represent the uncertainty over the classifier estimates of payoff. A novel interpretation of classifier and an extension of the accuracy concept are presented. The probabilistic approach is aimed at increasing XCS learning capabilities and tendency to evolve accurate, maximally general classifiers, especially when uncertainty affects the environment or the reward function. We show that BXCS can approximate optimal solutions in stochastic environments with a high level of uncertainty.

This paper deals with transport user equilibrium. A modified version of the ant colony system is proposed where the ant colony heuristic is adapted in order to take into account all aspects characterizing the transport problem: multiple ODs (Origin-Destination pairs), link congestion, non-separable cost link functions, elasticity of demand, multi classes in demand.

This contribution focuses on the application of Artificial Neural Networks (ANNs) in landslide susceptibility zonation taking into consideration both the prediction capability assessment and the sensitivity to measurement errors in the obtained models. We suggest a general procedure to perform susceptibility analysis by means of ANNS, introducing robustness analysis as the final step in the susceptibility modelling in order to test the reliability of the obtained maps with respect to errors in measuring, and calculating, the conditioning factors. Such robustness analysis has been performed by calculating a robustness index both for each conditioning factors and for the total model; this allowed us to find the errors in the conditioning factors which affect the neural computation in a greater way and the overall robustness of the model. The experimental results, obtained on the Deba Valley database, suggest that ANNs are a proper method to analyze a complex relationship between conditioning factors and landslides, and that the robustness analysis is a crucial step in the susceptibility modeling, specially as an iterative procedure for variables selection.

In this paper, we present a dead reckoning sensor to support reliable odometry on mobile robots. This sensor is based on a pair of optical mice rigidly connected to the robot body and its main advantages are 1) this localization system is independent from the kinematics of the robot, 2) the measurement given by the mice is not subject to slipping, since they are independent from the traction wheels, nor to crawling, since they measure displacements in any direction 3) it is a low-cost solution with a precision comparable to classical shaft encoders. Since we have redundant measures it is possible to detect non-systematic errors; in this paper, an automatic procedure to reduce non-systematic errors of the sensor is presented and validated with experimental results on a real mobile robot.

In this paper we present a genetic approach that uses a Bayesian fitness function to the design of rich neural network topologies in order to find an optimal domain-specific non-linear function approximator with good generalization performance. Rich neural networks have a feed-forward topology with shortcut connections and arbitrary activation functions at each layer. This kind of topologies is particularly well suited for non-linear regression tasks, but it may suffer for overfilling issues. In this paper we present a Bayesian fitness function to effectively apply genetic algorithms with these models obtaining, in a completely automated way, models well-matched to the problem, with good generalization capability, and low complexity.

In this paper, we present a dead reckoning sensor to support reliable odometry on mobile robots. This sensor is based on a pair of optical mice rigidly connected to the robot body and its main advantages are 1) this localization system is independent from the kinematics of the robot, 2) the measurement given by the mice is not subject to slipping, since they are independent from the traction wheels, nor to crawling, since they measure displacements in any direction 3) it is a low-cost solution with a precision comparable to classical shaft encoders. We present the mathematical model of the sensor, its implementation, and some experimental evaluations using the standard UMBmark benchmark for odometry.

The success of any reinforcement learning (RL) application is in large part due to the design of an appropriate reinforcement function. A methodological framework to support the design of reinforcement functions has not been defined yet, and this critical and often underestimated activity is left to the ability of the RL application designer. We propose an approach to support reinforcement function design in RL applications concerning learning behaviors for autonomous agents. We define some dimensions along which we can describe reinforcement functions; we consider the distribution of reinforcement values, their coherence and their matching with the designer's perspective. We give hints to define measures that objectively describe the reinforcement function; we discuss the trade-offs that should be considered to improve learning and we introduce the dimensions along which this improvement can be expected. The approach we are presenting is general enough to be adopted in a large number of RL projects. We show how to apply it in the design of learning classifier systems (LCS) applications. We consider a simple, but quite complete case study in evolutionary robotics, and we discuss reinforcement function design issues in this sample context.