811 research outputs found
Multilabel Classification through Structured Output Learning - Methods and Applications
No full textMultilabel classification is an important topic in machine learning that arises naturally from many real world applications. For example, in document classification, a research article can be categorized as “science”, “drug discovery” and “genomics” at the same time. The goal of multilabel classification is to reliably predict multiple outputs for a given input. As multiple interdependent labels can be “on” and “off” simultaneously, the central problem in multilabel classification is how to best exploit the correlation between labels to make accurate predictions. Compared to the previous flat multilabel classification approaches which treat multiple labels as a flat vector, structured output learning relies on an output graph connecting multiple labels to model the correlation between labels in a comprehensive manner. The main question studied in this thesis is how to tackle multilabel classification through structured output learning.
This thesis starts with an extensive review on the topic of classification learning including both single-label and multilabel classification. The first problem we address is how to solve the multilabel classification problem when the output graph is observed apriori. We discuss several well-established structured output learning algorithms and study the network response prediction problem within the context of social network analysis. As the current structured output learning algorithms rely on the output graph to exploit the dependency between labels, the second problem we address is how to use structured output learning when the output graph is not known. Specifically, we examine the potential of learning on a set of random output graphs when the “real” one is hidden. This problem is relevant as in most multilabel classification problems there does not exist any output graph that reveals the dependency between labels. The third problem we address is how to analyze the proposed learning algorithms in a theoretical manner. Specifically, we want to explain the intuition behind the proposed models and to study the generalization error.
The main contributions of this thesis are several new learning algorithms that widen the applicability of structured output learning. For the problem with an observed output graph, the proposed algorithm “SPIN” is able to predict an optimal directed acyclic graph from an observed underlying network that best responses to an input. For general multilabel classification problems without any known output graph, we proposed several learning algorithms that combine a set of structured output learners built on random output graphs. In addition, we develop a joint learning and inference framework which is based on max-margin learning over a random sample of spanning trees. The theoretic analysis also guarantees the generalization error of the proposed methods
Education and earnings inequality in Mexico
No full textEducation attainment levels increased dramatically for Mexico's labor force in the 1980s and early 1990s. In parallel, the country experienced a pronounced increase in earnings inequality from 1984-94, reflected in a higher dispersion of wages and an absolute decline in the real incomes of less educated, poorer Mexicans. This increased wage dispersion presents policymakers with a tradeoff between efficiency considerations (favoring increased spending on higher education) and equity considerations (favoring a more equal distribution of per student spending) in the allocation of fiscal resources to education. The author concludes that the best way to deal with this equity-efficiency tradeoff is to encourage greater private participation in higher education. His main findings are that: a) The accumulation of human capital during 1984-94, as proxied by education attainment, was accompanied by a more equal distribution of education attainment levels over that period and, thus, exerted an equalizing effect on the distribution of incomes. The increased income inequalityobserved over that period appears to be caused by an increased rate of skill-based technological change, whose transmission to Mexico and other developing countries may have been facilitated by the increased openness of their economies. b) The greater dispersion of wager observed in Mexico during the past decade raised the rates of return on investing in higher education, reversing the traditional pattern where primary education exhibits the highest rates of return. c) The social rates of return across levels of schooling were more uniform in 1994 than in 1984, suggesting a more efficient assignment of education spending. At the same time, the distribution of spending on education became more egalitarian, as per student spending in higher education declined markedly compared with per student spending at the primary level. This surprising coincidence in the pattern of spending on education was only possible because Mexico started out with a very distorted resource allocation in education that was both highly inequitable and inefficient. As Mexico's policymakers are on the way to correcting these distortions, the opportunities for avoiding the equity-efficiency tradeoff within Mexico's centralized education framework will become progressively exhausted. d) There is little reason to expect the pace of technological change, which appears mainly responsible for raising wage dispersion and the relative returns on higher education, to abate. Efficiency considerations dictate that Mexico should respond by devoting more resources to higher education. However, the federal budget, which traditionally has financed the lion's share of higher education costs in Mexico, is unable to accommodate additional spending on higher education, while spending cuts elsewhere in the education sector are bound to raise serious equity questions. Thus, to avoid falling behind in terms of human capital accumulation, greater private sector participation is necessary, at least, in terms of cost recovery from the main beneficiaries of higher education.Decentralization,Teaching and Learning,Environmental Economics&Policies,Public Health Promotion,Curriculum&Instruction,Teaching and Learning,Environmental Economics&Policies,Health Monitoring&Evaluation,Gender and Education,Curriculum&Instruction
Representation learning of natural language and its application to language understanding and generation
No full textHow to properly represent language is a crucial and fundamental problem in Natural Language Processing (NLP). Language representation learning aims to encode rich information such as the syntax and semantics of the language into dense vectors. It facilitates the modeling, manipulation and analysis of natural language in computational linguistics. Existing algorithms utilize corpus statistics such as word co-occurrences to learn general-purpose language representation. Recent advances in generic representation integrate intensive information such as contextualized features from unlabeled text corpora.
