A dynamical system approach to data assimilation in chaotic models

The Assimilation in the Unstable Subspace (AUS) was introduced by Trevisan and Uboldi in 2004, and developed by Trevisan, Uboldi and Carrassi, to minimize the analysis and forecast errors by exploiting the flow-dependent instabilities of the forecast-analysis cycle system, which may be thought of as a system forced by observations. In the AUS scheme the assimilation is obtained by confining the analysis increment in the unstable subspace of the forecast-analysis cycle system so that it will have the same structure of the dominant instabilities of the system. The unstable subspace is estimated by Breeding on the Data Assimilation System (BDAS). AUS- BDAS has already been tested in realistic models and observational configurations, including a Quasi-Geostrophicmodel and a high dimensional, primitive equation ocean model; the experiments include both fixed and“adaptive”observations. In these contexts, the AUS-BDAS approach greatly reduces the analysis error, with reasonable computational costs for data assimilation with respect, for example, to a prohibitive full Extended Kalman Filter. This is a follow-up study in which we revisit the AUS-BDAS approach in the more basic, highly nonlinear Lorenz 1963 convective model. We run observation system simulation experiments in a perfect model setting, and with two types of model error as well: random and systematic. In the different configurations examined, and in a perfect model setting, AUS once again shows better efficiency than other advanced data assimilation schemes. In the present study, we develop an iterative scheme that leads to a significant improvement of the overall assimilation performance with respect also to standard AUS. In particular, it boosts the efficiency of regime’s changes tracking, with a low computational cost. Other data assimilation schemes need estimates of ad hoc parameters, which have to be tuned for the specific model at hand. In Numerical Weather Prediction models, tuning of parameters — and in particular an estimate of the model error covariance matrix — may turn out to be quite difficult. Our proposed approach, instead, may be easier to implement in operational models.

The Enhanced Single-dish Control System and wide surveys of compact sources

The Italian radio telescopes currently undergo a major upgrade period in response to the growing demand for deep radio observations, such as surveys on large sky areas or observations of vast samples of compact radio sources. The optimised employment of the Italian antennas, at first constructed mainly for VLBI activities and provided with a control system (FS – Field System) not tailored to single-dish observations, required important modifications in particular of the guiding software and data acquisition system. The production of a completely new control system called ESCS (Enhanced Single-dish Control System) for the Medicina dish started in 2007, in synergy with the software development for the forthcoming Sardinia Radio Telescope (SRT). The aim is to produce a system optimised for single-dish observations in continuum, spectrometry and polarimetry. ESCS is also planned to be installed at the Noto site. A substantial part of this thesis work consisted in designing and developing subsystems within ESCS, in order to provide this software with tools to carry out large maps, spanning from the implementation of On-The-Fly fast scans (following both conventional and innovative observing strategies) to the production of single-dish standard output files and the realisation of tools for the quick-look of the acquired data. The test period coincided with the commissioning phase for two devices temporarily installed – while waiting for the SRT to be completed – on the Medicina antenna: a 18-26 GHz 7-feed receiver and the 14-channel analogue backend developed for its use. It is worth stressing that it is the only K-band multi-feed receiver at present available worldwide. The commissioning of the overall hardware/software system constituted a considerable section of the thesis work. Tests were led in order to verify the system stability and its capabilities, down to sensitivity levels which had never been reached in Medicina using the previous observing techniques and hardware devices. The aim was also to assess the scientific potential of the multi-feed receiver for the production of wide maps, exploiting its temporary availability on a mid-sized antenna. Dishes like the 32-m antennas at Medicina and Noto, in fact, offer the best conditions for large-area surveys, especially at high frequencies, as they provide a suited compromise between sufficiently large beam sizes to cover quickly large areas of the sky (typical of small-sized telescopes) and sensitivity (typical of large-sized telescopes). The KNoWS (K-band Northern Wide Survey) project is aimed at the realisation of a full-northern-sky survey at 21 GHz; its pilot observations, performed using the new ESCS tools and a peculiar observing strategy, constituted an ideal test-bed for ESCS itself and for the multi-feed/backend system. The KNoWS group, which I am part of, supported the commissioning activities also providing map-making and source-extraction tools, in order to complete the necessary data reduction pipeline and assess the general system scientific capabilities. The K-band observations, which were carried out in several sessions along the December 2008-March 2010 period, were accompanied by the realisation of a 5 GHz test survey during the summertime, which is not suitable for high-frequency observations. This activity was conceived in order to check the new analogue backend separately from the multi-feed receiver, and to simultaneously produce original scientific data (the 6-cm Medicina Survey, 6MS, a polar cap survey to complete PMN-GB6 and provide an all-sky coverage at 5 GHz).

