Supporting Scientific Research Through Machine and Deep Learning: Fluorescence Microscopy and Operational Intelligence Use Cases

Although the debate of what data science is has a long history and has not reached a complete consensus yet, Data Science can be summarized as the process of learning from data. Guided by the above vision, this thesis presents two independent data science projects developed in the scope of multidisciplinary applied research. The first part analyzes fluorescence microscopy images typically produced in life science experiments, where the objective is to count how many marked neuronal cells are present in each image. Aiming to automate the task for supporting research in the area, we propose a neural network architecture tuned specifically for this use case, cell ResUnet (c-ResUnet), and discuss the impact of alternative training strategies in overcoming particular challenges of our data. The approach provides good results in terms of both detection and counting, showing performance comparable to the interpretation of human operators. As a meaningful addition, we release the pre-trained model and the Fluorescent Neuronal Cells dataset collecting pixel-level annotations of where neuronal cells are located. In this way, we hope to help future research in the area and foster innovative methodologies for tackling similar problems. The second part deals with the problem of distributed data management in the context of LHC experiments, with a focus on supporting ATLAS operations concerning data transfer failures. In particular, we analyze error messages produced by failed transfers and propose a Machine Learning pipeline that leverages the word2vec language model and K-means clustering. This provides groups of similar errors that are presented to human operators as suggestions of potential issues to investigate. The approach is demonstrated on one full day of data, showing promising ability in understanding the message content and providing meaningful groupings, in line with previously reported incidents by human operators.

Enabling 5G and beyond: millimeter wave propagation and cloud RAN evolution

In the last few years, mobile wireless technology has gone through a revolutionary change. Web-enabled devices have evolved into essential tools for communication, information, and entertainment. The fifth generation (5G) of mobile communication networks is envisioned to be a key enabler of the next upcoming wireless revolution. Millimeter wave (mmWave) spectrum and the evolution of Cloud Radio Access Networks (C-RANs) are two of the main technological innovations of 5G wireless systems and beyond. Because of the current spectrum-shortage condition, mmWaves have been proposed for the next generation systems, providing larger bandwidths and higher data rates. Consequently, new radio channel models are being developed. Recently, deterministic ray-based models such as Ray-Tracing (RT) are getting more attractive thanks to their frequency-agility and reliable predictions. A modern RT software has been calibrated and used to analyze the mmWave channel. Knowledge of the electromagnetic properties of materials is therefore essential. Hence, an item-level electromagnetic characterization of common construction materials has been successfully achieved to obtain information about their complex relative permittivity. A complete tuning of the RT tool has been performed against indoor and outdoor measurement campaigns at 27 and 38 GHz, setting the basis for the future development of advanced beamforming techniques which rely on deterministic propagation models (as RT). C-RAN is a novel mobile network architecture which can address a number of challenges that network operators are facing in order to meet the continuous customers’ demands. C-RANs have already been adopted in advanced 4G deployments; however, there are still some issues to deal with, especially considering the bandwidth requirements set by the forthcoming 5G systems. Open RAN specifications have been proposed to overcome the new 5G challenges set on C-RAN architectures, including synchronization aspects. In this work it is described an FPGA implementation of the Synchronization Plane for an O-RAN-compliant radio system.

