The main focus of this dissertation is on the optimal design of motion compensation scheme for predictive coding in video compression. The “sub-optimality” of conventional block-based motion compensation scheme, wherein the impact of a motion vector is confined within in a rigid rectangular block, motivates our design of a multi-hypothesis motion compensation scheme. We explicitly treat motion vectors as pointers to observation sources. Given the high-correlation between adjacent pixels in nature images, the motion vectors of neighboring blocks can point to relevant estimates of the current target. This dissertation work builds on this paradigm and demonstrates advanced techniques devised by incorporating the information of the entire motion vector field.

We first directly formulate the problem of motion compensation with multiple estimates as a linear estimation problem, and design a training method to derive the optimal linear coefficients to avoid overfitting as well as to minimize the ultimate reconstruction errors rather than prediction errors. As a single set of coefficients cannot capture the varying statistics of video sequences, we design K sets of coefficients which are trained off-line through “K-mode” iterative clustering techniques. By switching between the predefined sets of coefficients, the encoder can adapt to local statistics. This approach is then extended to the setting of variable block size partitioning to enjoy the substantial gain provided by the flexibility of dividing blocks to approximate object shapes. As the additional side information to indicate the set of prediction coefficients used to generate the final prediction is generally not negligible, a parametric framework is proposed to model the statistics of estimates and target pixels. The model leverages the first-order Markov property for image signals and relationships between motion vectors in the motion field. As a result, the coefficients derived from the model can automatically adapt to local variations without additional side information. Moreover, using the parametric approach, we can combine estimates from any number of motion vectors at any pixel location, which allows us to completely break free from the block structure by allowing a motion vector’s influence to be of arbitrary shape.

The reminder of this dissertation is focused on optimization on the AV1 encoder, the open source video encoder founded by the Alliance of Open Media (AOM). We first introduce a new coding tool to extend the number of reference frames, and then we use the existing coding tools to design a coding structure, wherein the reference frames are allocated to cover a wider temporal range to offer more diversities. This new design allows the encoder to better capture temporal variations. Finally, we introduce a complementary compound prediction mode, which is designed and optimized for the blocks where the existing compound modes fall short. Simulations provide experimental evidences for the efficiency of proposed mode with consistent coding gain across all bit-rate range.

Plant cell walls (PCWs) are the most abundant biomass on planet Earth. Unlike vertebrate animals, many invertebrates, especially insects, can digest PCWs, either to access the protoplasm or to break the wall itself into digestible, nutritive oligomers and monomers (Pauchet, Wilkinson et al. 2010, Calderon-Cortes, Quesada et al. 2012). This ability of many insects to break down recalcitrant PCWs has contributed greatly to their biodiversity and success around the world (Tokuda 2019). Traditionally, the breakdown of lignocellulosic and cellulosic material by insects was thought to be dependent on gastrointestinal microbes. Recent research with molecular and omics techniques (genomics, transcriptomics, and proteomics) has provided evidence to reveal PCW-degrading mechanisms in insects that use various arsenals of endogenous digestive enzymes. A few studies have reported on the endogenous PCW-degrading enzymes (PCWDEs) of Phasmatodea or stick insects. Much is still unknown with respect to how and where PCWDEs perform in a phasmid’s digestive tract, in addition to the question of whether gut microbial PCWDEs are present to contribute. This dissertation applies RNA-Seq and metagenomic analysis to identify the repertoire of endogenous PCWDEs in Aretaon asperrimus and Medauroidea extradentata, to assess the expression profiles of these enzymes through the alimentary canal, and to investigate the potential digestive role that could be played by gut microbes. We identify abundant endogenous PCWDE repertoires in each species, including GH9 endoglucanases, GH1 cellobiases, and GH28 polygalacturonases. Most PCWDE families are highly expressed in the anterior section of the midgut, while beta-glucosidases show a global expression pattern in multiple tissues, including the head. We observed very low expressions of GH9 and GH28 in hindgut tissues (ileum and rectum). Hemicellulases from GH2, GH3, and GH30 also showed global expression profiles without high expression in the AMG. The compartmentalization of phasmid digestion is investigated by an annotation and enrichment test for differentially expressed genes and the 1,000 most highly expressed genes in each tissue. The metagenomic study indicates that the dominant bacterial phylum, Proteobacteria, in both species’ gut communities, shows a great number of polygalacturonase sequences in the posterior section, which infers a potential pectin degradation synergism between endogenous and bacterial pectinases in different sections of a phasmid’s midgut. A lesser capability for PCW-degrading was observed in both species’ midgut microbiota, with a lack of cellulolytic Fibrobacteres and Spirochaetes. Overall, the identified transcripts of various glycoside hydrolase families in this dissertation further support that phasmids have uncommonly rich endogenous repertoires of PCWDEs. All these sequences could benefit future evolutionary analyses of these gene families and, more broadly, the phylogenetic analyses in Phasmatodea. Our results from the expression profiles of identified PCWDEs take a step further to illustrate how these endogenous digestive enzymes perform differently among tissues. The discovery of many bacterial PCWDEs in phasmid’s midgut in this dissertation indicates a possible synergism in the degradation of PCWs, which contradicts previous claims. All these findings contribute to the understanding of the digestive physiology in Phasmatodea and provide genomic data for future studies.