In this dissertation, we continue this line of research to incorporate rich knowledge into generic embeddings. We show that word representation could be enriched with various information including temporal and spatial variations as well as syntactic functionalities, and that text representation could be refined with topical knowledge. Moreover, we develop an insight into the geometry of pre-trained representation, and connect it to the semantic understanding such as identifying the idiomatic word usage.
Besides generic representation, task-dependent representation is also extensively studied in downstream applications, where the representation is trained to encode domain information from labeled datasets. This dissertation leverages the capability of neural network models to integrate the task-specific supervision into language representations. We introduce new deep learning models and algorithms to train representations with external knowledge in annotated data. It is shown that the learned representation can assist in various downstream tasks in language understanding such as text classification and language generation such as text style transfer.Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2022-05-01The student, Hongyu Gong, accepted the attached license on 2020-04-14 at 14:10.The student, Hongyu Gong, submitted this Dissertation for approval on 2020-04-14 at 14:33.This Dissertation was approved for publication on 2020-04-15 at 11:07.DSpace SAF Submission Ingestion Package generated from Vireo submission #14978 on 2020-08-25 at 17:27:37Made available in DSpace on 2020-08-26T23:54:33Z (GMT). No. of bitstreams: 3
GONG-DISSERTATION-2020.pdf: 4385148 bytes, checksum: d8a85c86ec73d08dab4ac2ed3f3dea25 (MD5)
LICENSE.txt: 4208 bytes, checksum: 74c7fbb182d7130e64419d4e044b31f8 (MD5)
PROQUEST_LICENSE.txt: 4554 bytes, checksum: 87e4b6e6fd78a49c5bb719f0950b7222 (MD5)
Previous issue date: 2020-04-15Embargo set by: Seth Robbins for item 115720
Lift date: 2022-08-26T23:54:40Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemEmbargo set by: Seth Robbins for item 115720
Lift date: 2022-08-26T23:55:59Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemEmbargo set by: Seth Robbins for item 115720
Lift date: 2022-08-26T23:57:28Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemEmbargo set by: Seth Robbins for item 115720
Lift date: 2022-08-26T23:58:55Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemAuthor requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemU of I Onl
The Comparative Life Cycle Assessment of Structural Retrofit Techniques
No full textabstract: The current study conducts a comparative LCA of two alternative structural retrofit/ strengthening techniques - steel jacketing, and the carbon fiber reinforced polymer (CFRP) retrofit. A cradle-to-gate system boundary is used for both techniques. The results indicated that the CFRP retrofit technique has merits over the conventional steel jacketing in all three impact categories covered by this study. This is primarily attribute to the much less material consumption for CFRP retrofit as compared to steel jacketing for achieving the same load carrying capability of the retrofitted bridge structures. Even though the transoceanic transportation of carbon fiber has been taken into account in this study, the energy consumption and environmental impacts of CFRP transportation is still much smaller than steel due to it light weight property. The impacts of CFRP retrofit are mainly focused in the material manufacturing phase, which implies that the improvements in the carbon fiber manufacturing technology could potentially further reduce the environmental impacts of CFRP retrofit
Sources of tropospheric ozone along the Asian Pacific Rim : an analysis of ozonesonde observations
No full textAuthor name used in this publication: Hongyu LiuAuthor name used in this publication: Lo Yin ChanAuthor name used in this publication: Samuel J. OltmansAuthor name used in this publication: Joyce M. Harris2002-2003 > Academic research: refereed > Publication in refereed journalVersion of RecordPublishedVoR allowe
Buckling of Bulk Structures With Finite Prebuckling Deformation