QoS-aware architectures, technologies, and middleware for the cloud continuum

The recent trend of moving Cloud Computing capabilities to the Edge of the network is reshaping how applications and their middleware supports are designed, deployed, and operated. This new model envisions a continuum of virtual resources between the traditional cloud and the network edge, which is potentially more suitable to meet the heterogeneous Quality of Service (QoS) requirements of diverse application domains and next-generation applications. Several classes of advanced Internet of Things (IoT) applications, e.g., in the industrial manufacturing domain, are expected to serve a wide range of applications with heterogeneous QoS requirements and call for QoS management systems to guarantee/control performance indicators, even in the presence of real-world factors such as limited bandwidth and concurrent virtual resource utilization. The present dissertation proposes a comprehensive QoS-aware architecture that addresses the challenges of integrating cloud infrastructure with edge nodes in IoT applications. The architecture provides end-to-end QoS support by incorporating several components for managing physical and virtual resources. The proposed architecture features: i) a multilevel middleware for resolving the convergence between Operational Technology (OT) and Information Technology (IT), ii) an end-to-end QoS management approach compliant with the Time-Sensitive Networking (TSN) standard, iii) new approaches for virtualized network environments, such as running TSN-based applications under Ultra-low Latency (ULL) constraints in virtual and 5G environments, and iv) an accelerated and deterministic container overlay network architecture. Additionally, the QoS-aware architecture includes two novel middlewares: i) a middleware that transparently integrates multiple acceleration technologies in heterogeneous Edge contexts and ii) a QoS-aware middleware for Serverless platforms that leverages coordination of various QoS mechanisms and virtualized Function-as-a-Service (FaaS) invocation stack to manage end-to-end QoS metrics. Finally, all architecture components were tested and evaluated by leveraging realistic testbeds, demonstrating the efficacy of the proposed solutions.

Data driven regularization models of non-linear ill-posed inverse problems in imaging

Imaging technologies are widely used in application fields such as natural sciences, engineering, medicine, and life sciences. A broad class of imaging problems reduces to solve ill-posed inverse problems (IPs). Traditional strategies to solve these ill-posed IPs rely on variational regularization methods, which are based on minimization of suitable energies, and make use of knowledge about the image formation model (forward operator) and prior knowledge on the solution, but lack in incorporating knowledge directly from data. On the other hand, the more recent learned approaches can easily learn the intricate statistics of images depending on a large set of data, but do not have a systematic method for incorporating prior knowledge about the image formation model. The main purpose of this thesis is to discuss data-driven image reconstruction methods which combine the benefits of these two different reconstruction strategies for the solution of highly nonlinear ill-posed inverse problems. Mathematical formulation and numerical approaches for image IPs, including linear as well as strongly nonlinear problems are described. More specifically we address the Electrical impedance Tomography (EIT) reconstruction problem by unrolling the regularized Gauss-Newton method and integrating the regularization learned by a data-adaptive neural network. Furthermore we investigate the solution of non-linear ill-posed IPs introducing a deep-PnP framework that integrates the graph convolutional denoiser into the proximal Gauss-Newton method with a practical application to the EIT, a recently introduced promising imaging technique. Efficient algorithms are then applied to the solution of the limited electrods problem in EIT, combining compressive sensing techniques and deep learning strategies. Finally, a transformer-based neural network architecture is adapted to restore the noisy solution of the Computed Tomography problem recovered using the filtered back-projection method.

An architecture for network acceleration as a service in the cloud continuum

The pervasive availability of connected devices in any industrial and societal sector is pushing for an evolution of the well-established cloud computing model. The emerging paradigm of the cloud continuum embraces this decentralization trend and envisions virtualized computing resources physically located between traditional datacenters and data sources. By totally or partially executing closer to the network edge, applications can have quicker reactions to events, thus enabling advanced forms of automation and intelligence. However, these applications also induce new data-intensive workloads with low-latency constraints that require the adoption of specialized resources, such as high-performance communication options (e.g., RDMA, DPDK, XDP, etc.). Unfortunately, cloud providers still struggle to integrate these options into their infrastructures. That risks undermining the principle of generality that underlies the cloud computing scale economy by forcing developers to tailor their code to low-level APIs, non-standard programming models, and static execution environments. This thesis proposes a novel system architecture to empower cloud platforms across the whole cloud continuum with Network Acceleration as a Service (NAaaS). To provide commodity yet efficient access to acceleration, this architecture defines a layer of agnostic high-performance I/O APIs, exposed to applications and clearly separated from the heterogeneous protocols, interfaces, and hardware devices that implement it. A novel system component embodies this decoupling by offering a set of agnostic OS features to applications: memory management for zero-copy transfers, asynchronous I/O processing, and efficient packet scheduling. This thesis also explores the design space of the possible implementations of this architecture by proposing two reference middleware systems and by adopting them to support interactive use cases in the cloud continuum: a serverless platform and an Industry 4.0 scenario. A detailed discussion and a thorough performance evaluation demonstrate that the proposed architecture is suitable to enable the easy-to-use, flexible integration of modern network acceleration into next-generation cloud platforms.

