1731 research outputs found
Sort by
Nanodiamonds for biological applications: Synthesis by laser ablation and sensing of local magnetic environment by optical spectroscopy of NV centers
Get PDFNanodiamonds (NDs) are the subject of intense investigation for their unique physical and chemical properties. Due to high hardness, optical transparency and biocompatibility, NDs find applications in tribology, catalysis and drug delivery. When enriched with nitrogen-vacancy (NV) centers, NDs can be used in bioimaging and biosensing. While the field is progressing rapidly, a number of problems are still open. In this dissertation I have tackled two important aspects for the development of NDs as biosensors:
1) production of NDs with controlled size and properties;
2) characterization and optimization of commercial fluorescent NDs as probes of paramagnetic species.
In the first part of my thesis, I report a novel synthesis route for NDs by pulsed laser ablation (PLA) in water. PLA can directly produce diamonds on a nanoscopic scale, with potential advantages over alternative methods, like grinding of bulk crystals or detonation techniques. Specifically, I demonstrate synthesis of nanometric diamond crystals by PLA in an aqueous environment and investigate the thermodynamics of this process. Indeed, the synthesis of NDs by PLA is related to a drastic change in the thermodynamic state of the target upon laser irradiation. Fast laser-induced heating results in melting and superheating of the target, followed by a strong boiling, a process named “phase explosion”, and then by a fast cooling of the molten material in water. I provide a theoretical description of both superheated and undercooled liquids and of the mechanism of phase explosion. The investigation of the link between the metastable liquids (superheated or undercooled) and the synthesis of nanoparticles is carried out by theoretical analyses, computer simulations and comparison of our experimental data with previous literature.
In the second part of the thesis I turn to commercial NDs enriched with (NV) centers. The purpose of the investigation is to explore the use of fluorescent NDs for sensing of paramagnetic species of biological interest. To this end, I explored the effects of size and surface coating on the optical properties and sensing capabilities of fluorescent NDs.
Following a theoretical introduction to the basic properties of the NV centers and to the ground state spin dynamics of these color centers, I describe the set up used for the experimental characterization of the NDs. All NDs used in my experiments, characterized by different sizes and coatings, presented high fluorescence levels, the result of a relatively high concentration of NV center. In all NDs, I observed a fast loss in coherence due to interactions between the NV centers and with the external environment. The most striking and unexpected result concerns the dynamics of the spin-lattice relaxation time T1. Differently from previous reports, spin dynamics after polarization of NV centers could not be described by a single exponential decay, but showed a sharp signal increase that I attribute to charge dynamics and charge conversion between the negative and neutral forms of the NV center. Unexpectedly, I found that coupled charge and spin dynamics are strongly affected by paramagnetic interactions, yielding unprecedented sensitivity to subnanomolar concentrations of gadolinium, a strong paramagnetic contrast agent. The connection between relaxation dynamics and concentration of paramagnetic species can open new perspectives in biosensing and in bioimaging. As a demonstration of a practical application, I tested the sensitivity of NDs in the detection of deoxyhemoglobin, an endogenous paramagnetic species in blood
Laser diagnostics of non-equilibrium plasmas
Get PDFIn the context of the electrification of the chemical industry, this thesis sets itself the goal of studying two promising plasma reactors: a nanosecond repetitively pulsed (NRP) discharge and a microwave discharge. The NRP reactor has been investigated for the dissociation of CO2. The microwave discharge has been used for studies of vibrational energy loading into N2 , as a first step towards the non-thermal synthesis of NO for fertilizer production. The complicated system that a non-equilibrium plasma represents requires sophisticated diagnostics. Such diagnostics have to be species specific and provide spatial and time resolved information about the quantities of interest, such as temperature (vibrational and rotational), product concentration and energy deposition. Given these require ments, diagnostics based upon the use of pulsed lasers are usually employed to study systems where fast kinetics are at play. The diagnostics of choice are Laser Induced Fluorescence (LIF) and vibrational Raman scattering
The Social Gaze: social visual orienting in typical and atypical development
Get PDFSocial visual orienting in typical and atypical development
Mixture SGR: un nuovo paradigma nella gestione delle manipolazioni di dati self report
Get PDFIl comportamento di faking è un problema trasversale che riguarda la ricerca sul comportamento umano in molti contesti quali la psicologia (Hopwood, Talbert, Morey,
& Rogers, 2008), le scienze sociali (Van der Geest & Sarkodie, 1998) ed economiche (Crawford, 2003), la psichiatria (Beaber, Marston, Michelli, & Mills, 1985), la medicina forense (Gray, MacCulloch, Smith, Morris, & Snowden, 2003) e le frodi scientifiche (Marshall, 2000). Esso è un problema importante sia perché va ad alterare l’affidabilità di un test quando lo si usa per le decisioni in ambito inferenziale, sia perché i dati così raccolti perdono di validità e quindi le relazioni tra variabili misurate perdono di significato.
