245 research outputs found
Fast algorithms for Bayesian variable selection
No full textVariable selection of regression and classification models is an important but challenging problem. There are generally two approaches, one based on penalized likelihood, and the other based on Bayesian framework. We focus on the Bayesian framework in which a hierarchical prior is imposed on all unknown parameters including the unknown variable set. The Bayesian approach has many advantages, for example, we can access unknown obtain the posterior distribution of the sub-models. And more accurate prediction may be obtained by model averaging.
However, as the posterior distribution of the model parameters is usually not in closed form, posterior inference that relies on Markov Chain Monte Carlo (MCMC) has high computational cost especially in high-dimensional settings, which makes Bayesian approaches undesirable. In order to deal with datasets with large number of features, we aim to develop fast algorithms for Bayesian variable selection, which approximate the true posterior distribution, but yet still return the right inference (at least asymptotically).
In this thesis, we start with a variational algorithm for linear regression. Our algorithm is based on the work by Carbonetto and Stephens (2012), and with essential modifications including updating scheme and truncation of posterior inclusion probabilities. We have shown that our algorithm achieves both frequentist and Bayesian variable selection consistency.
Then we extend our variational algorithm to logistic regression by incorporating the Polya-Gamma data-augmentation trick (Polson et al., 2013), which links our algorithm for linear regression with logistic regression. However, as the variational algorithm needs to update the variational distribution of all the latent Polya-Gamma random variables of the same size of the observations at every iteration, this algorithm is slow when there are huge amount of observations, or even be infeasible when the data is too large to be loaded into computer memory. We propose an online algorithm for the logistic regression, under the framework of online convex optimization. Our algorithm is fast, and achieves similar accuracy (log-loss) as the state-of-art algorithm (Follow-the-Regularized-Proximal algorithm).Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2019-08-01The student, Xichen Huang, accepted the attached license on 2017-07-07 at 15:07.The student, Xichen Huang, submitted this Dissertation for approval on 2017-07-07 at 15:23.This Dissertation was approved for publication on 2017-07-10 at 12:40.DSpace SAF Submission Ingestion Package generated from Vireo submission #11339 on 2017-09-29 at 10:46:46Made available in DSpace on 2017-09-29T17:45:38Z (GMT). No. of bitstreams: 2
HUANG-DISSERTATION-2017.pdf: 610142 bytes, checksum: 132966a902cf66c70d837edb312274ec (MD5)
LICENSE.txt: 4209 bytes, checksum: 41e50f48224de044e3c2f39d5bf69ddf (MD5)
Previous issue date: 2017-07-10Embargo set by: Colleen Fallaw for item 103486
Lift date: 2019-09-29T17:48:06Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemEmbargo set by: Seth Robbins for item 103486
Lift date: 2020-03-02T19:56:41Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemEmbargo set by: Seth Robbins for item 103486
Lift date: 2020-03-02T19:59:52Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemEmbargo set by: Seth Robbins for item 103486
Lift date: 2020-03-02T20:02:46Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemU of I Only Restriction Lifted for Item 103486 on 2020-03-03T10:15:29Z
Recommended from our members
Daoism in the Twentieth Century: Between Eternity and Modernity
No full textIn Daoism in the Twentieth Century, an interdisciplinary group of scholars explores the social history and anthropology of Daoism from the late nineteenth century to the present, focusing on the evolution of traditional forms of practice and community, as well as modern reforms and reinventions both within China and on the global stage. Essays investigate ritual specialists, body cultivation and meditation traditions, monasticism, new religious movements, state-sponsored institutionalization, and transnational networks.“This pioneering work not only explores the ways in which Daoism was able to adapt and reinvent itself during China’s modern era, but sheds new light on how Daoism helped structure the development of Chinese religious culture. The authors also demonstrate Daoism’s role as a world religion, particularly in terms of emigration and identity. The book’s sophisticated approach transcends previous debates over how to define the term ‘Daoism,’ and should help inspire a new wave of research on Chinese religious movements.” Paul R. Katz, Academia Sinica, TaiwanContributors: Kenneth Dean, Fan Guangchun, Vincent Goossaert, Adeline Herrou, Lai Chi-tim, Lee Fongmao, Xun Liu, Lü Xichen, David A. Palmer, Kristofer Schipper, Elijah Siegler, Yang Der-rueySeries: New Perspectives on Chinese Culture and Society, vol.
Magnetic sensing supported by machine learning
No full textThis dissertation explores how magnetic sensing can be advanced by integrating machine learning (ML) with magnetically responsive materials. Grounded in the concept of bioinspired adaptation, the core approach leverages soft ferromagnetic assemblies, magnetoelectric interfaces, and mouldable magnetic soft composites as reconfigurable platforms that readily adjust sensing sensitivity and functionality under varying stimuli. Coupled with ML methods—from basic classification to optimization strategies—these magnetically driven sensors transcend conventional static designs, enabling tasks such as shape detection, adaptive mechanosensing, secure information encoding, and magnetic field measurement.
