Dipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica - Tesi di Dottorato
Permanent URI for this collectionhttps://lisa.unical.it/handle/10955/31
Questa collezione raccoglie le Tesi di Dottorato afferenti al Dipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica dell'Università della Calabria.
Browse
216 results
Search Results
Item Ensemble of deep learning prediction models for data analytics(Università della Calabria, 2021-06-21) Zicari, Paolo; Fortino, Giancarlo; Folino, GianluigiThe abundance of available unstructured or raw text requires the automatic extraction of information for di↵erent tasks. One of the most relevant, Text Classification, extracts this information by assigning informative labels to raw texts from a pre-defined set. Deep Learning (DL) o↵ers challenging solutions to the automatic text classification problem. Despite the great potentialities of DL-based text classifiers, current solutions are exposed to a number of challenging issues that frequently occur in scenarios where text categorization is used in reallife applications. First of all, a large number of labelled data are usually necessary to train a deep model adequately, while labelling texts is timeconsuming, expensive, and very often requires specific knowledge. Moreover, configuring the structure and hyper-parameters of a Deep Neural Network (DNN) architecture is a difficult task, which entails long and careful design and tuning activities to make the DNN perform well. Typical scenarios are characterized by the fact that classes are often imbalanced. These issues entail a high risk of eventually obtaining a DNN-based classifier that overfits the training data and relies on non-general, biased and unreliable classification patterns. On the other hand, the black-box nature of a DNN model does not allow for easy reasoning on which features of a data instance drove the model to its classification decision. The work in this thesis, starting from the general problem of text classification, focuses on some challenging aspects associated with using an ensemble of deep learning methods to classify raw texts. More in detail, this work focuses on the analysis, exploration, study and test of algorithms and learning models to be employed in the proposal of novel techniques of Ensemble Deep Learning (EDL) aimed at performing classification and explanation tasks and on the research of semi-supervised strategies based on pseudo-labelling for improving classifier prediction performances in case of scarcity of labelled data. To this aim, this thesis proposes a complete framework based on the paradigm of ensembles of deep learning algorithms. The proposed framework is designed to furnish a valid instrument for exploring, validating and testing the proposed novel deep ensemble techniques contextualised in reallife applications, covering the entire classification process, including preprocessing, learning model building, explanation of the results, self-training for scarce labelled data, human-in-the-loop validating and model refining. Even though the methods proposed in this work could be used in any field of interest, the problem of extracting information from the raw text was specialised for two specific application contexts: automatic customer support ticket classification and the problem of fake detection. The first application scenario deals with the necessity of the Customer Care Department of most companies to answer their customer requests applied as tickets through several common channels like email, short message texts, social posts, etc. Ticket classification is necessary for automatic answer generation and routing to the specific human operator. Limiting the spread of misinformation, related to the high growth of social media dissemination and sharing of information, has raised the issue of distinguishing true news from fakes, with the challenging problem of processing long texts like news for fake detection. For this reason, the second scenario deals with the critical problem of discerning fake news from the vast amount of information circulating on the Web. In these research areas, the ensemble paradigm has been adopted only recently; thus, discovering the possible advantages when applying this technique is challenging. Experimental tests conducted on real data collected by two Customer Relationship Management (CRM) systems have proven the framework’s effectiveness in di↵erent ticket categorisation tasks and the practical value of their associated explanations. In addition, experiments conducted on two fake news datasets have proven the e↵ectiveness of the proposed semisupervised self-training ensemble-based strategy for improving performances when a few labelled data are available.Item Feature Selection in Classification by means of Optimization and Multi-Objective Optimization(Università della Calabria, 2023-05-10) Pirouz, Behzad; Fortino, Giancarlo; Gaudioso, ManlioThe thesis is in the area of mathematical optimization with application to Machine Learning. The focus is on Feature Selection (FS) in the framework of binary classification via Support Vector Machine paradigm. We concentrate on the use of sparse optimization techniques, which are widely considered as the election tool for tackling FS. We study the problem both in terms of single and multi-objective optimization. We propose first a novel Mixed-Integer Nonlinear Programming (MINLP) model for sparse optimization based on the polyhedral k-norm. We introduce a new way to take into account the k-norm for sparse optimization by setting a model based on fractional programming (FP). Then we address the continuous relaxation of the problem, which is reformulated via a DC (Difference of Convex) decomposition. On the other hand, designing supervised learning systems, in general, is a multi-objective problem. It requires finding appropriate trade-offs between several objectives, for example, between the number of misclassified training data (minimizing the squared error) and the number of nonzero elements separating the hyperplane (minimizing the number of nonzero elements). When we deal with multi-objective