High performance and energy-efficient instruction cache design and optimisation for ultra-low-power multi-core clusters

High Energy efficiency and high performance are the key regiments for Internet of Things (IoT) end-nodes. Exploiting cluster of multiple programmable processors has recently emerged as a suitable solution to address this challenge. However, one of the main bottlenecks for multi-core architectures is the instruction cache. While private caches fall into data replication and wasting area, fully shared caches lack scalability and form a bottleneck for the operating frequency. Hence we propose a hybrid solution where a larger shared cache (L1.5) is shared by multiple cores connected through a low-latency interconnect to small private caches (L1). However, it is still limited by large capacity miss with a small L1. Thus, we propose a sequential prefetch from L1 to L1.5 to improve the performance with little area overhead. Moreover, to cut the critical path for better timing, we optimized the core instruction fetch stage with non-blocking transfer by adopting a 4 x 32-bit ring buffer FIFO and adding a pipeline for the conditional branch. We present a detailed comparison of different instruction cache architectures' performance and energy efficiency recently proposed for Parallel Ultra-Low-Power clusters. On average, when executing a set of real-life IoT applications, our two-level cache improves the performance by up to 20% and loses 7% energy efficiency with respect to the private cache. Compared to a shared cache system, it improves performance by up to 17% and keeps the same energy efficiency. In the end, up to 20% timing (maximum frequency) improvement and software control enable the two-level instruction cache with prefetch adapt to various battery-powered usage cases to balance high performance and energy efficiency.

The FOOT experiment: Trigger and Data Acquisition (TDAQ) development and data analysis

Hadrontherapy employs high-energy beams of charged particles (protons and heavier ions) to treat deep-seated tumours: these particles have a favourable depth-dose distribution in tissue characterized by a low dose in the entrance channel and a sharp maximum (Bragg peak) near the end of their path. In these treatments nuclear interactions have to be considered: beam particles can fragment in the human body releasing a non-zero dose beyond the Bragg peak while fragments of human body nuclei can modify the dose released in healthy tissues. These effects are still in question given the lack of interesting cross sections data. Also space radioprotection can profit by fragmentation cross section measurements: the interest in long-term manned space missions beyond Low Earth Orbit is growing in these years but it has to cope with major health risks due to space radiation. To this end, risk models are under study: however, huge gaps in fragmentation cross sections data are currently present preventing an accurate benchmark of deterministic and Monte Carlo codes. To fill these gaps in data, the FOOT (FragmentatiOn Of Target) experiment was proposed. It is composed by two independent and complementary setups, an Emulsion Cloud Chamber and an electronic setup composed by several subdetectors providing redundant measurements of kinematic properties of fragments produced in nuclear interactions between a beam and a target. FOOT aims to measure double differential cross sections both in angle and kinetic energy which is the most complete information to address existing questions. In this Ph.D. thesis, the development of the Trigger and Data Acquisition system for the FOOT electronic setup and a first analysis of 400 MeV/u 16O beam on Carbon target data acquired in July 2021 at GSI (Darmstadt, Germany) are presented. When possible, a comparison with other available measurements is also reported.

