Continual learning for computer vision applications

One of the most visionary goals of Artificial Intelligence is to create a system able to mimic and eventually surpass the intelligence observed in biological systems including, ambitiously, the one observed in humans. The main distinctive strength of humans is their ability to build a deep understanding of the world by learning continuously and drawing from their experiences. This ability, which is found in various degrees in all intelligent biological beings, allows them to adapt and properly react to changes by incrementally expanding and refining their knowledge. Arguably, achieving this ability is one of the main goals of Artificial Intelligence and a cornerstone towards the creation of intelligent artificial agents. Modern Deep Learning approaches allowed researchers and industries to achieve great advancements towards the resolution of many long-standing problems in areas like Computer Vision and Natural Language Processing. However, while this current age of renewed interest in AI allowed for the creation of extremely useful applications, a concerningly limited effort is being directed towards the design of systems able to learn continuously. The biggest problem that hinders an AI system from learning incrementally is the catastrophic forgetting phenomenon. This phenomenon, which was discovered in the 90s, naturally occurs in Deep Learning architectures where classic learning paradigms are applied when learning incrementally from a stream of experiences. This dissertation revolves around the Continual Learning field, a sub-field of Machine Learning research that has recently made a comeback following the renewed interest in Deep Learning approaches. This work will focus on a comprehensive view of continual learning by considering algorithmic, benchmarking, and applicative aspects of this field. This dissertation will also touch on community aspects such as the design and creation of research tools aimed at supporting Continual Learning research, and the theoretical and practical aspects concerning public competitions in this field.

Scene Mapping and Understanding by Robotic Vision

The first mechanical Automaton concept was found in a Chinese text written in the 3rd century BC, while Computer Vision was born in the late 1960s. Therefore, visual perception applied to machines (i.e. the Machine Vision) is a young and exciting alliance. When robots came in, the new field of Robotic Vision was born, and these terms began to be erroneously interchanged. In short, we can say that Machine Vision is an engineering domain, which concern the industrial use of Vision. The Robotic Vision, instead, is a research field that tries to incorporate robotics aspects in computer vision algorithms. Visual Servoing, for example, is one of the problems that cannot be solved by computer vision only. Accordingly, a large part of this work deals with boosting popular Computer Vision techniques by exploiting robotics: e.g. the use of kinematics to localize a vision sensor, mounted as the robot end-effector. The remainder of this work is dedicated to the counterparty, i.e. the use of computer vision to solve real robotic problems like grasping objects or navigate avoiding obstacles. Will be presented a brief survey about mapping data structures most widely used in robotics along with SkiMap, a novel sparse data structure created both for robotic mapping and as a general purpose 3D spatial index. Thus, several approaches to implement Object Detection and Manipulation, by exploiting the aforementioned mapping strategies, will be proposed, along with a completely new Machine Teaching facility in order to simply the training procedure of modern Deep Learning networks.

Methodologies for visual correspondence

Visual correspondence is a key computer vision task that aims at identifying projections of the same 3D point into images taken either from different viewpoints or at different time instances. This task has been the subject of intense research activities in the last years in scenarios such as object recognition, motion detection, stereo vision, pattern matching, image registration. The approaches proposed in literature typically aim at improving the state of the art by increasing the reliability, the accuracy or the computational efficiency of visual correspondence algorithms. The research work carried out during the Ph.D. course and presented in this dissertation deals with three specific visual correspondence problems: fast pattern matching, stereo correspondence and robust image matching. The dissertation presents original contributions to the theory of visual correspondence, as well as applications dealing with 3D reconstruction and multi-view video surveillance.

Sistemi riconfigurabili a basso consumo per applicazioni di monitoraggio distribuito

