DissertationsCopyright (c) 2014 Carnegie Mellon University All rights reserved.
http://repository.cmu.edu/dissertations
Recent documents in Dissertationsen-usThu, 18 Sep 2014 01:44:26 PDT3600Nonparametric Discovery of Human Behavior Patterns from Multimodal Data
http://repository.cmu.edu/dissertations/359
http://repository.cmu.edu/dissertations/359Tue, 16 Sep 2014 07:44:57 PDT
Recent advances in sensor technologies and the growing interest in context- aware applications, such as targeted advertising and location-based services, have led to a demand for understanding human behavior patterns from sensor data. People engage in routine behaviors. Automatic routine discovery goes beyond low-level activity recognition such as sitting or standing and analyzes human behaviors at a higher level (e.g., commuting to work). The goal of the research presented in this thesis is to automatically discover high-level semantic human routines from low-level sensor streams. One recent line of research is to mine human routines from sensor data using parametric topic models. The main shortcoming of parametric models is that they assume a fixed, pre-specified parameter regardless of the data. Choosing an appropriate parameter usually requires an inefficient trial-and-error model selection process. Furthermore, it is even more difficult to find optimal parameter values in advance for personalized applications. The research presented in this thesis offers a novel nonparametric framework for human routine discovery that can infer high-level routines without knowing the number of latent low-level activities beforehand. More specifically, the frame-work automatically finds the size of the low-level feature vocabulary from sensor feature vectors at the vocabulary extraction phase. At the routine discovery phase, the framework further automatically selects the appropriate number of latent low-level activities and discovers latent routines. Moreover, we propose a new generative graphical model to incorporate multimodal sensor streams for the human activity discovery task. The hypothesis and approaches presented in this thesis are evaluated on public datasets in two routine domains: two daily-activity datasets and a transportation mode dataset. Experimental results show that our nonparametric framework can automatically learn the appropriate model parameters from multimodal sensor data without any form of manual model selection procedure and can outperform traditional parametric approaches for human routine discovery tasks.
]]>
Feng-Tso SunThe Natural Woman: Science and Sentimentality in Nineteenth - Century America
http://repository.cmu.edu/dissertations/358
http://repository.cmu.edu/dissertations/358Tue, 16 Sep 2014 05:10:52 PDTSheila LimingControlled-mobile Sensor Networks for Dynamic Sensing and Monitoring Applications
http://repository.cmu.edu/dissertations/357
http://repository.cmu.edu/dissertations/357Thu, 11 Sep 2014 07:37:36 PDT
Many potential indoor sensing and monitoring applications are characterized by hazardous and constantly-changing operating environments. For example, consider emergency response scenarios such as urban fire rescue. Traditionally, first responders have little access to situational information. In-situ information about the conditions, such as the extent and evolution of the indoor fire, can augment rescue efforts and reduce risk to emergency personnel. Static sensor networks that are pre-deployed or manually deployed have been proposed for such applications, but are less practical due to need for large infrastructure, lack of adaptivity and limited coverage. The main hypothesis of this thesis is that controlled-mobile networked sensing – the capability of nodes to move as per network needs, is a novel, feasible, and beneficial approach to monitoring dynamic and hazardous environments. Controlled-mobility in sensor networks can provide the desired autonomy and adaptability to overcome the limitations of static sensors. The research focuses on four of the major challenges in realizing controlled-mobile sensor networking systems: Understanding the trade-off between cost, weight, and sensing and actuation capabilities in designing a hardware platform for controlled-mobile sensing together with a complementary firmware architecture. Designing simulation environments for controlled-mobile sensing platforms that adequately incorporate both the cyber (network, processing, planning) and physical (motion, environment) components of such systems. Investigating the effects of controlled-mobility on network group discovery and maintenance protocols and designing approaches that meet the mobility, latency and energy constraints. Exploring novel low-overhead infrastructure-less mechanisms for collaborative coverage, deployment and navigation of resource-constrained controlled-mobile nodes in previously unseen environments. The thesis validates and evaluates the presented architecture, tools, and algorithms for controlled-mobile sensing systems through extensive simulations and a real-system test-bed implementation. The results show that controlled-mobility is feasible and can enable new class of sensing and monitoring applications.
