Bidirectional Helmholtz Machines

L'entraînement sans surveillance efficace et inférence dans les modèles génératifs profonds reste un problème difficile. Une approche assez simple, la machine de Helmholtz, consiste à entraîner du haut vers le bas un modèle génératif dirigé qui sera utilisé plus tard pour l'inférence approximative. Des résultats récents suggèrent que de meilleurs modèles génératifs peuvent être obtenus par de meilleures procédures d'inférence approximatives. Au lieu d'améliorer la procédure d'inférence, nous proposons ici un nouveau modèle, la machine de Helmholtz bidirectionnelle, qui garantit qu'on peut calculer efficacement les distributions de haut-vers-bas et de bas-vers-haut. Nous y parvenons en interprétant à les modèles haut-vers-bas et bas-vers-haut en tant que distributions d'inférence approximative, puis ensuite en définissant la distribution du modèle comme étant la moyenne géométrique de ces deux distributions. Nous dérivons une borne inférieure pour la vraisemblance de ce modèle, et nous démontrons que l'optimisation de cette borne se comporte en régulisateur. Ce régularisateur sera tel que la distance de Bhattacharyya sera minisée entre les distributions approximatives haut-vers-bas et bas-vers-haut. Cette approche produit des résultats de pointe en terme de modèles génératifs qui favorisent les réseaux significativement plus profonds. Elle permet aussi une inférence approximative amérliorée par plusieurs ordres de grandeur. De plus, nous introduisons un modèle génératif profond basé sur les modèles BiHM pour l'entraînement semi-supervisé.

The Processing of Preposition-Stranding Constructions in English

One of the prominent questions in modern psycholinguistics is the relationship between the grammar and the parser. Within the approach of Generative Grammar, this issue has been investigated in terms of the role that Principles of Universal Grammar may play in language processing. The aim of this research experiment is to investigate this topic. Specifically, this experiment aims to test whether the Minimal Structure Principle (MSP) plays a role in the processing of Preposition-Stranding versus Pied-Piped Constructions. This investigation is made with a self-paced reading task, an on-line processing test that measures participants’ unconscious reaction to language stimuli. Monolingual English speakers’ reading times of sentences with Preposition-Stranding and Pied-Piped Constructions are compared. Results indicate that neither construction has greater processing costs, suggesting that factors other than the MSP are active during language processing.

Bidirectional Helmholtz Machines

L'entraînement sans surveillance efficace et inférence dans les modèles génératifs profonds reste un problème difficile. Une approche assez simple, la machine de Helmholtz, consiste à entraîner du haut vers le bas un modèle génératif dirigé qui sera utilisé plus tard pour l'inférence approximative. Des résultats récents suggèrent que de meilleurs modèles génératifs peuvent être obtenus par de meilleures procédures d'inférence approximatives. Au lieu d'améliorer la procédure d'inférence, nous proposons ici un nouveau modèle, la machine de Helmholtz bidirectionnelle, qui garantit qu'on peut calculer efficacement les distributions de haut-vers-bas et de bas-vers-haut. Nous y parvenons en interprétant à les modèles haut-vers-bas et bas-vers-haut en tant que distributions d'inférence approximative, puis ensuite en définissant la distribution du modèle comme étant la moyenne géométrique de ces deux distributions. Nous dérivons une borne inférieure pour la vraisemblance de ce modèle, et nous démontrons que l'optimisation de cette borne se comporte en régulisateur. Ce régularisateur sera tel que la distance de Bhattacharyya sera minisée entre les distributions approximatives haut-vers-bas et bas-vers-haut. Cette approche produit des résultats de pointe en terme de modèles génératifs qui favorisent les réseaux significativement plus profonds. Elle permet aussi une inférence approximative amérliorée par plusieurs ordres de grandeur. De plus, nous introduisons un modèle génératif profond basé sur les modèles BiHM pour l'entraînement semi-supervisé.

Propuesta de un análisis metafórico-poético en una novela de ciencia ficción: 2001, una odisea del espacio