The term Ambient Intelligence (AmI) refers to a vision on the future of the information society where smart, electronic environment are sensitive and responsive to the presence of people and their activities (Context awareness). In an ambient intelligence world, devices work in concert to support people in carrying out their everyday life activities, tasks and rituals in an easy, natural way using information and intelligence that is hidden in the network connecting these devices. This promotes the creation of pervasive environments improving the quality of life of the occupants and enhancing the human experience. AmI stems from the convergence of three key technologies: ubiquitous computing, ubiquitous communication and natural interfaces. Ambient intelligent systems are heterogeneous and require an excellent cooperation between several hardware/software technologies and disciplines, including signal processing, networking and protocols, embedded systems, information management, and distributed algorithms. Since a large amount of fixed and mobile sensors embedded is deployed into the environment, the Wireless Sensor Networks is one of the most relevant enabling technologies for AmI. WSN are complex systems made up of a number of sensor nodes which can be deployed in a target area to sense physical phenomena and communicate with other nodes and base stations. These simple devices typically embed a low power computational unit (microcontrollers, FPGAs etc.), a wireless communication unit, one or more sensors and a some form of energy supply (either batteries or energy scavenger modules). WNS promises of revolutionizing the interactions between the real physical worlds and human beings. Low-cost, low-computational power, low energy consumption and small size are characteristics that must be taken into consideration when designing and dealing with WSNs. To fully exploit the potential of distributed sensing approaches, a set of challengesmust be addressed. Sensor nodes are inherently resource-constrained systems with very low power consumption and small size requirements which enables than to reduce the interference on the physical phenomena sensed and to allow easy and low-cost deployment. They have limited processing speed,storage capacity and communication bandwidth that must be efficiently used to increase the degree of local ”understanding” of the observed phenomena. A particular case of sensor nodes are video sensors. This topic holds strong interest for a wide range of contexts such as military, security, robotics and most recently consumer applications. Vision sensors are extremely effective for medium to long-range sensing because vision provides rich information to human operators. However, image sensors generate a huge amount of data, whichmust be heavily processed before it is transmitted due to the scarce bandwidth capability of radio interfaces. In particular, in video-surveillance, it has been shown that source-side compression is mandatory due to limited bandwidth and delay constraints. Moreover, there is an ample opportunity for performing higher-level processing functions, such as object recognition that has the potential to drastically reduce the required bandwidth (e.g. by transmitting compressed images only when something ‘interesting‘ is detected). The energy cost of image processing must however be carefully minimized. Imaging could play and plays an important role in sensing devices for ambient intelligence. Computer vision can for instance be used for recognising persons and objects and recognising behaviour such as illness and rioting. Having a wireless camera as a camera mote opens the way for distributed scene analysis. More eyes see more than one and a camera system that can observe a scene from multiple directions would be able to overcome occlusion problems and could describe objects in their true 3D appearance. In real-time, these approaches are a recently opened field of research. In this thesis we pay attention to the realities of hardware/software technologies and the design needed to realize systems for distributed monitoring, attempting to propose solutions on open issues and filling the gap between AmI scenarios and hardware reality. The physical implementation of an individual wireless node is constrained by three important metrics which are outlined below. Despite that the design of the sensor network and its sensor nodes is strictly application dependent, a number of constraints should almost always be considered. Among them: • Small form factor to reduce nodes intrusiveness. • Low power consumption to reduce battery size and to extend nodes lifetime. • Low cost for a widespread diffusion. These limitations typically result in the adoption of low power, low cost devices such as low powermicrocontrollers with few kilobytes of RAMand tenth of kilobytes of program memory with whomonly simple data processing algorithms can be implemented. However the overall computational power of the WNS can be very large since the network presents a high degree of parallelism that can be exploited through the adoption of ad-hoc techniques. Furthermore through the fusion of information from the dense mesh of sensors even complex phenomena can be monitored. In this dissertation we present our results in building several AmI applications suitable for a WSN implementation. The work can be divided into two main areas:Low Power Video Sensor Node and Video Processing Alghoritm and Multimodal Surveillance . Low Power Video Sensor Nodes and Video Processing Alghoritms In comparison to scalar sensors, such as temperature, pressure, humidity, velocity, and acceleration sensors, vision sensors generate much higher bandwidth data due to the two-dimensional nature of their pixel array. We have tackled all the constraints listed above and have proposed solutions to overcome the current WSNlimits for Video sensor node. We have designed and developed wireless video sensor nodes focusing on the small size and the flexibility of reuse in different applications. The video nodes target a different design point: the portability (on-board power supply, wireless communication), a scanty power budget (500mW),while still providing a prominent level of intelligence, namely sophisticated classification algorithmand high level of reconfigurability. We developed two different video sensor node: The device architecture of the first one is based on a low-cost low-power FPGA+microcontroller system-on-chip. The second one is based on ARM9 processor. Both systems designed within the above mentioned power envelope could operate in a continuous fashion with Li-Polymer battery pack and solar panel. Novel low power low cost video sensor nodes which, in contrast to sensors that just watch the world, are capable of comprehending the perceived information in order to interpret it locally, are presented. Featuring such intelligence, these nodes would be able to cope with such tasks as recognition of unattended bags in airports, persons carrying potentially dangerous objects, etc.,which normally require a human operator. Vision algorithms for object detection, acquisition like human detection with Support Vector Machine (SVM) classification and abandoned/removed object detection are implemented, described and illustrated on real world data. Multimodal surveillance: In several setup the use of wired video cameras may not be possible. For this reason building an energy efficient wireless vision network for monitoring and surveillance is one of the major efforts in the sensor network community. Energy efficiency for wireless smart camera networks is one of the major efforts in distributed monitoring and surveillance community. For this reason, building an energy efficient wireless vision network for monitoring and surveillance is one of the major efforts in the sensor network community. The Pyroelectric Infra-Red (PIR) sensors have been used to extend the lifetime of a solar-powered video sensor node by providing an energy level dependent trigger to the video camera and the wireless module. Such approach has shown to be able to extend node lifetime and possibly result in continuous operation of the node.Being low-cost, passive (thus low-power) and presenting a limited form factor, PIR sensors are well suited for WSN applications. Moreover techniques to have aggressive power management policies are essential for achieving long-termoperating on standalone distributed cameras needed to improve the power consumption. We have used an adaptive controller like Model Predictive Control (MPC) to help the system to improve the performances outperforming naive power management policies.