]]>
Aveek PurohitQuenched Stresses And Linear Elastic Response Of Random Packings Of Frictionless Particles Near Jamming
http://repository.cmu.edu/dissertations/356
http://repository.cmu.edu/dissertations/356Thu, 11 Sep 2014 07:17:08 PDT
We study stress correlations and elastic response in large-scale computer simulations of particle packings near jamming. We show that there are characteristic lengths in both the stresses and elastic response that diverge in similar ways as the confining pressure approaches zero from above. For the case of the stress field, we show that the power spectrum of the hydrostatic pressure and shear stress agrees with a field-theoretic framework proposed by Henkes and Chakraborty [15] at short to intermediate wavelengths (where the power is flat in Fourier space), but contains significant excess power at wavelengths larger than about 50 to 100 particle diameters, with the specific crossover point going to larger wavelength at decreasing pressure, consistent with a divergence at p=0.These stress correlations were missed in previous studies by other groups due to limited system size. For the case of the elastic response, we probe the system in three ways: i) point forcing, ii) constrained homogeneous deformation where the system is driven with no-slip boundary conditions, and iii) free periodic homogeneous deformation. For the point force, we see distinct characteristic lengths for longitudinal and transverse modes each of which diverges in a different way with decreasing pressure with ET⇠p^{-1/4} and EL⇠p ^{-0.4} respectively. For the constrained homogeneous deformation we see a scaling of the local shear modulus with the size of the probing region consistent with E⇠p^{-1/2} similar to the EL⇠p^{-0.4} observed in the longitudinal component of the point response and in perfect agreement with the rigidity length discussed in recently proposed scenarios for jamming. Finally, we show that the transverse and longitudinal contributions to the strain field in response to unconstrained deformation (either volumetric or shear) have markedly different behavior. The transverse contribution is surprisingly invariant with respect to p with localized shear transformations dominating the response down to surprisingly small pressures. The longitudinal contribution develops a feature at small wavelength that intensifies with decreasing p but does not show any appreciable change in length. We interpret this pressure-invariant length as the characteristic shear zone size.
]]>
Kamran KarimiBiophysical Modeling and Optimal Control of DNA Amplification
http://repository.cmu.edu/dissertations/355
http://repository.cmu.edu/dissertations/355Thu, 11 Sep 2014 06:26:55 PDT
DNA amplification or the Polymerase Chain Reactions (PCR) is the workhorse of nearly every modern molecular biology laboratory, as well as the burgeoning discipline of personalized medicine. Despite the apparent simplicity of the PCR reaction, the method is often fraught with difficulties that can decrease the cycle efficiency or result in competitive amplification of undesired side products. The focus of this work is to derive an optimal reaction condition for a given PCR using the engineering discipline of control theory that can automatically derive prescriptions for the optimal temperature cycling protocols of a PCR reaction, if a suitable kinetic model exists. We first developed a theoretical model to estimate the sequence and temperature dependent rate parameters of a oligonucleotide hybridization or annealing reaction. Rate constants that were estimated using our model is in good agreement with the experimentally estimated rate constants of the same oligonucleotide hybridization reaction. Using the theory of enzyme processivity the kinetic parameters of enzyme binding and extension reactions were estimated experimentally. Thus, a first sequence-dependent biophysical model for DNA amplification has been developed. It is shown that amplification efficiency is affected by dynamic processes that are not accurately represented in simplified models of DNA amplification that are the basis of conventional temperature cycling protocols. Based on this analysis; a modified temperature protocol that improves the PCR efficiency is suggested. Use of this sequence- dependent kinetic model in a control theoretic framework to determine the optimal dynamic operating conditions of DNA amplification reactions, for any specified amplification objective, is discussed. Using these control systems, we demonstrate that there exists an optimal temperature cycling strategy for geometric amplification of any DNA sequence and formulate optimal control problems that can be used to derive the optimal temperature control. Strategies for the optimal synthesis of the DNA amplification control trajectory are proposed. Analogous methods can be used to formulate control problems for more advanced amplification objectives corresponding to the design of new types of DNA amplification reactions. Finally, a PCR optimal control problem is solved and an optimal temperature control that maximizes the desired DNA concentration as well as minimizes the total reaction is obtained.