Traffic assignment models are an important component in analyzing the relationship between demand and supply in the transportation network for design, planning, and control purposes. The static traffic assignment model has been used in practice for several decades. With the latest development in the area of Advanced Traffic Management Systems (ATMS) and Advanced Traveler Information Systems (ATIS), there is an increasing demand for dynamic traffic assignment models to serve as a basis for studying various issues in these areas.

Existing dynamic user-equilibrium traffic assignment (DUETA) models are mostly expanded from the static user-equilibrium traffic assignment model by intro- ducing the time dimension along with a group of additional constraints. Whereas the equivalency between the solution to the traffic assignment model and the user-equilibrium condition as defined by Wardrop is well established in the static case, the same may not be true for the dynamic case. This paper examines the general form of DUETA models as proposed in previous research and shows that, if queuing behavior is represented in the model at a minimal level, the solution to conventional DUETA models with an objective function of the form Et Xi Jti(t) f(w)dw may not necessarily converge to or approximate the Wardropian user-equilibrium condition in the dynamic sense as defined by many researchers.

This dissertation studies two important fields of development economics in China, including agriculture (chapters 1 and 3) and education (chapter 2). Guided by theories in learning and decision-making literature, I exploit randomized controlled trial (RCT), lab experiment, and natural experiments to causally estimate the impact of learning (failure) on the productivity-related outcomes.

In Chapter 1 (coauthored with Binkai Chen and Ao Wang), we investigate agents' simultaneous learning about multiple interacting technologies in the context of fertilizer application in China. We first present experimental evidence that farmers overuse nitrogen fertilizers and underuse phosphorus and potassium fertilizers, relative to the personalized fertilizer recommendations based on plot-level soil analysis. Our first-phase intervention that provides customized fertilizer recommendations leads to reduced nitrogen applications and increased phosphorus/potassium uses. Average yields and revenues are 5-7% higher, while total fertilizer costs remain unchanged. These results are also consistent with a meaningful reduction in greenhouse gas (N2O) emissions linked to nitrogen overuse. Survey data suggest that farmers overestimate the return to nitrogen because it produces a salient signal on crops by increasing greenness, but they underestimate the effectiveness of phosphorus and potassium because their effects are barely observable during the growing stages. Motivated by these facts, we then propose a model of misspecified learning in which agents face two technologies with unknown returns. In learning about the effectiveness of both technologies, the overestimation of the return to the first technology causes an undervaluation and underuse of the second technology. To further test the model, we design a second-phase intervention that distributes leaf color charts to farmers to correct their overestimation of the return to greenness. Consistent with the model prediction, the intervention not only reduces farmers’ nitrogen use immediately, but also induces gradual learning of phosphorus and potassium; the proportion of farmers using phosphorus and potassium both increase by 6 percentage points, relative to 4% and 9% in the control group.