Diversi studi hanno dimostrato che istruendo a dovere i partecipanti a un questionario, essi sono in grado di modificare le risposte fornendo risposte più estreme
(es, Furnham, 1986; Hesketh, Griffin, & Grayson, 2004; McFarland & Ryan, 2000; Viswesvaran & Ones, 1999).
Gli effetti conosciuti del faking da un punto di vista psicometrico agiscono sulla struttura della covarianza delle scale distorte. Generalmente si è visto che i punteggi che hanno subito una manipolazione da faking, in una situazione in cui i soggetti venivano motivati, tendono ad avere una varianza più piccola e una minore affidabilità (Eysenck, Eysenck, & Shaw, 1974; Topping & O’Gorman, 1997; Ellingson, Smith, & Sackett, 2001; Hesketh et al., 2004). Allo stesso tempo però è stato osservato che quando è stato chiesto ai soggetti di mentire intenzionalmente, questo aumenta il grado di covariazione tra gli item manipolati (Ellingson, Sackett, & Hough, 1999; Galić, Jerneić, & Kovačić, 2012; Pauls & Crost, 2005; Zickar & Robie, 1999; Ziegler & Buehner, 2009). Il lavoro del capitolo 3 è stato il primo a iniziare e a essere terminato esso ci ha permesso di risollevare il problema dell’identificazione dei fakers e di definire quale riteniamo essere la motivazione principale della poca efficienza degli indici di fit individuale (PFS), cioè una non discriminazione tra i fakers e gli onesti sulla base della likelihood. Le somiglianze con altri tipi di manipolazioni contenute nella parte di review contenuta nello stesso capitolo (par. 3.2) ci hanno dato conferma, almeno a nostro avviso, che l’SGR si inserisce di diritto all’interno delle manipolazioni dei dati. Oltretutto non si limita al faking, consideriamo il faking solo uno dei possibili impieghi, basta utilizzare un adeguato modello manipolativo a seconda della specifica situazione.