]]>
Karthikeyan MarimuthuAdvanced and Alternative Fuel Vehicle Policies: Regulations and Incentives in the United States
http://repository.cmu.edu/dissertations/354
http://repository.cmu.edu/dissertations/354Thu, 11 Sep 2014 05:08:33 PDT
Transportation policy is playing an increasingly important role in the transition towards more fuel-efficient vehicles and alternative fuel vehicles (AFVs). Whether the policy seeks to promote adoption through mandatory requirements or through monetary incentives, or to address issues related to adoption of AFVs, it is clear that such policies can have large-ranging impacts on the future of the US transportation system. The work I conduct in my dissertation seeks to understand these policies, in the past, present, and future. I evaluate the effects of the Energy Policy Act of 2005 (EPACT) on the adoption of HEVs. As part of EPACT, a tax credit incentives program was implemented for consumers purchasing HEVs. Using a unique fixed effects regression approach with lagged instrumental variables, I am able to estimate the effects of the incentives. I find most significant responses occur when incentives exceed $1,000 in tax cd credit. Depending on the vehicle model the presence of EPACT yielded increases in sales of 5% to 15%. This increase is relatively smaller compared to many existing studies, which my work indicates is likely the result of over-attribution of sales to policy. I go on to examine the effects of the adoption of electric vehicles on funding for transportation infrastructure. A significant portion of revenue for transportation infrastructure comes from taxes on gasoline, these funds will likely be diminished to some extent as electric vehicles are adopted as they consume little to no gasoline as fuel. Using several existing electric vehicle models, I find that at the per-vehicle level, revenue generation can be upwards of 50% lower in certain states depending on how fees are charged. The total annual revenue generation at the federal level could decrease by as much as $200 million by 2025, though this is quite a small portion of total revenues for transportation infrastructure. I demonstrate that the revenue decrease can easily be made up through small policy fee changes in either flat fixed or through incremental increases in use fees, though implementation of such policies can be difficult politically. I also focus on the recent implementation of alternative fuel vehicle incentives in the 2009 update of the CAFE standards. I demonstrate that while the AFV incentives help spur the production and adoption of AFVs, there is a short-term emissions penalty due to the structure of the policy. i find that every AFV sold results in an increase in emissions rate for another vehicle of 50-400 grams of CO2 per mile, comparable to adding an additional conventional vehicle onto the road. The cumulative effect is an increase of 20 to 70 million metric tons of CO2 for vehicles sold between 2012 and 2025. I further extends this work by investigating how other policies promoting AFV sales interact with the CAFE policy. I focus specifically on the California ZEV mandate interaction and find that there is an increase of 120 million metric tons of CO2 for new cars sold between 2012 and 2025. The analysis also demonstrates a counter intuitive effect: the greater the success of ZEV in inducing adoption of AFVs, the greater the short-term emissions penalty due to the two policies. Finally I examines the response of driving behavior response to changes in gasoline prices. Using a unique dataset obtained from Pennsylvania's Department of Transportation, we are able to observe annual driving behavior at the individual vehicle level from 2000 through 2010. We observe heterogeneity of price elasticities using two methods: separating data by quantiles over the factors of interest and by interacting the factors of interest as categorical variables with gasoline prices. We find statistically significant variations in elasticities: for driving intensities we observe values of -0.172 increasing up to -0.0576 as the amount driven annually increases, for gasoline prices we observe a range of elasticities from -0.002 to -0.05 for prices below $4/gallon with a sudden increase to -0.182 for prices above $4/gallon, lastly for fuel economies we find that below 20 MPG elasticities are highest at -0.173 with decreasing responsiveness as vehicle fuel economy increases. Heterogeneity needs to be accounted for in order to properly understand policy effects: responses based on average elasticity values are likely to be incorrect.
]]>
Alan Theodore JennElementary Algorithms for Solving Convex Optimization Problems
http://repository.cmu.edu/dissertations/353
http://repository.cmu.edu/dissertations/353Wed, 10 Sep 2014 10:10:39 PDT
The rapid growth in data availability has led to modern large scale convex optimization problems that pose new practical and theoretical challenges. Examples include classification problems such as customer segmentation in retail and credit scoring in insurance. Classical optimization and machine learning techniques are typically inadequate to solve these large optimization problems because of high memory requirements and slow convergence guarantees. This thesis develops two research threads to address these issues. The first involves improving the effectiveness of a class of algorithms with simple computational steps for solving convex optimization problems arising in machine learning, data mining, and decision making. The second involves the refinement of conditioning and geometry of convex optimization problems via preconditioning. I elaborate on these two threads below. 1. The main theme of this thesis focuses on the class of elementary algorithms for solving convex optimization problems. These algorithms only involve simple operations such as matrix-vector multiplications, vector updates, and separation oracles. This simplicity makes the computational cost per iteration and memory requirement of these algorithms low. Thus, elementary algorithms are promising for solving emerging big data applications in areas such as classification, pattern recognition, and online learning. A major hurdle that needs to be overcome is the slow convergence of these algorithms. We develop new elementary algorithms that are enhanced via smoothing and dilation techniques. These enhancements yield algorithms that retain the attractive simplicity of the elementary algorithms while achieving substantially improved convergence rates. Thus, these enhanced algorithms are better suited for solving modern large scale convex optimization problems. 2. A significant difficulty when solving large convex optimization problems is poor conditioning caused by the existence of at and nearly at geometries. This thesis shows that a combination of two simple preprocessing steps generally improve the geometric structure of problem instances. We improve instances' geometric structure by reconciling the properties of three different but related notions of conditioning. More precisely, when one of these measures is large, in which case the problem instance is certainly poorly conditioned, our procedure reduces it without making the remaining measures worse. Our preconditioning procedures can be potentially useful for the convergence properties of a large class of iterative methods without changing their ultimate solution.