optimization problems, the optimization problem has yet to have a single solution that represents the best solution for all objectives simultaneously. Consequently, there is not a single solution but a set of solutions, known as the Pareto-optimal solutions. We overview the SVM models and the related Feature Selection in terms of multi-objective optimization. Our multi-objective approach considers two simultaneous objectives: minimizing the squared error and minimizing the number of nonzero elements of the normal vector of the separator hyperplane. In this thesis, we propose a multi-objective model for sparse optimization. Our primary purpose is to demonstrate the advantages of considering SVM models as multi-objective optimization problems. In multi-objective cases, we can obtain a set of Pareto optimal solutions instead of one in single-objective cases. Therefore, our main contribution in this thesis is of two levels: first, we propose a new model for sparse optimization based on the polyhedral k-norm for SVM classification, and second, use multi-objective optimization to consider this new model. The results of several numerical experiments on some classification datasets are reported. We used all the datasets for single-objective and multi-objective models.Item Design of physically unclonable functions in cmos and emerging technologies for hardware security applications(Università della Calabria, 2023-02-23) Vatalaro, Massimo; Fortino, Giancarlo; Crupi, FeliceThe advent of the IoT scenario heavily pushed the demand of preserving the information down to the chip level due to the increasing demand of interconnected devices. Novel algorithms and hardware architectures are developed every year with the aim of making these systems more and more secure. However, IoT devices operate with constrained area, energy and budget thus making the hardware implementation of these architectures not always feasible. Moreover, these algorithms require truly random key for guarantying a certain security degree. Typically, these secret keys are generated off chip and stored in a non-volatile manner. Unfortunately, this approach requires additional costs and suffers from reverse engineering attacks. Physically unclonable functions (PUFs) are emerging cryptographic primitives which exploit random phenomena, such as random process variations in CMOS manufacturing processes, for generating a unique, repeatable, random, and secure keys in a volatile manner, like a digital fingerprint. PUFs represent a secure and low-cost solution for implementing lightweight cryptographic algorithms. Ideally PUF data should be unique and repeatable even under noisy or different environmental conditions. Unfortunately, guarantying a proper stability is still challenging, especially under PVT variations, thus requiring stability enhancement techniques which overtake the PUF itself in terms of required area and energy. Nowadays, different PUF solutions have been proposed with the aim of achieving ever more stable responses while keeping the area overhead low. This thesis presents a novel class of static monostable PUFs based on a voltage divider between two nominally identical sub-circuits. The fully static behavior along with the use of nominally identical sub-circuits ensure that the correct output is always delivered even when on-chip noise occasionally flips the bit, and that randomness is always guaranteed regardless of the PVT conditions. Measurement results in 180-nm CMOS technology demonstrates the effectiveness of the proposed solution with a native instability (BER) of only 0.61% (0.13%) along with a low sensitivity to both temperature and voltage variations. However, these results were achieved at the cost of more area-hungry design (i.e., 7,222𝐹 ) compared to other relevant works. The proposed solution was also implemented with emerging paper based MoS2 nFETs by exploiting a LUT-based Verilog-A model, calibrated with experimental 𝐼 vs 𝑉 at different 𝑉 curves, whose variability was extracted from different 𝐼 vs 𝑉 curves of 27 devices from the same manufacturing lot. Simulations results demonstrate that these devices can potentially used as building block for next generation electronics targeting hardware security applications. Finally, this thesis also provides an application scenario, in which the proposed PUF solution is employed as TRNG module for implementing a smart tag targeting anti-counterfeiting applications.Item Distributed Big Social Data Analysis: Advanced Techniques and Execution Strategies(Università della Calabria, 2023-05-16) Cantini, Riccardo; Fortino, Giancarlo; Trunfio, Paolo; Marozzo, FabrizioItem A logical and ontological framework for metadata extraction and modelling from heterogeneous document sources(Università della Calabria, 2023-06-24) Cuconato, Simone; Fortino, Giancarlo; Folino, Antonietta; Cardillo, ElenaItem Design Methodologies for FPGA-based Deep Learning Accelerators and Their Characterization(Università della Calabria, 2023) Sestito, Cristian; Fortino, Giancarlo; Perri, Stefania; Corsonello, PasqualeDeep Neural Networks (DNNs) are widespread in many applications, including computer vision, speech recognition and robotics, thanks to the ability of such models to extract information by building a hierarchical representation of knowledge. Image processing benefits from the latter behavior by using Convolutional Neural Networks (CNNs), which consist of several Convolutional (CONV) layers to extract features from inputs at different levels of abstraction. However, CNNs usually require billions of computations to reach high accuracy levels. In order to sustain such computational load, proper hardware acceleration is needed. Field Programmable Gate Arrays (FPGAs) have been shown as promising candidates, because they are able to achieve high throughput at limited power dissipation. In addition, FPGAs are flexible architectures to accommodate several CNNs’ workloads. While the hardware acceleration of conventional CNN models has been widely investigated, the interest about more sophisticated tasks is still emerging. The latter includes