Experimental Analysis and Numerical Simulation of Quench in Superconducting HTS Tapes and Coils

The quench characteristics of second generation (2 G) YBCO Coated Conductor (CC) tapes are of fundamental importance for the design and safe operation of superconducting cables and magnets based on this material. Their ability to transport high current densities at high temperature, up to 77 K, and at very high fields, over 20 T, together with the increasing knowledge in their manufacturing, which is reducing their cost, are pushing the use of this innovative material in numerous system applications, from high field magnets for research to motors and generators as well as for cables. The aim of this Ph. D. thesis is the experimental analysis and numerical simulations of quench in superconducting HTS tapes and coils. A measurements facility for the characterization of superconducting tapes and coils was designed, assembled and tested. The facility consist of a cryostat, a cryocooler, a vacuum system, resistive and superconducting current leads and signal feedthrough. Moreover, the data acquisition system and the software for critical current and quench measurements were developed. A 2D model was developed using the finite element code COMSOL Multiphysics R . The problem of modeling the high aspect ratio of the tape is tackled by multiplying the tape thickness by a constant factor, compensating the heat and electrical balance equations by introducing a material anisotropy. The model was then validated both with the results of a 1D quench model based on a non-linear electric circuit coupled to a thermal model of the tape, to literature measurements and to critical current and quench measurements made in the cryogenic facility. Finally the model was extended to the study of coils and windings with the definition of the tape and stack homogenized properties. The procedure allows the definition of a multi-scale hierarchical model, able to simulate the windings with different degrees of detail.

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.

Test vector generation for the phase II ATLAS event filter trigger upgrade

In the near future, the LHC experiments will continue to be upgraded as the LHC luminosity will increase from the design 1034 to 7.5 × 1034, with the HL-LHC project, to reach 3000 × f b−1 of accumulated statistics. After the end of a period of data collection, CERN will face a long shutdown to improve overall performance by upgrading the experiments and implementing more advanced technologies and infrastructures. In particular, ATLAS will upgrade parts of the detector, the trigger, and the data acquisition system. It will also implement new strategies and algorithms for processing and transferring the data to the final storage. This PhD thesis presents a study of a new pattern recognition algorithm to be used in the trigger system, which is a software designed to provide the information necessary to select physical events from background data. The idea is to use the well-known Hough Transform mathematical formula as an algorithm for detecting particle trajectories. The effectiveness of the algorithm has already been validated in the past, independently of particle physics applications, to detect generic shapes in images. Here, a software emulation tool is proposed for the hardware implementation of the Hough Transform, to reconstruct the tracks in the ATLAS Trigger and Data Acquisition system. Until now, it has never been implemented on electronics in particle physics experiments, and as a hardware implementation it would provide overall latency benefits. A comparison between the simulated data and the physical system was performed on a Xilinx UltraScale+ FPGA device.

Kinematics of the Sicily and Calabria Subduction System from Elastic Block Modeling of GPS Data

We use data from about 700 GPS stations in the EuroMediterranen region to investigate the present-day behavior of the the Calabrian subduction zone within the Mediterranean-scale plates kinematics and to perform local scale studies about the strain accumulation on active structures. We focus attenction on the Messina Straits and Crati Valley faults where GPS data show extentional velocity gradients of ∼3 mm/yr and ∼2 mm/yr, respectively. We use dislocation model and a non-linear constrained optimization algorithm to invert for fault geometric parameters and slip-rates and evaluate the associated uncertainties adopting a bootstrap approach. Our analysis suggest the presence of two partially locked normal faults. To investigate the impact of elastic strain contributes from other nearby active faults onto the observed velocity gradient we use a block modeling approach. Our models show that the inferred slip-rates on the two analyzed structures are strongly impacted by the assumed locking width of the Calabrian subduction thrust. In order to frame the observed local deformation features within the present- day central Mediterranean kinematics we realyze a statistical analysis testing the indipendent motion (w.r.t. the African and Eurasias plates) of the Adriatic, Cal- abrian and Sicilian blocks. Our preferred model confirms a microplate like behaviour for all the investigated blocks. Within these kinematic boundary conditions we fur- ther investigate the Calabrian Slab interface geometry using a combined approach of block modeling and χ2ν statistic. Almost no information is obtained using only the horizontal GPS velocities that prove to be a not sufficient dataset for a multi-parametric inversion approach. Trying to stronger constrain the slab geometry we estimate the predicted vertical velocities performing suites of forward models of elastic dislocations varying the fault locking depth. Comparison with the observed field suggest a maximum resolved locking depth of 25 km.