]]>
Negar Soheili AzadEssays on Asset Pricing and Portfolio Choice with Time-Varying Uncertainty
http://repository.cmu.edu/dissertations/352
http://repository.cmu.edu/dissertations/352Wed, 10 Sep 2014 08:02:48 PDT
In the first essay, I present a parsimonious consumption-based asset pricing model that explains the pricing of equity index options. The model has two key ingredients, a recursive utility function that overweights left-tail outcomes and a process for endowment volatility that allows for shocks with different persistence levels. The utility function produces a high price for tail risks and allows the model to replicate the implied volatility smirk in times of high uncertainty, during which extreme events are more likely. In times of low uncertainty the smirk arises due to mean reversion in volatility, which results in substantial volatility feedback and a conditional return distribution that is strongly left-skewed. The presence of multiple shock frequencies gives the variance premium the ability to predict returns over short horizons and the price-dividend ratio the ability to predict returns over long horizons, as in the data. Consistent with recent empirical evidence, the equity and variance premiums in the model arise predominantly from a high price of tail risk. The second essay (joint with Jan Schneemeier, University of Chicago) investigates the role of time-varying stock return volatility in a consumption and portfolio choice problem for a life-cycle investor facing short-selling and borrowing constraints. Faced with a benchmark investment strategy that conditions on age and wealth only, we find that an investor is willing to pay a fee of up to 1% - 1.5% of total life time consumption in order to optimally condition on volatility. Tilts in the optimal asset allocation in response to volatility shocks are considerably more pronounced than tilts in response to wealth shocks, and almost as important as life-cycle effects. Lastly, we find that the correlation between volatility and permanent labor income shocks may explain the low equity share of young households in the data. The third essay analyzes whether cross-sectional differences in the variance premium and the implied volatility smirk are related to the underlying firms' exposure to market variance risk and common idiosyncratic variance (CIV) risk. Using both cross-sectional regressions and sorts based on firms' loadings I find that firms whose variance co-moves more with market variance have steeper smirks and larger (less negative) variance premia. The latter finding is surprising in light of the fact that the variance premium of the market is believed to be negative. I show that the result persists in different sub-samples and that it is robust to various ways of estimating variance loadings. Exposure to CIV is not related to firm level option prices in a robust way.
]]>
David SchreindorferEssays in Financial Economics: Currency Risk and Pricing Kernal Volatility, CDS and Sovereign Bond Market Liquidity, CDS as Sovereign Debt Collateral
http://repository.cmu.edu/dissertations/351
http://repository.cmu.edu/dissertations/351Wed, 10 Sep 2014 06:58:55 PDT
Essay 1: CDS and Sovereign Bond Market Liquidity During the recent debt crisis in Europe, policy makers responded to the controversy surrounding CDS by implementing a series of policies that banned CDS trading. I use these bans as quasi-natural experiments to identify how derivative markets affect liquidity of the underlying cash market. I document that a temporary CDS ban increased bond market liquidity but a permanent ban instead decreased bond market liquidity. To explain these patterns, I build a dynamic search-theoretic model of over-the-counter bond and CDS markets that features an endogenous liquidity interaction between the two markets. My model shows that these opposing patterns are due to the fact that bond and CDS markets are substitute markets in the short run but are complementary markets in the long run. My results challenge existing theories of liquidity interaction among multiple markets and the common perception that the CDS market is a more liquid market than the bond market. Essay 2: CDS as Sovereign Debt Collateral A defining friction of sovereign debt is the lack of collateral that can back sovereign borrowing. This paper shows that credit default swaps (CDS) can serve as collateral and thereby support more sovereign borrowing. By giving more bargaining power to lenders in ex-post debt renegotiations, CDS becomes a commitment device for lenders to extract more repayment from the debtor country. This ex-post disciplining effect during debt renegotiations better aligns the sovereign’s ex-ante incentives with that of the lender. CDS alleviates agency frictions that are present in any lending contracts but are particularly difficult to mitigate in sovereign debt context. Essay 3: Currency Risk and Pricing Kernel Volatility A basic tenet of lognormal asset pricing models is that a risky currency is associated with low pricing kernel volatility. Empirical evidence indicates that a risky currency is associated with a relatively high interest rate. Taken together, these two statements associate high-interest-rate currencies with low pricing kernel volatility. We document evidence suggesting that the opposite is true, thus contradicting a fundamental empirical restriction of lognormal models. Our identification strategy revolves around using interest rate volatility differentials to make inferences about pricing kernel volatility differentials. In most lognormal models the two are monotonic functions of one another. A risky currency, therefore, is one with relatively low pricing kernel volatility and relatively low interest rate volatility. In the data, however, we see the opposite. High interest rates are associated with high interest rate volatility. This indicates that lognormal models of currency risk are inadequate and that future work should emphasize distributions in which higher moments play an important role. Our results apply to a fairly broad class of models, including Gaussian affine term structure models and many recent consumption-based models.