Se da un lato la simulazione per individuare i fakers è stata piuttosto robusta e approfondita (par. 3.3), dall’altro la parte conclusiva riguardo lo studio della likelihood (par. 3.4), purtroppo, è stata trattata più velocemente solo attraverso esempi e la generalizzazione a tutti i questionari solo dedotta. Probabilmente sarebbe interessante approfondire quella sezione per farla diventare un robusto lavoro indipendente. Nella stesura della simulazione era già emerso il concetto della mixture, cioè che il comportamento del mentitore sia coerente all’interno del pattern di risposte e che il campione si suddivida in sinceri e fakers, ma non ulteriormente formalizzato. La situazione migliore per il mentitore è quando mente su quasi la totalità degli item, cercando di creare pattern menzogneri, ma coerenti e questo comportamento gli crea un vantaggio nel non essere identificato (figure da 3.1 a 3.4) rispetto al mentire solo a metà degli item. Rimanendo però con l’idea della non-indipendenza tra item, con l’obiettivo di innovare l’utilizzo della struttura SGR fino a quel momento utilizzata, ci siamo resi conto che un questionario di molti item è difficile da analizzare perché le combinazioni di risposte sono esponenziali al numero di item. Quindi abbiamo scelto di ricominciare il percorso analitico da due item per non perdere il nostro assunto di interdipendenza nelle risposte. La decisione di questo "passo indietro" si è mostrata vincente perché ci avrebbe successivamente permesso l’esplorazione matematica delle distribuzioni di probabilità nel caso bivariato. Purtroppo resta ancora difficile applicare le formule (cap. 4) a casi reali per la difficoltà a ipotizzare i veri modelli di manipolazione e la quantità di soggetti faker, ad esempio, per cercare di ricostruire i dati originali (formula 4.16). Nonostante i limiti empirici a livello campionario, lo studio matematico ha permesso lo studio a livello di popolazione e mostrare che gli effetti del faking possono diventare facilmente invasivi nell’esisto di una ricerca gonfiando o diminuendo correlazioni o, addirittura, invertendone il segno. In particolare, i nostri risultati si possono sovrapporre, almeno parzialmente, a quelli di Loken e Gelman (2017). Nell’articolo gli autori parlano di errore casuale di misura, il quale può modificare la dimensione dell’effetto principale, anche accentuandolo nei piccoli campioni. Siamo arrivati a conclusioni simili quasi contemporaneamente; anzi il nostro tipo di “errore” di misura è più evoluto e rappresenta un comportamento complesso, strategico e non indipendente (attraverso gli item) dei soggetti. In particolare gli autori ci servono su un piatto d’argento la giustificazione di tutte le ore di lavoro che abbiamo dedicato a questo progetto: "The situation become more complicated in problems with multiple predictors, or with nonindependent error " (pg. 585). Il faking potrebbe essere, appunto, un esempio di errore non-indipendende. Ovviamente buoni presupposti non sono garanzia di un ottimo risultato, ma sono almeno un buon punto di partenza
Towards Uncovering the True Use of Unlabeled Data in Machine Learning
Get PDFKnowing how to exploit unlabeled data is a fundamental problem in machine learning. This dissertation provides contributions in different contexts, including semi-supervised learning, positive unlabeled learning and representation learning. In particular, we ask (i) whether is possible to learn a classifier in the context of limited data, (ii) whether is possible to scale existing models for positive unlabeled learning, and (iii) whether is possible to train a deep generative model with a single minimization problem
Instabilities and dynamics of elastic rods in the presence of movable constraints
Get PDFA new trend in the mechanical design of devices for advanced technologies, such as soft robotics and micro/nano mechanics, is the exploitation of structures undergoing large deflections, in an attempt of achieving superior performances. Within this framework, non-linear modelling becomes a fundamental tool for the design of compliant structures and deformable mechanism. Two structural systems are investigated, both based on the planar elastica and subject to movable and configurational constraints. These two structures disclose unforeseen behaviours when the values of the
parameters defining the models are varied. The first structural system is an elastic rod constrained by a slowly rotating clamp, while the other end is loaded with a lumped mass weight. When this weight is lower than that
corresponding to buckling, the edge of the rod describes a closed curve, behaving as an elastica compass. Differently, when the load is higher than that of buckling, a release of elastic energy is observed, leading to a snapback of the structure, so that the rod realizes an elastica catapult. The clamp in the above described structure is replaced by a frictionless and fixed sliding sleeve in the second system considered in this thesis. The rod is subject to a sudden release from the underformed configuration, providing dynamic effects on the system. By means of the variational approach, the presence of a configurational force at the exit of the sliding sleeve is proven within the dynamical setting, extending previous results restricted to the quasi-static assumption. The configurational force is found to strongly affect the dynamics of the structure. In particular, two different behaviours are observed, in which the rod may either completely penetrate in ("injection") or be expelled from ("ejection") the sliding sleeve. In both the above problems, the theoretical predictions are corroborated through the experimental validation on physical models, which have been ad hoc invented and designed. A new insight is obtained in the design of flexible devices, paving the way to applications in soft robotics