Reliable and Variation-Tolerant Interconnection Network for Low Power MPSOCS

Multi-Processor SoC (MPSOC) design brings to the foreground a large number of challenges, one of the most prominent of which is the design of the chip interconnection. With a number of on-chip blocks presently ranging in the tens, and quickly approaching the hundreds, the novel issue of how to best provide on-chip communication resources is clearly felt. Scaling down of process technologies has increased process and dynamic variations as well as transistor wearout. Because of this, delay variations increase and impact the performance of the MPSoCs. The interconnect architecture inMPSoCs becomes a single point of failure as it connects all other components of the system together. A faulty processing element may be shut down entirely, but the interconnect architecture must be able to tolerate partial failure and variations and operate with performance, power or latency overhead. This dissertation focuses on techniques at different levels of abstraction to face with the reliability and variability issues in on-chip interconnection networks. By showing the test results of a GALS NoC testchip this dissertation motivates the need for techniques to detect and work around manufacturing faults and process variations in MPSoCs’ interconnection infrastructure. As a physical design technique, we propose the bundle routing framework as an effective way to route the Network on Chips’ global links. For architecture-level design, two cases are addressed: (I) Intra-cluster communication where we propose a low-latency interconnect with variability robustness (ii) Inter-cluster communication where an online functional testing with a reliable NoC configuration are proposed. We also propose dualVdd as an orthogonal way of compensating variability at the post-fabrication stage. This is an alternative strategy with respect to the design techniques, since it enforces the compensation at post silicon stage.

An architecture for scaling ontology networks

Constructing ontology networks typically occurs at design time at the hands of knowledge engineers who assemble their components statically. There are, however, use cases where ontology networks need to be assembled upon request and processed at runtime, without altering the stored ontologies and without tampering with one another. These are what we call "virtual [ontology] networks", and keeping track of how an ontology changes in each virtual network is called "multiplexing". Issues may arise from the connectivity of ontology networks. In many cases, simple flat import schemes will not work, because many ontology managers can cause property assertions to be erroneously interpreted as annotations and ignored by reasoners. Also, multiple virtual networks should optimize their cumulative memory footprint, and where they cannot, this should occur for very limited periods of time. We claim that these problems should be handled by the software that serves these ontology networks, rather than by ontology engineering methodologies. We propose a method that spreads multiple virtual networks across a 3-tier structure, and can reduce the amount of erroneously interpreted axioms, under certain raw statement distributions across the ontologies. We assumed OWL as the core language handled by semantic applications in the framework at hand, due to the greater availability of reasoners and rule engines. We also verified that, in common OWL ontology management software, OWL axiom interpretation occurs in the worst case scenario of pre-order visit. To measure the effectiveness and space-efficiency of our solution, a Java and RESTful implementation was produced within an Apache project. We verified that a 3-tier structure can accommodate reasonably complex ontology networks better, in terms of the expressivity OWL axiom interpretation, than flat-tree import schemes can. We measured both the memory overhead of the additional components we put on top of traditional ontology networks, and the framework's caching capabilities.