The research progresses from trainable, bioinspired sensor concepts to increasingly integrated systems that demonstrate broader functionality and higher autonomy along this spectrum of applications. Underpinning the whole work is the concept of magnetic fields serving not only as a stimulus but also as a tunable “control knob” to reconfigure material properties. Machine learning methods then classify complex patterns, interpret sensor data, and enhance sensing resolution by addressing performance trade-offs. Although the main emphasis is on magnetics, a final demonstration of pressure-based handwriting recognition illustrates the broader applicability of integrating advanced material engineering with ML. The result is a cohesive framework where ML augments magnetic sensing systems toward enhanced adaptability, robustness, and intelligence.
By seamlessly uniting magnetic field manipulation with data-driven algorithms, this dissertation proposes a framework for developing advanced sensing devices in applications ranging from soft robotics and biomedical diagnostics to secure communication and wearable electronics. Beyond individual device advancements, the work underscores the broader potential of cross-disciplinary research—where merging materials science, magnetics, and ML can catalyse transformative innovations in sensor design and functionality
The Appreciation and Analysis of two Pound’s Poems—The ‘Congruous’ Transfer From the Perspective of Readers’ Cognition
No full textLiu Ch'e and Separation on the river Kiang are both translated by Ezra Pound, respectively based on the poem written by Liu Che, the Wu Emperor of Western Han, and the poem created by Li Po, the famous poet of Tang Dynasty, From the perspective of readers' cognition, Pound achieved the vivid transfer of images in his translated poems through the employment of his Imagist aesthetic standard and unique translation strategy
Exploration of the Motion Graphics Educational tools based on the animations ‘Weddings’
No full textWith the development of science and technology, computer networks and smart devices are widely used in our life and strongly driving social growth, including the field of education. Cities are becoming bigger and bigger. People are getting busier in modern times. People have lots of free but short time slots every day, but attending traditional classes is limited by the factors of geographical location, distance, time, etc. Thus the demand for on-line meetings and remote learning is sharply increasing. People are seeking a convenient and effective method to gain knowledge or information with their fragment time. This thesis is under such a background and focuses on the exploration of the Motion Graphics Educational tools based on the animations ‘Weddings’.
This thesis project presents an applicable design solution to facilitate extensive online educational classes. The author designed a series of animations of weddings in two specific times as an Exploratorium attempt via motion graphics from four aspects, audio, time, graphics, and motion effects. This thesis describes the process discussion on the topics of design and the benefits based on these four factors. For further accessible cultural communication, the author set up a website of weddings to work as a public channel to display the animations to the audience. On purpose for better understand the benefits of motion graphic educational tools better and functional realization effectiveness of this project, the author conducted two surveys and analyzed the responses and confirmed the educational tools do work to convey the expected information to the audience.
This thesis project demonstrates how motion graphics are implicated as a useful educational tool based on the animations ‘Weddings.’ It creates a more learner-friendly and accessible manner for the audience to gain knowledge or information conveniently and efficiently
Correlation study on firing temperature and color of plain pottery excavated from the Tang Dynasty tomb of Liu Jing in Shaanxi, China
No full textAbstract Plain pottery excavated from the Tang Dynasty tomb of Liu Jing was taken as the research object. The color, chemical composition, microstructure, and phase were tested to investigate the influencing factors of color for plain pottery fragments. The results indicated that the contents of Fe2O3 and TiO2 in all fragments varied little, and the influence of humic acids in clay as well as the firing atmosphere on the appearance color of plain pottery was excluded. Therefore, the main factor affecting color saturation (C*) was identified as the firing temperature (T). More importantly, the correlation between C* and firing temperature was established by replicas fired at different temperatures. Before the appearance of the glass phase, iron-containing minerals played a major role in coloring, and after that, iron ions in the glass phase and iron crystallization rose the important function of coloring. Consequently, with the increase of firing temperature, C* value increased firstly and then decreased. The inflection point of the fitted C* − T curve corresponded to the glass phase formation temperature. By comparing the estimated firing temperatures obtained by the fitted C* − T correlation curve with the known firing temperature of replicas, it was demonstrated that the color measurement is an ideal method for deducing the firing temperatures of ancient plain pottery
An Optimization Model Applied to Active Solar Energy System for Buildings in Cold Plateau Area
No full textAbstractThe large-scale utilization of solar energy in buildings is one of the most promising technologies to solve the global energy shortage problem and reduce the carbon dioxide emissions. The present paper has proposed an optimization model coupled with solar thermal and photovoltaic systems. Optimization results of active solar energy system from the energy saving view and economical view have been obtained for typical hotel and office buildings in cold plateau area, respectively
The Coupling of Strain and Lithium Diffusion: A Theoretical Model Based on First-Principles Calculations
No full textMost electrodes undergo volume changes in lithium-ion batteries, and it turns out that the volume changes also significantly affect lithium diffusion kinetics based on first-principles calculations. To study the mechano-electro-chemical coupling, a theoretical model of spherical electrode has been developed and the effect of strain on diffusion coefficient is explicitly considered. The results show that strain-enhanced diffusion greatly alleviates diffusion-induced stresses, and the lithiation and delithiation are extremely asymmetric in lithium concentration distribution and stresses due to asymmetrical changes of diffusion coefficient with time and positions along the radial direction. (C) 2015 The Electrochemical Society.National Natural Science Foundation of China [11172231, 11372241]; ARPA-E [DE-AR0000396]; AFOSR [FA9550-12-1-0159]SCI(E)[email protected]; [email protected]
- …