Hydro-climatic shifts in the Alpine region under a changing climate: trends, drivers detection and scale issues
Get PDFThe impact of changing climate on the hydrological cycle in Alpine regions has attracted in the last decades a wealth of attention by the scientific community and decision makers. Indeed, the implications of changes in the intensity and in the temporal and spatial patterns of precipitation, temperature and other climatic forcing have been widely observed accompanied with an increased frequency of drought and flood events, and a general degradation of water quality and health of aquatic ecosystems. Accordingly, in the present thesis, the effect of changes in hydro-climatic variables on the hydrological cycle is investigated over a range of temporal and spatial scales. In particular, the research moves
along two main directions: 1) changes in historical time series of streamflow, precipitation and temperature, recorded in the Adige River Basin (i.e., Northeastern Italy), are analyzed with a water balance approach and compared to those of other large European river basins (i.e., Ebro and Sava) in order to quantify alterations of the main hydrological fluxes due to climate change and water uses and to disentangle their reciprocal effects; 2) a framework for evaluating the hydrological coherence of available gridded meteorological datasets, including one developed in the first part of the thesis, is introduced and tested. Regarding the first line of research,
hydro-climatic and water quality variables of some important European river basins have been analyzed in order to quantify the main alterations of streamflow and to understand the most important factors controlling them. Particular attention is drawn to the Adige River Basin (an Alpine catchment located in the North-East of Italy), for which in depth studies, data measures and analyses have been performed. At this purpose, advanced techniques, besides novel approaches, have been applied. In particular, statistical methods (i.e., Mann-Kendall trend tests, Sen’s slope estimates, multivariate data analyses and Kriging algorithms) have been used to assess the water budgets and the variations in time and space of the aforementioned variables. Disentangling climatic and human impacts on the hydrological fluxes is a difficult task and it has not been fully explored yet, since concurring drivers of hydrological alterations (e.g., climate and land use changes, hydropower and agricultural developments and increasing population) are intimately intertwined one to each other and
combined in a complex nonlinear manner. At this purpose, spatial and temporal patterns of change in the hydrological cycle of the Adige River Basin have been identified by comparing annual and seasonal water budgets performed in four representative sub-basins (sized from 207 to 9,852 km2) characterized by different climatic and water uses conditions. A
significant downward trend of streamflow is found in the lower part of the Adige since the ’70s , which can be attributed to the intense development of irrigated agriculture in the drainage area of the Noce River (one of the main tributaries of the Adige River). Conversely, headwater catchments showed a significant positive trend in streamflow due to a shift in the seasonal distribution of precipitation. These results suggest that climate
change is the main driver only in headwater basins, while water uses overcome its effect along the main stream and the lower end of the tributaries. Therefore, a comparative analysis of recent trends in hydro-climatic parameters in three climatologically different European watersheds (i.e., the Adige, Ebro and Sava River Basins) has been performed. The main
results suggest that the highest risk of increasing water scarcity refers to the Ebro, whereas the Adige shows better resilience to a changing climate. In the second part, this thesis deals with the uncertainty associated with climate datasets, that typically represents the largest part of the total uncertainty in hydrological modeling and, more in general, in climate change impact studies. In particular, this thesis describes a new framework for assessing the coherence of gridded meteorological datasets with streamflow observations (i.e., HyCoT - Hydrological Coherence Test). Application to the Adige catchment reveals that using inverse hydrological modeling allows testing the accuracy of gridded temperature and precipitation datasets and it may represent a tool for excluding those that are inconsistent with the hydrological response
Economic Growth and Public Debt: Beyond Debt-Thresholds. Theoretical and Empirical Issues.