3D Surveying and Data Management towards the Realization of a Knowledge System for Cultural Heritage

The research activities involved the application of the Geomatic techniques in the Cultural Heritage field, following the development of two themes: Firstly, the application of high precision surveying techniques for the restoration and interpretation of relevant monuments and archaeological finds. The main case regards the activities for the generation of a high-fidelity 3D model of the Fountain of Neptune in Bologna. In this work, aimed to the restoration of the manufacture, both the geometrical and radiometrical aspects were crucial. The final product was the base of a 3D information system representing a shared tool where the different figures involved in the restoration activities shared their contribution in a multidisciplinary approach. Secondly, the arrangement of 3D databases for a Building Information Modeling (BIM) approach, in a process which involves the generation and management of digital representations of physical and functional characteristics of historical buildings, towards a so-called Historical Building Information Model (HBIM). A first application was conducted for the San Michele in Acerboli’s church in Santarcangelo di Romagna. The survey was performed by the integration of the classical and modern Geomatic techniques and the point cloud representing the church was used for the development of a HBIM model, where the relevant information connected to the building could be stored and georeferenced. A second application regards the domus of Obellio Firmo in Pompeii, surveyed by the integration of the classical and modern Geomatic techniques. An historical analysis permitted the definitions of phases and the organization of a database of materials and constructive elements. The goal is the obtaining of a federate model able to manage the different aspects: documental, analytic and reconstructive ones.

Data-stream driven Fuzzy-granular approaches for system maintenance

Intelligent systems are currently inherent to the society, supporting a synergistic human-machine collaboration. Beyond economical and climate factors, energy consumption is strongly affected by the performance of computing systems. The quality of software functioning may invalidate any improvement attempt. In addition, data-driven machine learning algorithms are the basis for human-centered applications, being their interpretability one of the most important features of computational systems. Software maintenance is a critical discipline to support automatic and life-long system operation. As most software registers its inner events by means of logs, log analysis is an approach to keep system operation. Logs are characterized as Big data assembled in large-flow streams, being unstructured, heterogeneous, imprecise, and uncertain. This thesis addresses fuzzy and neuro-granular methods to provide maintenance solutions applied to anomaly detection (AD) and log parsing (LP), dealing with data uncertainty, identifying ideal time periods for detailed software analyses. LP provides deeper semantics interpretation of the anomalous occurrences. The solutions evolve over time and are general-purpose, being highly applicable, scalable, and maintainable. Granular classification models, namely, Fuzzy set-Based evolving Model (FBeM), evolving Granular Neural Network (eGNN), and evolving Gaussian Fuzzy Classifier (eGFC), are compared considering the AD problem. The evolving Log Parsing (eLP) method is proposed to approach the automatic parsing applied to system logs. All the methods perform recursive mechanisms to create, update, merge, and delete information granules according with the data behavior. For the first time in the evolving intelligent systems literature, the proposed method, eLP, is able to process streams of words and sentences. Essentially, regarding to AD accuracy, FBeM achieved (85.64+-3.69)%; eGNN reached (96.17+-0.78)%; eGFC obtained (92.48+-1.21)%; and eLP reached (96.05+-1.04)%. Besides being competitive, eLP particularly generates a log grammar, and presents a higher level of model interpretability.