]]>
Batchimeg SambalaibatApproximate Dynamic Programming for Commodity and Energy Merchant Operations
http://repository.cmu.edu/dissertations/350
http://repository.cmu.edu/dissertations/350Wed, 10 Sep 2014 05:31:11 PDT
We study the merchant operations of commodity and energy conversion assets. Examples of such assets include natural gas pipelines systems, commodity swing options, and power plants. Merchant operations involves managing these assets as real options on commodity and energy prices with the objective of maximizing the market value of these assets. The economic relevance of natural gas conversion assets has increased considerably since the occurrence of the oil and gas shale boom; for example, the Energy Information Agency expects natural gas to be the source of 30% of the world's electricity production by 2040 and the McKinsey Global Institute projects United States spending on energy infrastructure to be about 100 Billion dollars by 2020. Managing commodity and energy conversion assets can be formulated as intractable Markov decision problems (MDPs), especially when using high dimensional price models commonly employed in practice. We develop approximate dynamic programming (ADP) methods for computing near optimal policies and lower and upper bounds on the market value of these assets. We focus on overcoming issues with the standard math programming and financial engineering ADP methods, that is, approximate linear programing (ALP) and least squares Monte Carlo (LSM), respectively. In particular, we develop: (i) a novel ALP relaxation framework to improve the ALP approach and use it to derive two new classes of ALP relaxations; (ii) an LSM variant in the context of popular practice-based price models to alleviate the substantial computational overhead when estimating upper bounds on the market value using existing LSM variants; and (iii) a mixed integer programming based ADP method that is exact with respect to a policy performance measure, while methods in the literature are heuristic in nature. Computational experiments on realistic instances of natural gas storage and crude oil swing options show that both our ALP relaxations and LSM methods are efficient and deliver near optimal policies and tight lower and upper bounds. Our LSM variant is also between one and three orders of magnitude faster than existing LSM variants for estimating upper bounds. Our mixed integer programming ADP model is computationally expensive to solve but its exact nature motivates further research into its solution. We provide theoretical support for our methods: By deriving bounds on approximation error we establish the optimality of our best ALP relaxation class in limiting regimes of practical relevance and provide a theoretical perspective on the relative performance of our LSM variant and existing LSM variants. We also unify different ADP methods in the literature using our ALP relaxation framework, including the financial engineering based LSM method. In addition, we employ ADP to study the novel application of jointly managing storage and transport assets in a natural gas pipeline system; the literature studies these assets in isolation. We leverage our structural analysis of the optimal storage policy to extend an LSM variant for this problem. This extension computes near optimal policies and tight bounds on instances formulated in collaboration with a major natural gas trading company. We use our extension and these instances to answer questions relevant to merchants managing such assets. Overall, our findings highlight the role of math programming for developing ADP methods. Although we focus on managing commodity and energy conversion assets, the techniques in this thesis have potential broader relevance for solving MDPs in other application contexts, such as inventory control with demand forecast updating, multiple sourcing, and optimal medical treatment design.
]]>
Selvaprabu NadarajahEssays on the Economics of Education
http://repository.cmu.edu/dissertations/349
http://repository.cmu.edu/dissertations/349Mon, 08 Sep 2014 10:10:03 PDTJason ImbrognoThe Labor Economics of the Great Migration
http://repository.cmu.edu/dissertations/348
http://repository.cmu.edu/dissertations/348Mon, 08 Sep 2014 09:41:35 PDTJohn GardnerEssays on the Provision and Funding of Public Goods
http://repository.cmu.edu/dissertations/347
http://repository.cmu.edu/dissertations/347Mon, 08 Sep 2014 07:36:54 PDT
The first essay studies how political parties’ choice of public good provision and tax funding affect the risk of default to public debt investors. Past research has largely ignored the effects that political parties have on default risk of state governments. The objective of this paper is to address this policy question using data from Credit Default Swap contracts (CDS), and poll data from state gubernatorial elections. The findings of the paper suggest that state Republican governors have a significant positive effect on CDS spreads. On average, Republican governors reduce credit spreads by around six percent, more than half of CDS standard deviation during election race. Prospects of a Republican administration are good news for debtholders. The positive effect of Republican candidates is larger when: candidates signed the ``Taxpayer Protection Pledge'', Democrats control the state houses and for highly contested gubernatorial elections. The second essay studies profiling and affirmative action in the access to gifted programs, a common public good provided by school districts. For decades, colleges and universities have struggled to increase participation of minority and disadvantaged students. Urban school districts confront a parallel challenge; minority and disadvantaged students are underrepresented in selective programs that use merit-based admission. We analyze optimal school district policy and develop an econometric framework providing a unified treatment of affirmative action and profiling. Implementing the model for a central-city district, we find profiling by race and income, affirmative action for low-income students, and no affirmative action with respect race. Counterfactual analysis reveals that these policies achieve 80\% of African American enrollment that would be could be attained by race-based affirmative action. The third essay studies a new alternative mean of funding for States and local authorities called Build America Bonds (BAB). BABs were issued by municipalities for twenty months as part of the 2009 fiscal package. Unlike traditional tax-exempt municipals, BABs are taxable to the holder, but the Treasury rebates 35% of the coupon to the issuer. The stated purpose was to provide municipalities access to a more liquid market including foreign, tax-exempt, and tax-deferred investors. We find BABs do not exhibit greater liquidity than traditional municipals. BABs are more underpriced initially, particularly for interdealer trades. BABs also show a substitution from underwriter fees toward more underpricing, suggesting the underpricing is a strategic response to the tax subsidy.