The aim of this doctoral thesis is to analyse both the text (production) and the set of cognition processes which facilitate the understanding of a masterpiece of Science Fiction: 2001, A Space Odyssey from a new perspective. Unlike other literary theories, texts in cognitive poetics (i.e. Structuralism, Generative Linguistics, Literary Criticism) are projected on the readerś minds by means of cognitive procedures. Cognitive poetics uses tools of cognitive science in order to understand literature (and in this particular case, Science Fiction). This has lead to a great change in our understanding of texts, literary and otherwise. This literary analysis was carried out on the Science Fiction genre based on the assumptions of cognitive poetics. Due to the peculiar kind of subjects touched upon by this genre, it has been theorised that the type of metaphor used for both its creation and style are rather different from other genres. Science Fiction coincides with other genres of writing in that it contains its own specific ways of providing language with a meaning while making it so innovative that many literary theories point out that it is written differently or that it is necessary to possess single reading codes in order to be fully understood. The methodology used for this thesis required a metaphorical basic and poetic conceptual analysis according to the poetical metaphor theory developed by Lakoff and Turner (1989). The aim of this metaphorical study is to analyse the basic and conceptual structure of 2001, A Space Odyssey by A. C. Clarke. The parameters of literary conceptualization of Science Fiction established by Peter Stockwell in his Poetics of Science Fiction (2000) were also applied in order to analyse this novel. The ICMs (Idealised Cognitive Models) were then examined to determine the presence of isomorphism related to the production level...

Understanding social media in everyday life: Ethnomethodological and conversation analytic perspectives

Over the last decade, social media has become a hot topic for researchers of collaborative technologies (e.g., CSCW). The pervasive use of social media in our everyday lives provides a ready source of naturalistic data for researchers to empirically examine the complexities of the social world. In this talk I outline a different perspective informed by ethnomethodology and conversation analysis (EMCA) - an orientation that has been influential within CSCW, yet has only rarely been applied to social media use. EMCA approaches can complement existing perspectives through articulating how social media is embedded in everyday life, and how its social organisation is achieved by users of social media. Outlining a possible programme of research, I draw on a corpus of screen and ambient audio recordings of mobile device use to show how EMCA research can be generative for understanding social media through concepts such as adjacency pairs, sequential context, turn allocation / speaker selection, and repair. In doing so, I also raise questions about existing studies of social media use and the way they characterise interactional phenomena.

Un aporte a la comprensión de las estrategias emergentes en las organizaciones a través del concepto de improvisación y la metáfora musical

Este trabajo se inscribe en uno de los grandes campos de los estudios organizacionales: la estrategia. La perspectiva clásica en este campo promovió la idea de que proyectarse hacia el futuro implica diseñar un plan (una serie de acciones deliberadas). Avances posteriores mostraron que la estrategia podía ser comprendida de otras formas. Sin embargo, la evolución del campo privilegió en alguna medida la mirada clásica estableciendo, por ejemplo, múltiples modelos para ‘formular’ una estrategia, pero dejando en segundo lugar la manera en la que esta puede ‘emerger’. El propósito de esta investigación es, entonces, aportar al actual nivel de comprensión respecto a las estrategias emergentes en las organizaciones. Para hacerlo, se consideró un concepto opuesto —aunque complementario— al de ‘planeación’ y, de hecho, muy cercano en su naturaleza a ese tipo de estrategias: la improvisación. Dado que este se ha nutrido de valiosos aportes del mundo de la música, se acudió al saber propio de este dominio, recurriendo al uso de ‘la metáfora’ como recurso teórico para entenderlo y alcanzar el objetivo propuesto. Los resultados muestran que 1) las estrategias deliberadas y las emergentes coexisten y se complementan, 2) la improvisación está siempre presente en el contexto organizacional, 3) existe una mayor intensidad de la improvisación en el ‘como’ de la estrategia que en el ‘qué’ y, en oposición a la idea convencional al respecto, 4) se requiere cierta preparación para poder improvisar de manera adecuada.

Distributed Large-scale Mixed-Integer Optimization with Application to Energy and Multi-robot Networks

Several decision and control tasks in cyber-physical networks can be formulated as large- scale optimization problems with coupling constraints. In these "constraint-coupled" problems, each agent is associated to a local decision variable, subject to individual constraints. This thesis explores the use of primal decomposition techniques to develop tailored distributed algorithms for this challenging set-up over graphs. We first develop a distributed scheme for convex problems over random time-varying graphs with non-uniform edge probabilities. The approach is then extended to unknown cost functions estimated online. Subsequently, we consider Mixed-Integer Linear Programs (MILPs), which are of great interest in smart grid control and cooperative robotics. We propose a distributed methodological framework to compute a feasible solution to the original MILP, with guaranteed suboptimality bounds, and extend it to general nonconvex problems. Monte Carlo simulations highlight that the approach represents a substantial breakthrough with respect to the state of the art, thus representing a valuable solution for new toolboxes addressing large-scale MILPs. We then propose a distributed Benders decomposition algorithm for asynchronous unreliable networks. The framework has been then used as starting point to develop distributed methodologies for a microgrid optimal control scenario. We develop an ad-hoc distributed strategy for a stochastic set-up with renewable energy sources, and show a case study with samples generated using Generative Adversarial Networks (GANs). We then introduce a software toolbox named ChoiRbot, based on the novel Robot Operating System 2, and show how it facilitates simulations and experiments in distributed multi-robot scenarios. Finally, we consider a Pickup-and-Delivery Vehicle Routing Problem for which we design a distributed method inspired to the approach of general MILPs, and show the efficacy through simulations and experiments in ChoiRbot with ground and aerial robots.

