Bayesian Inference in Large-scale Problems

Many modern applications fall into the category of "large-scale" statistical problems, in which both the number of observations n and the number of features or parameters p may be large. Many existing methods focus on point estimation, despite the continued relevance of uncertainty quantification in the sciences, where the number of parameters to estimate often exceeds the sample size, despite huge increases in the value of n typically seen in many fields. Thus, the tendency in some areas of industry to dispense with traditional statistical analysis on the basis that "n=all" is of little relevance outside of certain narrow applications. The main result of the Big Data revolution in most fields has instead been to make computation much harder without reducing the importance of uncertainty quantification. Bayesian methods excel at uncertainty quantification, but often scale poorly relative to alternatives. This conflict between the statistical advantages of Bayesian procedures and their substantial computational disadvantages is perhaps the greatest challenge facing modern Bayesian statistics, and is the primary motivation for the work presented here.

Two general strategies for scaling Bayesian inference are considered. The first is the development of methods that lend themselves to faster computation, and the second is design and characterization of computational algorithms that scale better in n or p. In the first instance, the focus is on joint inference outside of the standard problem of multivariate continuous data that has been a major focus of previous theoretical work in this area. In the second area, we pursue strategies for improving the speed of Markov chain Monte Carlo algorithms, and characterizing their performance in large-scale settings. Throughout, the focus is on rigorous theoretical evaluation combined with empirical demonstrations of performance and concordance with the theory.

One topic we consider is modeling the joint distribution of multivariate categorical data, often summarized in a contingency table. Contingency table analysis routinely relies on log-linear models, with latent structure analysis providing a common alternative. Latent structure models lead to a reduced rank tensor factorization of the probability mass function for multivariate categorical data, while log-linear models achieve dimensionality reduction through sparsity. Little is known about the relationship between these notions of dimensionality reduction in the two paradigms. In Chapter 2, we derive several results relating the support of a log-linear model to nonnegative ranks of the associated probability tensor. Motivated by these findings, we propose a new collapsed Tucker class of tensor decompositions, which bridge existing PARAFAC and Tucker decompositions, providing a more flexible framework for parsimoniously characterizing multivariate categorical data. Taking a Bayesian approach to inference, we illustrate empirical advantages of the new decompositions.

Latent class models for the joint distribution of multivariate categorical, such as the PARAFAC decomposition, data play an important role in the analysis of population structure. In this context, the number of latent classes is interpreted as the number of genetically distinct subpopulations of an organism, an important factor in the analysis of evolutionary processes and conservation status. Existing methods focus on point estimates of the number of subpopulations, and lack robust uncertainty quantification. Moreover, whether the number of latent classes in these models is even an identified parameter is an open question. In Chapter 3, we show that when the model is properly specified, the correct number of subpopulations can be recovered almost surely. We then propose an alternative method for estimating the number of latent subpopulations that provides good quantification of uncertainty, and provide a simple procedure for verifying that the proposed method is consistent for the number of subpopulations. The performance of the model in estimating the number of subpopulations and other common population structure inference problems is assessed in simulations and a real data application.

In contingency table analysis, sparse data is frequently encountered for even modest numbers of variables, resulting in non-existence of maximum likelihood estimates. A common solution is to obtain regularized estimates of the parameters of a log-linear model. Bayesian methods provide a coherent approach to regularization, but are often computationally intensive. Conjugate priors ease computational demands, but the conjugate Diaconis--Ylvisaker priors for the parameters of log-linear models do not give rise to closed form credible regions, complicating posterior inference. In Chapter 4 we derive the optimal Gaussian approximation to the posterior for log-linear models with Diaconis--Ylvisaker priors, and provide convergence rate and finite-sample bounds for the Kullback-Leibler divergence between the exact posterior and the optimal Gaussian approximation. We demonstrate empirically in simulations and a real data application that the approximation is highly accurate, even in relatively small samples. The proposed approximation provides a computationally scalable and principled approach to regularized estimation and approximate Bayesian inference for log-linear models.

Another challenging and somewhat non-standard joint modeling problem is inference on tail dependence in stochastic processes. In applications where extreme dependence is of interest, data are almost always time-indexed. Existing methods for inference and modeling in this setting often cluster extreme events or choose window sizes with the goal of preserving temporal information. In Chapter 5, we propose an alternative paradigm for inference on tail dependence in stochastic processes with arbitrary temporal dependence structure in the extremes, based on the idea that the information on strength of tail dependence and the temporal structure in this dependence are both encoded in waiting times between exceedances of high thresholds. We construct a class of time-indexed stochastic processes with tail dependence obtained by endowing the support points in de Haan's spectral representation of max-stable processes with velocities and lifetimes. We extend Smith's model to these max-stable velocity processes and obtain the distribution of waiting times between extreme events at multiple locations. Motivated by this result, a new definition of tail dependence is proposed that is a function of the distribution of waiting times between threshold exceedances, and an inferential framework is constructed for estimating the strength of extremal dependence and quantifying uncertainty in this paradigm. The method is applied to climatological, financial, and electrophysiology data.

The remainder of this thesis focuses on posterior computation by Markov chain Monte Carlo. The Markov Chain Monte Carlo method is the dominant paradigm for posterior computation in Bayesian analysis. It has long been common to control computation time by making approximations to the Markov transition kernel. Comparatively little attention has been paid to convergence and estimation error in these approximating Markov Chains. In Chapter 6, we propose a framework for assessing when to use approximations in MCMC algorithms, and how much error in the transition kernel should be tolerated to obtain optimal estimation performance with respect to a specified loss function and computational budget. The results require only ergodicity of the exact kernel and control of the kernel approximation accuracy. The theoretical framework is applied to approximations based on random subsets of data, low-rank approximations of Gaussian processes, and a novel approximating Markov chain for discrete mixture models.