Chapter 2 (coauthored with Binkai Chen and Ao Wang) intends to investigate the causal impact of collegiate economics courses on individual learning and decision-making under a development context. By exploiting a Chinese college-admission system that quasi-randomly assigns students to economics/business majors given students' preferences and the College Entrance Exam's cutoff scores for economics/business majors, we are able to isolate the treatment effects of an economics education on students' responses to a decision-making survey. Specifically, we compare the survey responses of students who narrowly meet the cutoffs for the economics/business majors to those who do not and find that students educated in economics/business courses are more likely to be risk neutral and less prone to common biases in probabilistic beliefs. While students in economics/business majors do not show significant changes in social preferences, they appear more inclined to believe that others behave selfishly.

Chapter 3 is joint with Qianmiao Chen and Shaoda Wang. We investigate how initial land endowments affect household short/long-term decision-makings on labor allocation when land and labor markets are incomplete. The village-level and family size-based land redistribution scheme before 2003 in China exogenously allocated households with differentiate initial land endowment. Using a rich rural fixed point survey data which tracks roughly 20,000 households during 1986 - 2013, we compare the decision-making of households that had population change before and after the last redistribution, the timeline of which is barely predictable by households. The empirical findings suggest that most of households oversupplied labor in agriculture, which led to the marginal product of labor deviate the optimum. Households with fewer initial land endowments decided to input more intense labor in agriculture, which exacerbated the labor misallocation given pre-existing over-employment issues. After forming the land rental market, households with fewer land endowments per capita rented more land, though more land did not influence household agricultural labor allocation. We then examine the impact of labor market reform on alleviating over-supply of labor. The overall agricultural labor input was reduced by 50-70%.

The findings in this dissertation can deepen our understanding of agricultural development and education system in China and shed lights on the research on individual decision-making process in other fields of development economics and other countries.

Traffic data from single loop detectors are one of the dominant data sources widely used in many traffic operation centers and traveler information systems. Speed estimation from single loop detectors is mainly based on occupancy data, a conversion factor from occupancy to density (which is related to vehicle length), and the assumed relationship between flow, speed, and density. This paper investigates the discrepancy between the speed estimated with single loops and the speed measured directly from double loops. It was found that the inaccuracy of speed estimation with single loops is mainly caused by the irregular behavior of vehicle pace. Under congested or unstable traffic conditions, the distribution of vehicle pace within a given time interval often exhibits a large variance accompanied by a strong skewness. Accuracy in speed estimation can be improved by computing occupancy in a different way, using the median vehicle passage time over the detector as opposed to the mean vehicle passage time often used in the conventional method. The performance of the enhanced speed estimation method is very encouraging. The use of the median vehicle passage time reduces the skewness of pace data.

This report documents our validation effort on the Incremental Transfer (IT) model, which is a macroscopic traffic flow model capable of handling freeway systems with special lanes and priority vehicles. The validation study is performed with field data from a 1.7-mile long freeway segment with a congested off-ramp in Oakland, California. In our study, vehicles exiting the off- ramp of this segment are treated as "regular vehicles" and through vehicles as "special vehicles." By assuming that the exiting or "regular" vehicles must stay on the "near" side lanes, we examine the spatial evolution of separate versus coalesced queues developed in the region resulting from a capacity reduction on the exit ramp. The observed queuing pattern is then compared with that predicted by the IT model. Qualitatively, the IT model yields a queuing pattern similar to the one observed in the field data. The model predicts the separate versus coalesced queues in a way that no other macroscopic traffic flow models can do. However, the predictions of the exact queue location are not accurate. The errors may be caused by the lack of sufficient input information required by the model and the difference between the unique traffic characteristics of the test site and model assumptions.