Fish quality assessment through the application of chemico-physical, sensory and microbiological analyses

The quality of fish products is indispensably linked to the freshness of the raw material modulated by appropriate manipulation and storage conditions, specially the storage temperature after catch. The purpose of the research presented in this thesis, which was largely conducted in the context of a research project funded by Italian Ministry of Agricultural, Food and Forestry Policies (MIPAAF), concerned the evaluation of the freshness of farmed and wild fish species, in relation to different storage conditions, under ice (0°C) or at refrigeration temperature (4°C). Several specimens of different species, bogue (Boops boops), red mullet (Mullus barbatus), sea bream (Sparus aurata) and sea bass (Dicentrarchus labrax), during storage, under the different temperature conditions adopted, have been examined. The assessed control parameters were physical (texture, through the use of a dynamometer; visual quality using a computer vision system (CVS)), chemical (through footprint metabolomics 1H-NMR) and sensory (Quality Index Method (QIM). Microbiological determinations were also carried out on the species of hake (Merluccius merluccius). In general obtained results confirmed that the temperature of manipulation/conservation is a key factor in maintaining fish freshness. NMR spectroscopy showed to be able to quantify and evaluate the kinetics for unselected compounds during fish degradation, even a posteriori. This can be suitable for the development of new parameters related to quality and freshness. The development of physical methods, particularly the image analysis performed by computer vision system (CVS), for the evaluation of fish degradation, is very promising. Among CVS parameters, skin colour, presence and distribution of gill mucus, and eye shape modification evidenced a high sensibility for the estimation of fish quality loss, as a function of the adopted storage conditions. Particularly the eye concavity index detected on fish eye showed a high positive correlation with total QIM score.

Leopoli-Cencelle beyond virtual reality Documentazione, interpretazione e comprensione di una città medievale

La città medievale di Leopoli-Cencelle (fondata da Papa Leone IV nell‘854 d.C. non lontano da Civitavecchia) è stata oggetto di studio e di periodiche campagne di scavo a partire dal 1994. Le stratigrafie investigate con metodi tradizionali, hanno portato alla luce le numerose trasformazioni che la città ha subìto nel corso della sua esistenza in vita. Case, torri, botteghe e strati di vissuto, sono stati interpretati sin dall’inizio dello scavo basandosi sulla documentazione tradizionale e bi-dimensionale, legata al dato cartaceo e al disegno. Il presente lavoro intende re-interpretare i dati di scavo con l’ausilio delle tecnologie digitali. Per il progetto sono stati utilizzati un laser scanner, tecniche di Computer Vision e modellazione 3D. I tre metodi sono stati combinati in modo da poter visualizzare tridimensionalmente gli edifici abitativi scavati, con la possibilità di sovrapporre semplici modelli 3D che permettano di formulare ipotesi differenti sulla forma e sull’uso degli spazi. Modellare spazio e tempo offrendo varie possibilità di scelta, permette di combinare i dati reali tridimensionali, acquisiti con un laser scanner, con semplici modelli filologici in 3D e offre l’opportunità di valutare diverse possibili interpretazioni delle caratteristiche dell’edificio in base agli spazi, ai materiali, alle tecniche costruttive. Lo scopo del progetto è andare oltre la Realtà Virtuale, con la possibilità di analizzare i resti e di re-interpretare la funzione di un edificio, sia in fase di scavo che a scavo concluso. Dal punto di vista della ricerca, la possibilità di visualizzare le ipotesi sul campo favorisce una comprensione più profonda del contesto archeologico. Un secondo obiettivo è la comunicazione a un pubblico di “non-archeologi”. Si vuole offrire a normali visitatori la possibilità di comprendere e sperimentare il processo interpretativo, fornendo loro qualcosa in più rispetto a una sola ipotesi definitiva.