CNNs based on Dilated Convolutions (DCONVs) and Transposed Convolutions (TCONVs), which deal with filter and image dilations, respectively. Accordingly, higher computational complexity is exhibited by these architectures, thus requiring careful hardware management. This PhD dissertation deals with the FPGA acceleration of CNNs for Image Processing based on DCONVs and TCONVs. Specifically, several designs using both the Very High-Speed Integrated Circuits Hardware Description Language (VHDL) and the High-Level Synthesis (HLS) are presented. Detailed characterization is discussed, based on the evaluation of resources occupation, throughput, power dissipation, as well as the impact of data quantization. Overall, the proposed circuits show noticeable energyefficiency when compared to several state-of-the-art counterparts. For instance, hardware acceleration of run-time reconfigurable CONVs and TCONVs for super-resolution imaging has shown an energy-efficiency of up to 518.5 GOPS/W, by outperforming stateof- the-art competitors by up to 2.3 times.Item An active learning Approach based on learning models' parameters exploitation(Università della Calabria, 2023-07-06) Scala, Francesco; Flesca, SergioArti cial Intelligence (AI) techniques and in particular Machine and Deep Learning (ML and DL), have been widely adopted to enhance various aspects of human life. ML algorithms can be categorized into four main types: Supervised Learning, Unsupervised Learning, Semi-supervised Learning, and Reinforcement Learning. A signi cant challenge in these techniques is the requirement for su cient labeled data for training. Active Learning (AL) is a machine learning framework that addresses this issue by selecting instances to be labeled in a smart way to optimize model training, i.e., AL reduces labeling time and leads to better-performing models by dynamically selecting the most representative samples to be labeled during the training phase. AL was proven to be e ective in di erent scenarios and its choice of querying a label depends on the cost and gain of obtaining the information. In this thesis, are presented two novel approaches for active learning in meta-learning models. The proposed methods, called LAL-IGradV and LAL-IGradV-VAE, select instances to be labeled using an estimate of their impact on the current classi er. This is achieved by evaluating the importance of previously labeled instances in training the classi cation model and training another model that estimates the importance of unlabeled instances. The approaches can be instantiated with any classi er that is trainable through gradient descent optimization, and in this study, is provided a formulation using a deep neural network. These approaches have not been thoroughly investigated in previous learning-to-active-learn methods and experimental results demonstrate its promising performance in scenarios where there are only a limited number of initially labeled instances. 2Item Inverse source and scattering solutions for microwave imaging applications(Università della Calabria, 2023-02-02) Lopez, Giuseppe; Fortino, Giancarlo; Costanzo, SandraItem Global Optimization and Fractal Curves(Università della Calabria, 2022-07-13) Nasso, Maria Chiara; Crupi, Felice; Sergeev, YaroslavIl presente lavoro di tesi è principalmente dedicato all'ottimizzazione globale e in particolare a metodi numerici di ottimizzazione globale basati su frattali. Viene affrontato lo studio teorico di alcune curve frattali, vengono proposti nuovi algoritmi che si basano su approcci frattali per ridurre la dimensione del problema e vengono introdotti nuovi metodi di ottimizzazione globale basati sulla tecnica del Local Tuning. Ciascuno dei nuovi metodi proposti è stato implementato e studiato dal punto di vista teorico. Inoltre, gli esperimenti numerici, condotti su diverse centinaia di funzioni test, tratte dalla letteratura e generate in maniera random confermano i vantaggi degli algoritmi presentati.Item Heterogeneous FPGA-based Embedded Systems for Vision IoT Applications(Università della Calabria, 2020-04-23) Spagnolo, Fanny; Crupi, Felice; Perri, Stefania; Corsonello, PasqualeEmbedded sensor devices provided by processing capabilities are opening novel and exciting opportunities in the era of edge-computing Internet-of-Things (IoT). The workload decentralization leads to a plenty of benefits, including better reactivity and reliability and reduced data transfer costs. These advantages have a strong impact especially in the visual IoT field, for which the large bandwidth required by visual data is one of the most critical challenges. However, bringing vision technologies into smart nodes is not a trivial task, because of the stringent energy and performance requirements, in addition to the need of cost-effective and compact processing units. Heterogeneous architectures may represent the key to address these necessities. Among possible heterogeneous platforms, those based on reconfigurable devices such as Field Programmable Gate Arrays (FPGAs) show a high adaptability to a variety of workloads, which is an important goal for edge-computing. Therefore, their deployment in disparate IoT applications, ranging from video surveillance to autonomous driving, is emerging as a promising solution. This dissertation proposes a study on the suitability of modern heterogeneous FPGA System-on-Chips (SoCs) to implement embedded smart vision sensor nodes. To this purpose, several computer vision algorithms aimed to extract synthetic data from raw input frames have been analysed, and novel hardware-oriented solutions have been proposed to deploy them on heterogeneous SoCs. In all the presented cases, ranging from stereo vision to connected component analysis and deep learning, speed performances and/or energy efficiency are considerably improved with respect to state-of-the-art solutions. As an example, the proposed heterogeneous architecture for convolutional neural networks achieves a power efficiency up to 89.5% higher than competitive prior works, demonstrating its suitability in the scenario of energy-constrained and real-time IoT.