]]>
Dario CestauEssays in Service Operations Management
http://repository.cmu.edu/dissertations/346
http://repository.cmu.edu/dissertations/346Mon, 08 Sep 2014 06:51:23 PDT
In this dissertation, I discuss three problems within service operations management: identifying situational attributes that lead to positive customer outcomes under a Twitter-based customer service framework; the conditions for finite delay of first-in-first-out multiserver systems when confronted with integral loads; and the relative performance of different bargaining mechanisms for a seller of finite perishable inventory, with a further investigation of the consequences of modeling private information. First, we consider a large telecommunications company that provides customer support over Twitter. Using 10 months of service data, we apply model selection techniques to develop an ordinal logistic regression model assessing the probability that a given customer service interaction will result in a positive, neutral or negative resolution as determined by the customer’s sentiment expression. Our model incorporates customer, service and network explanatory attributes. We find that customers are less likely to experience a positive final sentiment as time passes, that is, those cases later in the 10 month period studied are less likely to experience positive resolution. This suggests that there is a drop-off in the likelihood of more positive resolution, but that this effect levels off. This finding may indicate a shift by the customer service team to harder to resolve cases as the program matures. Next, we consider conditions for finite expected delay in FIFO multiserver queues with integral loads. Scheller-Wolf and Vesilo (2006) find necessary and sufficient conditions for a finite rth moment of expected delay in a FIFO multiserver queue, assuming a non-integral load and a service time distribution belonging to class L_{1}^{B} . Removing the non-integral load assumption results in a gap between the identified necessary and sufficient conditions, as discussed by Foss (2009). We decrease the size of this gap through the application of domain of attraction results. Specifically, we find a stricter necessary condition for a GI/GI/K-server system with integral p that is more restrictive than those in the literature. Finally, we consider the problem of a seller with a finite supply of perishable inventory. We consider four price setting mechanisms: seller posted price, buyer posted price, split-the-difference, and the neutral bargaining solution. We rank the value of these different mechanisms analytically and numerically in the context of the symmetric uniform trading problem from the perspective of the seller. While the ordering of the mechanisms remains the same as compared to the infinite horizon case studied in the literature, we use a model analogous to the infinite horizon case to find numerically that the relative value of the split-the-difference mechanism increases when the seller ultimately faces a dead- line to complete the sales. The split-the-difference mechanism becomes more valuable as the ratio of available inventory to time remaining increases because it is more likely to result in a sale than the seller posted price mechanism. In general, modeling private information is more challenging for the split-the-difference and neutral bargaining solution mechanisms than for the two posted price mechanisms. To assess the importance of this added complication, we quantify the effect of modeling private information when computing the seller’s opportunity cost and find that while private information makes only a small difference in the neutral bargaining solution case, this modeling choice makes a large difference in the split-the-difference case when the seller is weak.
]]>
Michele DufallaCompetition and Cooperation in Global Supply Chain Networks
http://repository.cmu.edu/dissertations/345
http://repository.cmu.edu/dissertations/345Mon, 08 Sep 2014 06:20:25 PDTXin FangPower-Electronics-Enabled Transient Stabilization of Power Systems
http://repository.cmu.edu/dissertations/344
http://repository.cmu.edu/dissertations/344Thu, 04 Sep 2014 01:52:43 PDT
Transient stability of electric energy grids is defined as the ability of the power system to remain in synchronism during large disturbances. If the grid is not equipped with controllers capable of transiently stabilizing system dynamics, large disturbances could cause protection to trigger disconnecting the equipment and leading further to cascading system-wide blackouts. Today’s practice of tuning controllers generally does not guarantee a transiently stable response because it does not use a model for representing system-wide dynamic interactions. To overcome this problem, in this thesis we propose a new systems modeling and control design for provable transient stabilization of power systems against a given set of disturbances. Of particular interest are fast power-electronically-controlled Flexible Alternating Current Transmission System (FACTS) devices which have become a new major option for achieving transient stabilization. The first major contribution of this thesis is a framework for modeling of general interconnected power systems for very fast transient stabilization using FACTS devices. We recognize that a dynamic model for transient stabilization of power systems has to capture fast electromagnetic dynamics of the transmission grid and FACTS, in addition to the commonly-modeled generator dynamics. To meet this need, a nonlinear dynamic model of general interconnected electric power systems is derived using time-varying phasors associated with states of all dynamic components. The second major contribution of this thesis is a two-level approach to modeling and control which exploits the unique network structure and enables preserving only relevant dynamics in the nonlinear system model. This approach is fundamentally based on separating: a) internal dynamics model for ensuring stable local response of components; b) system-level model in terms of interaction variables for ensuring stability of the system when the components are interconnected. The two levels can be controlled separately which minimizes the need for communication between controllers. Both distributed and cooperative ectropy-based controllers are proposed to control the interaction-level of system dynamics. Proof of concept simulations are presented to illustrate and compare the promising performance of the derived controllers. Some of the most advanced FACTS industry installations are modeled and further generalized using our approach.