Exploring domain-informed and physics-guided learning in image-to-image translation

Image-to-image (i2i) translation networks can generate fake images beneficial for many applications in augmented reality, computer graphics, and robotics. However, they require large scale datasets and high contextual understanding to be trained correctly. In this thesis, we propose strategies for solving these problems, improving performances of i2i translation networks by using domain- or physics-related priors. The thesis is divided into two parts. In Part I, we exploit human abstraction capabilities to identify existing relationships in images, thus defining domains that can be leveraged to improve data usage efficiency. We use additional domain-related information to train networks on web-crawled data, hallucinate scenarios unseen during training, and perform few-shot learning. In Part II, we instead rely on physics priors. First, we combine realistic physics-based rendering with generative networks to boost outputs realism and controllability. Then, we exploit naive physical guidance to drive a manifold reorganization, which allowed generating continuous conditions such as timelapses.

Hierarchical active inference for cognitive architectures

This thesis explores the methods based on the free energy principle and active inference for modelling cognition. Active inference is an emerging framework for designing intelligent agents where psychological processes are cast in terms of Bayesian inference. Here, I appeal to it to test the design of a set of cognitive architectures, via simulation. These architectures are defined in terms of generative models where an agent executes a task under the assumption that all cognitive processes aspire to the same objective: the minimization of variational free energy. Chapter 1 introduces the free energy principle and its assumptions about self-organizing systems. Chapter 2 describes how from the mechanics of self-organization can emerge a minimal form of cognition able to achieve autopoiesis. In chapter 3 I present the method of how I formalize generative models for action and perception. The architectures proposed allow providing a more biologically plausible account of more complex cognitive processing that entails deep temporal features. I then present three simulation studies that aim to show different aspects of cognition, their associated behavior and the underlying neural dynamics. In chapter 4, the first study proposes an architecture that represents the visuomotor system for the encoding of actions during action observation, understanding and imitation. In chapter 5, the generative model is extended and is lesioned to simulate brain damage and neuropsychological patterns observed in apraxic patients. In chapter 6, the third study proposes an architecture for cognitive control and the modulation of attention for action selection. At last, I argue how active inference can provide a formal account of information processing in the brain and how the adaptive capabilities of the simulated agents are a mere consequence of the architecture of the generative models. Cognitive processing, then, becomes an emergent property of the minimization of variational free energy.

On the effect of confounding in linear regression models: an approach based on the theory of quadratic forms

The main topic of this thesis is confounding in linear regression models. It arises when a relationship between an observed process, the covariate, and an outcome process, the response, is influenced by an unmeasured process, the confounder, associated with both. Consequently, the estimators for the regression coefficients of the measured covariates might be severely biased, less efficient and characterized by misleading interpretations. Confounding is an issue when the primary target of the work is the estimation of the regression parameters. The central point of the dissertation is the evaluation of the sampling properties of parameter estimators. This work aims to extend the spatial confounding framework to general structured settings and to understand the behaviour of confounding as a function of the data generating process structure parameters in several scenarios focusing on the joint covariate-confounder structure. In line with the spatial statistics literature, our purpose is to quantify the sampling properties of the regression coefficient estimators and, in turn, to identify the most prominent quantities depending on the generative mechanism impacting confounding. Once the sampling properties of the estimator conditionally on the covariate process are derived as ratios of dependent quadratic forms in Gaussian random variables, we provide an analytic expression of the marginal sampling properties of the estimator using Carlson’s R function. Additionally, we propose a representative quantity for the magnitude of confounding as a proxy of the bias, its first-order Laplace approximation. To conclude, we work under several frameworks considering spatial and temporal data with specific assumptions regarding the covariance and cross-covariance functions used to generate the processes involved. This study allows us to claim that the variability of the confounder-covariate interaction and of the covariate plays the most relevant role in determining the principal marker of the magnitude of confounding.