Get PDFThe idea that public debt may represent a burden for the economic system as a whole has distant origins and focuses on who and how should pay for debt, and with what consequences on the economy. Nevertheless, particularly influential both for academic research and the implementation of the fiscal corrective policies was the empirical paper proposed by Reinhart and Rogoff in 2010 at the dawn of the crisis. Reinhart and Rogoff (2010), in a large panel of countries, identified a critical threshold of 90% of the debt-to-GDP ratio beyond which debt is harmful to growth. Several countries in the world were fast approaching that threshold or already were well beyond it. Though Reinhart and Rogoff’s work was affected by many flaws, it has spurred buoyant empirical research in search of the general debt thresholds above which growth is jeopardised by public debt. Further works have supported the existence of critical debt-to-GDP ratios under various time and space observational fields, but results of these researches are inconclusive or controversial, as discussed in Chapter 2. Country-specific characteristics and contingencies play in fact a prominent role, thus prompting a branch of literature that attempts to comprehensively understand the debt-growth relationship and its determinants (see for instance Panizza and Presbitero, 2014; Eberhardt and Presbitero, 2015). In contrast with the findings of the broad threshold literature and of many theoretical models, the idea that public debt is always harmful to economic growth has partially been reconsidered in the last few years. Nevertheless, the existence of a linkage between debt and growth has not been rejected: the long-run relationship between such macroeconomic variables is inevitably and broadly affected by heterogeneous factors. However, in retrospect and as emerges in Chapter 1, one may say that the empirical pursuit of the debt-to-GDP threshold harmful to growth lacks deeper foundational work: why should we expect a negative public debt-growth relationship? In addition, if such a relationship exists, why should it take the specific form of a threshold of the debt-to-GDP ratio, and why should we expect this threshold to be equally valid across time and space? These questions are the starting point of this Doctoral Thesis, which is organised as follows. Chapter 1 surveys the theoretical literature concerning public debt and economic growth, aiming at finding a theoretical foundation for the debt-threshold literature. Overall, there is no clear and straightforward answer to the questions of why we should expect a negative public debt-growth relationship in the first place, why it should take the specific form of a threshold of the debt-to-GDP ratio, and why we should expect this threshold to be equally valid across time and space. Or, from another perspective, there are many possible answers and many elements affecting them, thus reflecting the complexity of the argument, as well as the variety of the empirical situations. In particular, the literature that I examine, on the one hand offers a rich variety of explanations and insights to researchers of the debt-growth relationship but, on the other, it does not provide any one-way conclusion: the relationship may be negative, positive, or even no relationship may exist, both from a theoretical and an empirical point of view. Even less is theoretically founded the existence of a general debt-to-GDP threshold above which growth is consistently stifled. Each country’s specific characteristics, circumstances, and events have an overwhelming importance that cannot be encapsulated in a single general law. In Chapter 1, I also present a fiscal model of endogenous growth that may help address the theoretical issues in an orderly and consistent manner along two specific coordinates of debt assessment: sustainability/unsustainability, and efficiency/inefficiency. The thrust of the model is that no meaningful assessment of debt and its effect on growth at any point in time is possible without reference to the whole debt trajectory and the specific state of the economy along the trajectory. Chapter 2 reviews the empirical literature and focuses on the debt-growth relationship from an econometric point of view. As before, it is difficult to derive a univocal conclusion on the nature of such a relationship on the basis of the literature’s findings: the existence of a significant negative relationship between debt and growth is the predominant thinking, though in contrast with the conclusions of several works. For these reasons, the aim of Chapter 2 is to go to the roots of the debt-growth relationship, to investigate whether the outstanding debt and the GDP are linked. To this end, I have adopted a research methodology that differs from the most common employed in the literature on debt-to-GDP thresholds. First, my analysis does not hinge on any specific theory, and it should not be considered as a proof of a specific theoretical statement. Rather, it is based on the approach outlined by Hoover et al. (2008) and aims at understanding "what the data say" without imposing aprioristic theoretical structures. A second methodological choice consistent with this approach is to treat the (growth of the) amount of public debt and (the growth of) GDP as the two genuine primitives, without imposing the debt-to-GDP ratio as a