]]>
Milos CvetkovicA Complexity Theory for VLSI
http://repository.cmu.edu/dissertations/343
http://repository.cmu.edu/dissertations/343Wed, 27 Aug 2014 09:34:03 PDT
The established methodologies for studying computational complexity can be applied to the new problems posed by very large-scale integrated (VLSI) circuits. This thesis develops a ''VLSI model of computation'' and derives upper and lower bounds on the silicon area and time required to solve the problems of sorting and discrete Fourier transformation. In particular, the area A and time T taken by any VLSI chip using any algorithm to perform an N-point Fourier transform must satisfy AT^{2} ≥ c N^{2 }log^{2 }N, for some fixed c > 0. A more general result for both sorting and Fourier transformation is that AT^{2x} = Ω(N^{1 + x }log^{2x }N) for any x in the range 0 < x < 1. Also, the energy dissipated by a VLSI chip during the solution of either of these problems is at least Ω(N^{3/2 }log N). The tightness of these bounds is demonstrated by the existence of nearly optimal circuits for both sorting and Fourier transformation. The circuits based on the shuffle-exchange interconnection pattern are fast but large: T = O(log^{2 }N) for Fourier transformation, T = O(log^{3 }N) for sorting; both have area A of at most O(N^{2} / log^{1/2 }N). The circuits based on the mesh interconnection pattern are slow but small: T = O(N^{1/2 }loglogN), A = O(N log^{2}N).
]]>
C. D. ThompsonSemi-Supervised and Latent-Variable Models of Natural Language Semantics
http://repository.cmu.edu/dissertations/342
http://repository.cmu.edu/dissertations/342Mon, 18 Aug 2014 05:40:45 PDT
This thesis focuses on robust analysis of natural language semantics. A primary bottleneck for semantic processing of text lies in the scarcity of high-quality and large amounts of annotated data that provide complete information about the semantic structure of natural language expressions. In this dissertation, we study statistical models tailored to solve problems in computational semantics, with a focus on modeling structure that is not visible in annotated text data. We first investigate supervised methods for modeling two kinds of semantic phenomena in language. First, we focus on the problem of paraphrase identification, which attempts to recognize whether two sentences convey the same meaning. Second, we concentrate on shallow semantic parsing, adopting the theory of frame semantics (Fillmore, 1982). Frame semantics offers deep linguistic analysis that exploits the use of lexical semantic properties and relationships among semantic frames and roles. Unfortunately, the datasets used to train our paraphrase and frame-semantic parsing models are too small to lead to robust performance. Therefore, a common trait in our methods is the hypothesis of hidden structure in the data. To this end, we employ conditional log-linear models over structures, that are firstly capable of incorporating a wide variety of features gathered from the data as well as various lexica, and secondly use latent variables to model missing information in annotated data. Our approaches towards solving these two problems achieve state-of-the-art accuracy on standard corpora. For the frame-semantic parsing problem, we present fast inference techniques for jointly modeling the semantic roles of a given predicate. We experiment with linear program formulations, and use a commercial solver as well as an exact dual decomposition technique that breaks the role labeling problem into several overlapping components. Continuing with the theme of hypothesizing hidden structure in data for modeling natural language semantics, we present methods to leverage large volumes of unlabeled data to improve upon the shallow semantic parsing task. We work within the framework of graph-based semi-supervised learning, a powerful method that associates similar natural language types, and helps propagate supervised annotations to unlabeled data. We use this framework to improve frame-semantic parsing performance on unknown predicates that are absent in annotated data. We also present a family of novel objective functions for graph-based learning that result in sparse probability measures over graph vertices, a desirable property for natural language types. Not only are these objectives easier to numerically optimize, but also they result in smoothed distributions over predicates that are smaller in size. The experiments presented in this dissertation empirically demonstrates that missing information in text corpora contain considerable semantic information that can be incorporated into structured models for semantics, to significant benefit over the current state of the art. The methods in this thesis were originally presented by Das and Smith (2009, 2011, 2012), and Das et al. (2010, 2012). The thesis gives a more thorough exposition, relating and comparing the methods, and also presents several extensions of the aforementioned papers.
