Tensor-Train decomposition for image classification problems

In these last years a great effort has been put in the development of new techniques for automatic object classification, also due to the consequences in many applications such as medical imaging or driverless cars. To this end, several mathematical models have been developed from logistic regression to neural networks. A crucial aspect of these so called classification algorithms is the use of algebraic tools to represent and approximate the input data. In this thesis, we examine two different models for image classification based on a particular tensor decomposition named Tensor-Train (TT) decomposition. The use of tensor approaches preserves the multidimensional structure of the data and the neighboring relations among pixels. Furthermore the Tensor-Train, differently from other tensor decompositions, does not suffer from the curse of dimensionality making it an extremely powerful strategy when dealing with high-dimensional data. It also allows data compression when combined with truncation strategies that reduce memory requirements without spoiling classification performance. The first model we propose is based on a direct decomposition of the database by means of the TT decomposition to find basis vectors used to classify a new object. The second model is a tensor dictionary learning model, based on the TT decomposition where the terms of the decomposition are estimated using a proximal alternating linearized minimization algorithm with a spectral stepsize.

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.

Image analysis in the morphological and morphometric study of teeth

The subject of this doctoral dissertation concerns the definition of a new methodology for the morphological and morphometric study of fossilized human teeth, and therefore strives to provide a contribution to the reconstruction of human evolutionary history that proposes to extend to the different species of hominid fossils. Standardized investigative methodologies are lacking both regarding the orientation of teeth subject to study and in the analysis that can be carried out on these teeth once they are oriented. The opportunity to standardize a primary analysis methodology is furnished by the study of certain early Neanderthal and preneanderthal molars recovered in two caves in southern Italy [Grotta Taddeo (Taddeo Cave) and Grotta del Poggio (Poggio Cave), near Marina di Camerata, Campania]. To these we can add other molars of Neanderthal and modern man of the upper Paleolithic era, specifically scanned in the paleoanthropology laboratory of the University of Arkansas (Fayetteville, Arkansas, USA), in order to increase the paleoanthropological sample data and thereby make the final results of the analyses more significant. The new analysis methodology is rendered as follows: 1. Standardization of an orientation system for primary molars (superior and inferior), starting from a scan of a sample of 30 molars belonging to modern man (15 M1 inferior and 15 M1 superior), the definition of landmarks, the comparison of various systems and the choice of a system of orientation for each of the two dental typologies. 2. The definition of an analysis procedure that considers only the first 4 millimeters of the dental crown starting from the collar: 5 sections parallel to the plane according to which the tooth has been oriented are carried out, spaced 1 millimeter between them. The intention is to determine a method that allows for the differentiation of fossilized species even in the presence of worn teeth. 3. Results and Conclusions. The new approach to the study of teeth provides a considerable quantity of information that can better be evaluated by increasing the fossil sample data. It has been demonstrated to be a valid tool in evolutionary classification that has allowed (us) to differentiate the Neanderthal sample from that of modern man. In a particular sense the molars of Grotta Taddeo, which up until this point it has not been possible to determine with exactness their species of origin, through the present research they are classified as Neanderthal.

Computational modelling of human stem cell-derived cardiomyocytes - Texture descriptors for biological image processing

This thesis investigates two distinct research topics. The main topic (Part I) is the computational modelling of cardiomyocytes derived from human stem cells, both embryonic (hESC-CM) and induced-pluripotent (hiPSC-CM). The aim of this research line lies in developing models of the electrophysiology of hESC-CM and hiPSC-CM in order to integrate the available experimental data and getting in-silico models to be used for studying/making new hypotheses/planning experiments on aspects not fully understood yet, such as the maturation process, the functionality of the Ca2+ hangling or why the hESC-CM/hiPSC-CM action potentials (APs) show some differences with respect to APs from adult cardiomyocytes. Chapter I.1 introduces the main concepts about hESC-CMs/hiPSC-CMs, the cardiac AP, and computational modelling. Chapter I.2 presents the hESC-CM AP model, able to simulate the maturation process through two developmental stages, Early and Late, based on experimental and literature data. Chapter I.3 describes the hiPSC-CM AP model, able to simulate the ventricular-like and atrial-like phenotypes. This model was used to assess which currents are responsible for the differences between the ventricular-like AP and the adult ventricular AP. The secondary topic (Part II) consists in the study of texture descriptors for biological image processing. Chapter II.1 provides an overview on important texture descriptors such as Local Binary Pattern or Local Phase Quantization. Moreover the non-binary coding and the multi-threshold approach are here introduced. Chapter II.2 shows that the non-binary coding and the multi-threshold approach improve the classification performance of cellular/sub-cellular part images, taken from six datasets. Chapter II.3 describes the case study of the classification of indirect immunofluorescence images of HEp2 cells, used for the antinuclear antibody clinical test. Finally the general conclusions are reported.