Integrated communication and learning towards autonomous mobile radio networks

The integration of distributed and ubiquitous intelligence has emerged over the last years as the mainspring of transformative advancements in mobile radio networks. As we approach the era of “mobile for intelligence”, next-generation wireless networks are poised to undergo significant and profound changes. Notably, the overarching challenge that lies ahead is the development and implementation of integrated communication and learning mechanisms that will enable the realization of autonomous mobile radio networks. The ultimate pursuit of eliminating human-in-the-loop constitutes an ambitious challenge, necessitating a meticulous delineation of the fundamental characteristics that artificial intelligence (AI) should possess to effectively achieve this objective. This challenge represents a paradigm shift in the design, deployment, and operation of wireless networks, where conventional, static configurations give way to dynamic, adaptive, and AI-native systems capable of self-optimization, self-sustainment, and learning. This thesis aims to provide a comprehensive exploration of the fundamental principles and practical approaches required to create autonomous mobile radio networks that seamlessly integrate communication and learning components. The first chapter of this thesis introduces the notion of Predictive Quality of Service (PQoS) and adaptive optimization and expands upon the challenge to achieve adaptable, reliable, and robust network performance in dynamic and ever-changing environments. The subsequent chapter delves into the revolutionary role of generative AI in shaping next-generation autonomous networks. This chapter emphasizes achieving trustworthy uncertainty-aware generation processes with the use of approximate Bayesian methods and aims to show how generative AI can improve generalization while reducing data communication costs. Finally, the thesis embarks on the topic of distributed learning over wireless networks. Distributed learning and its declinations, including multi-agent reinforcement learning systems and federated learning, have the potential to meet the scalability demands of modern data-driven applications, enabling efficient and collaborative model training across dynamic scenarios while ensuring data privacy and reducing communication overhead.

Regularization meets GreenAI: a new framework for image reconstruction in life sciences applications

Ill-conditioned inverse problems frequently arise in life sciences, particularly in the context of image deblurring and medical image reconstruction. These problems have been addressed through iterative variational algorithms, which regularize the reconstruction by adding prior knowledge about the problem's solution. Despite the theoretical reliability of these methods, their practical utility is constrained by the time required to converge. Recently, the advent of neural networks allowed the development of reconstruction algorithms that can compute highly accurate solutions with minimal time demands. Regrettably, it is well-known that neural networks are sensitive to unexpected noise, and the quality of their reconstructions quickly deteriorates when the input is slightly perturbed. Modern efforts to address this challenge have led to the creation of massive neural network architectures, but this approach is unsustainable from both ecological and economic standpoints. The recently introduced GreenAI paradigm argues that developing sustainable neural network models is essential for practical applications. In this thesis, we aim to bridge the gap between theory and practice by introducing a novel framework that combines the reliability of model-based iterative algorithms with the speed and accuracy of end-to-end neural networks. Additionally, we demonstrate that our framework yields results comparable to state-of-the-art methods while using relatively small, sustainable models. In the first part of this thesis, we discuss the proposed framework from a theoretical perspective. We provide an extension of classical regularization theory, applicable in scenarios where neural networks are employed to solve inverse problems, and we show there exists a trade-off between accuracy and stability. Furthermore, we demonstrate the effectiveness of our methods in common life science-related scenarios. In the second part of the thesis, we initiate an exploration extending the proposed method into the probabilistic domain. We analyze some properties of deep generative models, revealing their potential applicability in addressing ill-posed inverse problems.

Latent representations for traditional music analysis and generation

We create and study a generative model for Irish traditional music based on Variational Autoencoders and analyze the learned latent space trying to find musically significant correlations in the latent codes' distributions in order to perform musical analysis on data. We train two kinds of models: one trained on a dataset of Irish folk melodies, one trained on bars extrapolated from the melodies dataset, each one in five variations of increasing size. We conduct the following experiments: we inspect the latent space of tunes and bars in relation to key, time signature, and estimated harmonic function of bars; we search for links between tunes in a particular style (i.e. "reels'") and their positioning in latent space relative to other tunes; we compute distances between embedded bars in a tune to gain insight into the model's understanding of the similarity between bars. Finally, we show and evaluate generative examples. We find that the learned latent space does not explicitly encode musical information and is thus unusable for musical analysis of data, while generative results are generally good and not strictly dependent on the musical coherence of the model's internal representation.

Generazione di dati sintetici di risonanza magnetica cerebrale

L’applicazione degli algoritmi di Intelligenza Artificiale (AI) al settore dell’imaging medico potrebbe apportare numerosi miglioramenti alla qualità delle cure erogate ai pazienti. Tuttavia, per poterla mettere a frutto si devono ancora superare alcuni limiti legati alla necessità di grandi quantità di immagini acquisite su pazienti reali, utili nell’addestramento degli stessi algoritmi. Il principale limite è costituito dalle norme che tutelano la privacy di dati sensibili, tra cui sono incluse le immagini mediche. La generazione di grandi dataset di immagini sintetiche, ottenute con algoritmi di Deep Learning (DL), sembra essere la soluzione a questi problemi.