Application Platforms for the Internet of Things: Theory, Architecture, Protocols, Data Formats, and Privacy

The Internet of Things (IoT) is the next industrial revolution: we will interact naturally with real and virtual devices as a key part of our daily life. This technology shift is expected to be greater than the Web and Mobile combined. As extremely different technologies are needed to build connected devices, the Internet of Things field is a junction between electronics, telecommunications and software engineering. Internet of Things application development happens in silos, often using proprietary and closed communication protocols. There is the common belief that only if we can solve the interoperability problem we can have a real Internet of Things. After a deep analysis of the IoT protocols, we identified a set of primitives for IoT applications. We argue that each IoT protocol can be expressed in term of those primitives, thus solving the interoperability problem at the application protocol level. Moreover, the primitives are network and transport independent and make no assumption in that regard. This dissertation presents our implementation of an IoT platform: the Ponte project. Privacy issues follows the rise of the Internet of Things: it is clear that the IoT must ensure resilience to attacks, data authentication, access control and client privacy. We argue that it is not possible to solve the privacy issue without solving the interoperability problem: enforcing privacy rules implies the need to limit and filter the data delivery process. However, filtering data require knowledge of how the format and the semantics of the data: after an analysis of the possible data formats and representations for the IoT, we identify JSON-LD and the Semantic Web as the best solution for IoT applications. Then, this dissertation present our approach to increase the throughput of filtering semantic data by a factor of ten.

Neural network modelling of the wave-structure interaction processes

This thesis presents a new Artificial Neural Network (ANN) able to predict at once the main parameters representative of the wave-structure interaction processes, i.e. the wave overtopping discharge, the wave transmission coefficient and the wave reflection coefficient. The new ANN has been specifically developed in order to provide managers and scientists with a tool that can be efficiently used for design purposes. The development of this ANN started with the preparation of a new extended and homogeneous database that collects all the available tests reporting at least one of the three parameters, for a total amount of 16’165 data. The variety of structure types and wave attack conditions in the database includes smooth, rock and armour unit slopes, berm breakwaters, vertical walls, low crested structures, oblique wave attacks. Some of the existing ANNs were compared and improved, leading to the selection of a final ANN, whose architecture was optimized through an in-depth sensitivity analysis to the training parameters of the ANN. Each of the selected 15 input parameters represents a physical aspect of the wave-structure interaction process, describing the wave attack (wave steepness and obliquity, breaking and shoaling factors), the structure geometry (submergence, straight or non-straight slope, with or without berm or toe, presence or not of a crown wall), or the structure type (smooth or covered by an armour layer, with permeable or impermeable core). The advanced ANN here proposed provides accurate predictions for all the three parameters, and demonstrates to overcome the limits imposed by the traditional formulae and approach adopted so far by some of the existing ANNs. The possibility to adopt just one model to obtain a handy and accurate evaluation of the overall performance of a coastal or harbor structure represents the most important and exportable result of the work.

Optimizing AI at the Edge: from network topology design to MCU deployment

The first topic analyzed in the thesis will be Neural Architecture Search (NAS). I will focus on two different tools that I developed, one to optimize the architecture of Temporal Convolutional Networks (TCNs), a convolutional model for time-series processing that has recently emerged, and one to optimize the data precision of tensors inside CNNs. The first NAS proposed explicitly targets the optimization of the most peculiar architectural parameters of TCNs, namely dilation, receptive field, and the number of features in each layer. Note that this is the first NAS that explicitly targets these networks. The second NAS proposed instead focuses on finding the most efficient data format for a target CNN, with the granularity of the layer filter. Note that applying these two NASes in sequence allows an "application designer" to minimize the structure of the neural network employed, minimizing the number of operations or the memory usage of the network. After that, the second topic described is the optimization of neural network deployment on edge devices. Importantly, exploiting edge platforms' scarce resources is critical for NN efficient execution on MCUs. To do so, I will introduce DORY (Deployment Oriented to memoRY) -- an automatic tool to deploy CNNs on low-cost MCUs. DORY, in different steps, can manage different levels of memory inside the MCU automatically, offload the computation workload (i.e., the different layers of a neural network) to dedicated hardware accelerators, and automatically generates ANSI C code that orchestrates off- and on-chip transfers with the computation phases. On top of this, I will introduce two optimized computation libraries that DORY can exploit to deploy TCNs and Transformers on edge efficiently. I conclude the thesis with two different applications on bio-signal analysis, i.e., heart rate tracking and sEMG-based gesture recognition.