primitive itself. In fact, for this to be possible, the two underlying primitives should display well defend statistical properties, namely cointegration and convergence towards a long-term equilibrium value, which are usually not tested in the literature. Third, I believe that the heterogeneity, or non-generality, of results that I have pointed out before should be taken as an intrinsic feature of the problem at hand, so that a viable strategy is to restrict, rather than expand, the observational field. I have set time and space limits to my dataset by purpose: my analysis is based on a panel dataset including quarterly data for 25 Eastern and Western European countries from 1999Q1 to 2015Q4. The Eurozone represents a unique "field experiment" of a large number of countries where some key conditioning factors of fiscal policy are common and exogenous, namely fiscal targets and rules, monetary policy, and the exchange rate with the rest of the world. The main result is that a long-run equilibrium relationship between GDP and debt exists for some countries ― and debt and GDP tend to adjust towards it ― but it is not generalisable. Where a relationship exists, it does not always imply that the debt-to-GDP ratio may be the appropriate variable for describing it. Moreover, cross-country heterogeneity and the role of the financial crisis and of the austerity periods remain substantial and overwhelming factors. Therefore, a unique equation describing the GDP-debt relationship does not seem to exist, which entails the impossibility to derive a meaningful general debt-to-GDP threshold. Thus far I have focused on the general relationship between debt and growth from both the theoretical and the empirical points of view. Turning to the analysis of the Sovereign Debt Crisis and of the austerity period, Chapter 3 attempts to explain what has driven austerity ― measured as the first difference of the cyclically adjusted structural primary balance ― within a dataset of 28 European countries. In the first part of this chapter I present a correlation analysis that describes the relationship between the variable austerity and each of the considered determinants, that are brought back to four main sets of variables: fiscal discipline, market discipline, fiscal consolidation, and macroeconomic stabilisation. The second part implements a panel econometric analysis based on the principal component factor analysis and on the pooled partial common correlation effect estimator. Results show that the variables and factors of the analysis are not able to fully explain austerity, though an important contribution is provided by the enforcement of the Eurozone fiscal rules (the adoption of excessive deficit procedures) and is partially counterbalanced by the cyclical position of the economy. The last chapter, Chapter 4, aims at gaining insight into the role of debt and government expectations and their impact on growth under uncertainty conditions. In fact, it is possible that the effects of austerity measures in some countries, for instance the so-called PIIGS, were amplified by uncertainty. My ambition is to relate austerity with consumers’ expectations, thus studying whether and when consumers’ beliefs about public debt and government intervention affect their consumption, savings, and tax compliance choices with a direct impact, at the aggregate level, on economic growth. Therefore, Chapter 4 implements a laboratory experiment to study how people react in a generalized framework in which public debt may be unexpectedly reduced. The debt dynamics arises endogenously: within a public good game, taxes are collected from all participants and are used to cover a given level of public expenditure, which is then equally distributed to the same participants at the beginning of each round. If the collected amount of taxes is lower than what the public expenditure would require, a deficit is generated. Moreover, reproducing a forced withdrawal, the outstanding amount of public debt can be reduced upon accessing subjects’ savings. Within this setting, expectations are directly elicited by asking subjects if they believe that public debt is going to be reduced, and if they think that the other subjects believe that public debt is sustainable. Therefore, it is possible to identify whether and how agents’ allocations and expectations are affected by the public debt path. As mentioned above, a peculiarity of my approach is the endogenous dynamics of public debt: not only it avoids introducing predetermined dynamics, but also increases the ecological validity of the experiment. Participants are indeed more psychologically involved in the debt mechanism and they might feel responsible for the raise in debt. On the other hand, an exogenous dynamics could depict public debt and tax compliance as irrelevant. Results show that this experimental framework is characterized by relatively high and often increasing aggregate savings and relatively low and decreasing aggregate consumption. Interestingly, an increase in the debt-reduction expectations and a decrease in the perceived debt sustainability are also found to explain savings and consumption behaviours, as is shown in the econometric part of Chapter 4
Non-farm Entrepreneurial Activities and the role of Non-cognitive Skills in Agriculture. Theoretical framework and Empirical Evidence from Ethiopia.