No full textThe prebuckling deformation of structures is neglected in most of the conventional buckling theory (CBT) and numerical method (CNM), because it is usually very small in conventional concepts. In the preceding paper (Su et al., 2019), we found a class of structures from the emerging field of stretchable electronics, of which the prebuckling deformation became large and essential for determining the critical buckling load, and developed a systematic buckling theory for 3D beams considering the effects of finite prebuckling deformation (FPD). For bulk structures that appear vastly in the advanced structures, a few buckling theories consider the effects of the prebuckling deformation in constitutive equations by energy method, which are significantly important but not straightforward and universal enough. In this paper, a systematic and straightforward theory for the FPD buckling of bulk structures is developed with the use of two constitutive models. The variables for the prebuckling deformation serve as the coefficients of the incremental displacements, deformation components, and stress in the buckling analysis. Four methods, including the CBT, CNM, DLU (disturbing-loading-unloading method) method and FPD buckling theory, are applied to the classic problems, including buckling of an elastic semi-plane solid and buckling of an elastic rectangular solid, respectively. Compared with the accurate buckling load from the DLU method, the FPD buckling theory is able to give a good prediction, while the CBT and CNM may yield unacceptable results (with 70% error for the buckling of an elastic semi-plane solid)
Investment planning under uncertainty in energy systems: Modelling and algorithms
No full textThis thesis applies operational research methods for the investment planning of energy systems under uncertainty for the energy transition. We develop new models and solution methods.
On the modelling side, we first focus on modelling hydrogen-based offshore energy hubs in an offshore energy system. A mixed-integer linear program is developed for the investment planning of offshore energy systems with offshore energy hubs. The model is then extended to (1) include uncertainty using a multi-horizon stochastic programming approach and (2) include the European onshore and offshore energy systems. Finally, some major extensions are made to the model, which leads to the REORIENT model. The REORIENT model is a multi-horizon mixed-integer linear stochastic program for integrated investment, retrofit, and abandonment planning of energy systems under short-term and long-term uncertainty. This is the first model that integrates different alternatives and investigates the role of existing energy infrastructure in the energy transition. The REORIENT model features the main modelling contributions in this thesis. In addition, we also extend the modelling of an existing model, EMPIRE, which is a stochastic linear program for the European power system investment planning, by modelling the heat and industry sectors with a strong focus on endogenous decisions regarding industry decarbonisation, hydrogen and carbon capture and storage.
On the methodology side, we develop algorithms that exploit the structure of multi-horizon stochastic programming. The algorithms developed can also be applied in general multi-stage stochastic programs. We develop enhanced Benders decomposition and Lagrangean decomposition algorithms. The enhanced Benders decomposition utilises adaptive oracles. We also propose to stabilise the adaptive Benders decomposition with (1) a novel dynamic level method and (2) a novel centre point strategy. Also, we propose parallelised Lagrangean decomposition with primal reduction. The scenario subproblems are solved in parallel, and the primal problem is reduced based on the structure of multi-horizon stochastic programming and solved in parallel. We apply the proposed algorithms to solve the REORIENT model and its variations and compare them with standard Benders, unstabilised adaptive Benders, and standard Lagrangean decomposition.
The proposed models and algorithms contribute to operational research and provide useful insights for the energy transition
- …