]]>
Dipanjan DasStatistical Source Expansion for Question Answering
http://repository.cmu.edu/dissertations/341
http://repository.cmu.edu/dissertations/341Sun, 17 Aug 2014 10:19:19 PDT
A source expansion algorithm automatically extends a given text corpus with related information from large, unstructured sources. While the expanded corpus is not intended for human consumption, it can be leveraged in question answering (QA) and other information retrieval or extraction tasks to find more relevant knowledge and to gather additional evidence for evaluating hypotheses. In this thesis, we propose a novel algorithm that expands a collection of seed documents by (1) retrieving related content from the Web or other large external sources, (2) extracting self-contained text nuggets from the related content, (3) estimating the relevance of the text nuggets with regard to the topics of the seed documents using a statistical model, and (4) compiling new pseudo-documents from nuggets that are relevant and complement existing information. In an intrinsic evaluation on a dataset comprising 1,500 hand-labeled web pages, the most elective statistical relevance model ranked text nuggets by relevance with 81% MAP, compared to 43% when relying on rankings generated by a web search engine, and 75% when using a multi-document summarization algorithm. These differences are statistically significant and result in noticeable gains in search performance in a task-based evaluation on QA datasets. The statistical models use a comprehensive set of features to predict the topicality and quality of text nuggets based on topic models built from seed content, search engine rankings and surface characteristics of the retrieved text. Linear models that evaluate text nuggets individually are compared to a sequential model that estimates their relevance given the surrounding nuggets. The sequential model leverages features derived from text segmentation algorithms to dynamically predict transitions between relevant and irrelevant passages. It slightly outperforms the best linear model while using fewer parameters and requiring less training time. In addition, we demonstrate that active learning reduces the amount of labeled data required to fit a relevance model by two orders of magnitude with little loss in ranking performance. This facilitates the adaptation of the source expansion algorithm to new knowledge domains and applications. Applied to the QA task, the proposed method yields consistent and statistically significant performance gains across different datasets, seed corpora and retrieval strategies. We evaluated the impact of source expansion on search performance and end-to-end accuracy using Watson and the OpenEphyra QA system, and datasets comprising over 6,500 questions from the Jeopardy! quiz show and TREC evaluations. By expanding various seed corpora with web search results, we were able to improve the QA accuracy of Watson from 66% to 71% on regular Jeopardy! questions, from 45% to 51% on Final Jeopardy! questions and from 59% to 64% on TREC factoid questions. We also show that the source expansion approach can be adapted to extract relevant content from locally stored sources without requiring a search engine, and that this method yields similar performance gains. When combined with the approach that uses web search results, Watson's accuracy further increases to 72% on regular Jeopardy! data, 54% on Final Jeopardy! and 67% on TREC questions.
]]>
Nico SchlaeferContinuous Graphical Models for Static and Dynamic Distributions: Application to Structural Biology
http://repository.cmu.edu/dissertations/340
http://repository.cmu.edu/dissertations/340Sun, 17 Aug 2014 08:42:30 PDT
Generative models of protein structure enable researchers to predict the behavior of proteins under different conditions. Continuous graphical models are powerful and efficient tools for modeling static and dynamic distributions, which can be used for learning generative models of molecular dynamics. In this thesis, we develop new and improved continuous graphical models, to be used in modeling of protein structure. We first present von Mises graphical models, and develop consistent and efficient algorithms for sparse structure learning and parameter estimation, and inference. We compare our model to sparse Gaussian graphical model and show it outperforms GGMs on synthetic and Engrailed protein molecular dynamics datasets. Next, we develop algorithms to estimate Mixture of von Mises graphical models using Expectation Maximization, and show that these models outperform Von Mises, Gaussian and mixture of Gaussian graphical models in terms of accuracy of prediction in imputation test of non-redundant protein structure datasets. We then use non-paranormal and nonparametric graphical models, which have extensive representation power, and compare several state of the art structure learning methods that can be used prior to nonparametric inference in reproducing kernel Hilbert space embedded graphical models. To be able to take advantage of the nonparametric models, we also propose feature space embedded belief propagation, and use random Fourier based feature approximation in our proposed feature belief propagation, to scale the inference algorithm to larger datasets. To improve the scalability further, we also show the integration of Coreset selection algorithm with the nonparametric inference, and show that the combined model scales to large datasets with very small adverse effect on the quality of predictions. Finally, we present time varying sparse Gaussian graphical models, to learn smoothly varying graphical models of molecular dynamics simulation data, and present results on CypA protein
]]>
Narges Sharif Razavian