Data augmentation Gibbs samplers are arguably the most popular class of algorithm for approximately sampling from the posterior distribution for the parameters of generalized linear models. The truncated Normal and Polya-Gamma data augmentation samplers are standard examples for probit and logit links, respectively. Motivated by an important problem in quantitative advertising, in Chapter 7 we consider the application of these algorithms to modeling rare events. We show that when the sample size is large but the observed number of successes is small, these data augmentation samplers mix very slowly, with a spectral gap that converges to zero at a rate at least proportional to the reciprocal of the square root of the sample size up to a log factor. In simulation studies, moderate sample sizes result in high autocorrelations and small effective sample sizes. Similar empirical results are observed for related data augmentation samplers for multinomial logit and probit models. When applied to a real quantitative advertising dataset, the data augmentation samplers mix very poorly. Conversely, Hamiltonian Monte Carlo and a type of independence chain Metropolis algorithm show good mixing on the same dataset.

Testing of Interposer-Based 2.5D Integrated Circuits

The unprecedented and relentless growth in the electronics industry is feeding the demand for integrated circuits (ICs) with increasing functionality and performance at minimum cost and power consumption. As predicted by Moore's law, ICs are being aggressively scaled to meet this demand. While the continuous scaling of process technology is reducing gate delays, the performance of ICs is being increasingly dominated by interconnect delays. In an effort to improve submicrometer interconnect performance, to increase packing density, and to reduce chip area and power consumption, the semiconductor industry is focusing on three-dimensional (3D) integration. However, volume production and commercial exploitation of 3D integration are not feasible yet due to significant technical hurdles.

At the present time, interposer-based 2.5D integration is emerging as a precursor to stacked 3D integration. All the dies and the interposer in a 2.5D IC must be adequately tested for product qualification. However, since the structure of 2.5D ICs is different from the traditional 2D ICs, new challenges have emerged: (1) pre-bond interposer testing, (2) lack of test access, (3) limited ability for at-speed testing, (4) high density I/O ports and interconnects, (5) reduced number of test pins, and (6) high power consumption. This research targets the above challenges and effective solutions have been developed to test both dies and the interposer.

The dissertation first introduces the basic concepts of 3D ICs and 2.5D ICs. Prior work on testing of 2.5D ICs is studied. An efficient method is presented to locate defects in a passive interposer before stacking. The proposed test architecture uses e-fuses that can be programmed to connect or disconnect functional paths inside the interposer. The concept of a die footprint is utilized for interconnect testing, and the overall assembly and test flow is described. Moreover, the concept of weighted critical area is defined and utilized to reduce test time. In order to fully determine the location of each e-fuse and the order of functional interconnects in a test path, we also present a test-path design algorithm. The proposed algorithm can generate all test paths for interconnect testing.

In order to test for opens, shorts, and interconnect delay defects in the interposer, a test architecture is proposed that is fully compatible with the IEEE 1149.1 standard and relies on an enhancement of the standard test access port (TAP) controller. To reduce test cost, a test-path design and scheduling technique is also presented that minimizes a composite cost function based on test time and the design-for-test (DfT) overhead in terms of additional through silicon vias (TSVs) and micro-bumps needed for test access. The locations of the dies on the interposer are taken into consideration in order to determine the order of dies in a test path.

To address the scenario of high density of I/O ports and interconnects, an efficient built-in self-test (BIST) technique is presented that targets the dies and the interposer interconnects. The proposed BIST architecture can be enabled by the standard TAP controller in the IEEE 1149.1 standard. The area overhead introduced by this BIST architecture is negligible; it includes two simple BIST controllers, a linear-feedback-shift-register (LFSR), a multiple-input-signature-register (MISR), and some extensions to the boundary-scan cells in the dies on the interposer. With these extensions, all boundary-scan cells can be used for self-configuration and self-diagnosis during interconnect testing. To reduce the overall test cost, a test scheduling and optimization technique under power constraints is described.

In order to accomplish testing with a small number test pins, the dissertation presents two efficient ExTest scheduling strategies that implements interconnect testing between tiles inside an system on chip (SoC) die on the interposer while satisfying the practical constraint that the number of required test pins cannot exceed the number of available pins at the chip level. The tiles in the SoC are divided into groups based on the manner in which they are interconnected. In order to minimize the test time, two optimization solutions are introduced. The first solution minimizes the number of input test pins, and the second solution minimizes the number output test pins. In addition, two subgroup configuration methods are further proposed to generate subgroups inside each test group.

Finally, the dissertation presents a programmable method for shift-clock stagger assignment to reduce power supply noise during SoC die testing in 2.5D ICs. An SoC die in the 2.5D IC is typically composed of several blocks and two neighboring blocks that share the same power rails should not be toggled at the same time during shift. Therefore, the proposed programmable method does not assign the same stagger value to neighboring blocks. The positions of all blocks are first analyzed and the shared boundary length between blocks is then calculated. Based on the position relationships between the blocks, a mathematical model is presented to derive optimal result for small-to-medium sized problems. For larger designs, a heuristic algorithm is proposed and evaluated.

In summary, the dissertation targets important design and optimization problems related to testing of interposer-based 2.5D ICs. The proposed research has led to theoretical insights, experiment results, and a set of test and design-for-test methods to make testing effective and feasible from a cost perspective.