Concepts and methods for efficient decentralized learning in federated settings

Deep Neural Networks (DNNs) have revolutionized a wide range of applications beyond traditional machine learning and artificial intelligence fields, e.g., computer vision, healthcare, natural language processing and others. At the same time, edge devices have become central in our society, generating an unprecedented amount of data which could be used to train data-hungry models such as DNNs. However, the potentially sensitive or confidential nature of gathered data poses privacy concerns when storing and processing them in centralized locations. To this purpose, decentralized learning decouples model training from the need of directly accessing raw data, by alternating on-device training and periodic communications. The ability of distilling knowledge from decentralized data, however, comes at the cost of facing more challenging learning settings, such as coping with heterogeneous hardware and network connectivity, statistical diversity of data, and ensuring verifiable privacy guarantees. This Thesis proposes an extensive overview of decentralized learning literature, including a novel taxonomy and a detailed description of the most relevant system-level contributions in the related literature for privacy, communication efficiency, data and system heterogeneity, and poisoning defense. Next, this Thesis presents the design of an original solution to tackle communication efficiency and system heterogeneity, and empirically evaluates it on federated settings. For communication efficiency, an original method, specifically designed for Convolutional Neural Networks, is also described and evaluated against the state-of-the-art. Furthermore, this Thesis provides an in-depth review of recently proposed methods to tackle the performance degradation introduced by data heterogeneity, followed by empirical evaluations on challenging data distributions, highlighting strengths and possible weaknesses of the considered solutions. Finally, this Thesis presents a novel perspective on the usage of Knowledge Distillation as a mean for optimizing decentralized learning systems in settings characterized by data heterogeneity or system heterogeneity. Our vision on relevant future research directions close the manuscript.

Computational methods applied to undeciphered scripts from the Aegean and Cyprus: Linear A and Cypro-Minoan

The study of ancient, undeciphered scripts presents unique challenges, that depend both on the nature of the problem and on the peculiarities of each writing system. In this thesis, I present two computational approaches that are tailored to two different tasks and writing systems. The first of these methods is aimed at the decipherment of the Linear A afraction signs, in order to discover their numerical values. This is achieved with a combination of constraint programming, ad-hoc metrics and paleographic considerations. The second main contribution of this thesis regards the creation of an unsupervised deep learning model which uses drawings of signs from ancient writing system to learn to distinguish different graphemes in the vector space. This system, which is based on techniques used in the field of computer vision, is adapted to the study of ancient writing systems by incorporating information about sequences in the model, mirroring what is often done in natural language processing. In order to develop this model, the Cypriot Greek Syllabary is used as a target, since this is a deciphered writing system. Finally, this unsupervised model is adapted to the undeciphered Cypro-Minoan and it is used to answer open questions about this script. In particular, by reconstructing multiple allographs that are not agreed upon by paleographers, it supports the idea that Cypro-Minoan is a single script and not a collection of three script like it was proposed in the literature. These results on two different tasks shows that computational methods can be applied to undeciphered scripts, despite the relatively low amount of available data, paving the way for further advancement in paleography using these methods.

Mind the gaps: cognitive-inspired AI for high-level visual sensemaking. Towards abstract concept image classification

The abundance of visual data and the push for robust AI are driving the need for automated visual sensemaking. Computer Vision (CV) faces growing demand for models that can discern not only what images "represent," but also what they "evoke." This is a demand for tools mimicking human perception at a high semantic level, categorizing images based on concepts like freedom, danger, or safety. However, automating this process is challenging due to entropy, scarcity, subjectivity, and ethical considerations. These challenges not only impact performance but also underscore the critical need for interoperability. This dissertation focuses on abstract concept-based (AC) image classification, guided by three technical principles: situated grounding, performance enhancement, and interpretability. We introduce ART-stract, a novel dataset of cultural images annotated with ACs, serving as the foundation for a series of experiments across four key domains: assessing the effectiveness of the end-to-end DL paradigm, exploring cognitive-inspired semantic intermediaries, incorporating cultural and commonsense aspects, and neuro-symbolic integration of sensory-perceptual data with cognitive-based knowledge. Our results demonstrate that integrating CV approaches with semantic technologies yields methods that surpass the current state of the art in AC image classification, outperforming the end-to-end deep vision paradigm. The results emphasize the role semantic technologies can play in developing both effective and interpretable systems, through the capturing, situating, and reasoning over knowledge related to visual data. Furthermore, this dissertation explores the complex interplay between technical and socio-technical factors. By merging technical expertise with an understanding of human and societal aspects, we advocate for responsible labeling and training practices in visual media. These insights and techniques not only advance efforts in CV and explainable artificial intelligence but also propel us toward an era of AI development that harmonizes technical prowess with deep awareness of its human and societal implications.