Get PDFThe dissertation is articulated in three empirical applications which focus on the economic development of Ethiopian smallholders either on and off-farm from 2012 to 2016. The complete title of the thesis is “Non-farm Entrepreneurial Activities and the Role of Non-cognitive Skills in Agriculture. Theoretical Framework and Empirical Evidence from Ethiopia”. In fact, one of the earliest and most central insights of the literature on economic development is that development entails structural change. The countries that manage to pull out of poverty and get richer are those that are able to diversify away from agriculture and other traditional products. As labor and other resources move from agriculture into modern economic activities, overall productivity rises and incomes expand. However, the enhancing technologies developed by modern economic sectors may be reinvested on farm increasing the agricultural productivity too. The production on the farm side is analyzed in the first two papers of the thesis, while the third one focuses on the non-farm sector.
The first two papers share the same data set which is a cross-section survey collected in two rural regions in Ethiopia during the crop year 2012. It was financed by the Food and Agriculture Organization (FAO), with the collaboration of the CEIS of the University of Torvergata and implemented by the University of Addis Ababa. The survey collected also non-cognitive skills of the 501 households in the survey using two well-known psychometric tests: the Big Five Inventory test and the Emotion Regulation questionnaire. These traits were used for the empirical estimation of the agricultural production function and the agricultural input equations in order to see how these traits affect productive and allocative efficiency in rural Ethiopia. Results show a statistical relevance of personality traits in affecting both production and input choice/application. Furthermore, when non-cognitive skills variables are explicated in the models is possible to confirm the “recursive structure” between output and inputs usage proposed by Zellner et al (1966). Namely, the use of the OLS technique for the production function estimation is (weakly) validated thanks to the explication personality traits variables in the model generally omitted in the neoclassical model specifications.
The third paper analyses whether the economic activities off-farm in Ethiopia are the results of push or pull factors. We wanted to assess if households were attracted in these activities because of the remunerative opportunities or if they were pushed due to ex ante coping mechanisms. For the empirical application, we rely on longitudinal panel data “Living Standard Measurement Study- Integrated Survey on Agriculture” collected by World Bank (WB). We use the three rounds available (2011/2012, 2013/2014, and 2015/2016) to track households overtime. Furthermore, the empirical strategy relies on different kind of econometric techniques: cluster analysis, multivariate regressions and Heckman correction models. Then, we focused on the impact of household’s non-farm engagement on households’ wellbeing (consumption and agricultural input expenditure). Results show that push factors are the main drivers of households’ participation in the non-farm sector, and without further investments in infrastructure and education these activities are not likely to be the engine for further economic development in Ethiopia
Online Adaptive Neural Machine Translation: from single- to multi-domain scenarios
No full textIn this thesis we investigate methods for deploying machine translation (MT) in real-world application scenarios related to the use of MT in computer assisted translation (CAT), where human translators post-edit MT outputs. In particular, we investigate (in chronological order) MT adaptation under two working conditions: single-domain and multi-domain. In the former, we assume that MT receives requests by a single user working on a single domain, while in the latter we assume the MT system to receive requests i) from multiple users working on different domains, ii) with no predefined order, and iii) without domain information. In the single-domain case, we first focus on word alignment, a core component of online adaptive phrase-based MT (PBMT) that is crucial for extracting features from a post-edited segment. In particular, we concentrate on improving word alignment in presence of out-of-vocabulary words observed in the source sentences or introduced by the post-editor. In the multi-domain scenario we turned our focus to the neural MT (NMT) paradigm. In particular, we introduce a scalable solution that adapts on-the-fly a generic NMT model to each incoming translation request. It relies on a procedure that locally fine-tunes the model to each input sentence using samples retrieved from a pool of parallel data. Our instance-based adaptation uses a more general formulation of the log-likelihood approach to control the contribution of relevant and irrelevant words during model update. Finally, we test our approach on a simulated continuous learning setting, where the system receives user feedback under form of post-editing