Datasets, features, learning, and models in visual recognition

Visual recognition is a fundamental research topic in computer vision. This dissertation explores datasets, features, learning, and models used for visual recognition. In order to train visual models and evaluate different recognition algorithms, this dissertation develops an approach to collect object image datasets on web pages using an analysis of text around the image and of image appearance. This method exploits established online knowledge resources (Wikipedia pages for text; Flickr and Caltech data sets for images). The resources provide rich text and object appearance information. This dissertation describes results on two datasets. The first is Berg’s collection of 10 animal categories; on this dataset, we significantly outperform previous approaches. On an additional set of 5 categories, experimental results show the effectiveness of the method. Images are represented as features for visual recognition. This dissertation introduces a text-based image feature and demonstrates that it consistently improves performance on hard object classification problems. The feature is built using an auxiliary dataset of images annotated with tags, downloaded from the Internet. Image tags are noisy. The method obtains the text features of an unannotated image from the tags of its k-nearest neighbors in this auxiliary collection. A visual classifier presented with an object viewed under novel circumstances (say, a new viewing direction) must rely on its visual examples. This text feature may not change, because the auxiliary dataset likely contains a similar picture. While the tags associated with images are noisy, they are more stable when appearance changes. The performance of this feature is tested using PASCAL VOC 2006 and 2007 datasets. This feature performs well; it consistently improves the performance of visual object classifiers, and is particularly effective when the training dataset is small. With more and more collected training data, computational cost becomes a bottleneck, especially when training sophisticated classifiers such as kernelized SVM. This dissertation proposes a fast training algorithm called Stochastic Intersection Kernel Machine (SIKMA). This proposed training method will be useful for many vision problems, as it can produce a kernel classifier that is more accurate than a linear classifier, and can be trained on tens of thousands of examples in two minutes. It processes training examples one by one in a sequence, so memory cost is no longer the bottleneck to process large scale datasets. This dissertation applies this approach to train classifiers of Flickr groups with many group training examples. The resulting Flickr group prediction scores can be used to measure image similarity between two images. Experimental results on the Corel dataset and a PASCAL VOC dataset show the learned Flickr features perform better on image matching, retrieval, and classification than conventional visual features. Visual models are usually trained to best separate positive and negative training examples. However, when recognizing a large number of object categories, there may not be enough training examples for most objects, due to the intrinsic long-tailed distribution of objects in the real world. This dissertation proposes an approach to use comparative object similarity. The key insight is that, given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. This dissertation develops a regularized kernel machine algorithm to use this category dependent similarity regularization. Experiments on hundreds of categories show that our method can make significant improvement for categories with few or even no positive examples.

Ecological and evolutionary effects of dispersal on freshwater zooplankton

A recent focus on contemporary evolution and the connections between communities has sought to more closely integrate the fields of ecology and evolutionary biology. Studies of coevolutionary dynamics, life history evolution, and rapid local adaptation demonstrate that ecological circumstances can dictate evolutionary trajectories. Thus, variation in species identity, trait distributions, and genetic composition may be maintained among ecologically divergent habitats. New theories and hypotheses (e.g., metacommunity theory and the Monopolization hypothesis) have been developed to understand better the processes occurring in spatially structured environments and how the movement of individuals among habitats contributes to ecology and evolution at broader scales. As few empirical studies of these theories exist, this work seeks to further test these concepts. Spatial and temporal dispersal are the mechanisms that connect habitats to one another. Both processes allow organisms to leave conditions that are suboptimal or unfavorable, and enable colonization and invasion, species range expansion, and gene flow among populations. Freshwater zooplankton are aquatic crustaceans that typically develop resting stages as part of their life cycle. Their dormant propagules allow organisms to disperse both temporally and among habitats. Additionally, because a number of species are cyclically parthenogenetic, they make excellent model organisms for studying evolutionary questions in a controlled environment. Here, I use freshwater zooplankton communities as model systems to explore the mechanisms and consequences of dispersal and to test these nascent theories on the influence of spatial structure in natural systems. In Chapter one, I use field experiments and mathematical models to determine the range of adult zooplankton dispersal over land and what vectors are moving zooplankton. Chapter two focuses on prolonged dormancy of one aquatic zooplankter, Daphnia pulex. Using statistical models with field and mesocosm experiments, I show that variation in Daphnia dormant egg hatching is substantial among populations in nature, and some of that variation can be attributed to genetic differences among the populations. Chapters three and four explore the consequences of dispersal at multiple levels of biological organization. Chapter three seeks to understand the population level consequences of dispersal over evolutionary time on current patterns of population genetic differentiation. Nearby populations of D. pulex often exhibit high population genetic differentiation characteristic of very low dispersal. I explore two alternative hypotheses that seek to explain this pattern. Finally, chapter four is a case study of how dispersal has influenced patterns of variation at the community, trait and genetic levels of biodiversity in a lake metacommunity.

Investigating wetting characteristics on microstructured surfaces for superhydrophobicity and metal microcasting

The engineering of liquid behavior on surfaces is important for infrastructure, transportation, manufacturing, and sensing. Surfaces can be rendered superhydrophobic by microstructuring, and superhydrophobic devices could lead to practical corrosion inhibition, self-cleaning, fluid flow control, and surface drag reduction. To more fully understand how liquid interacts with microstructured surfaces, this dissertation introduces a direct method for determining droplet solid-liquid-vapor interfacial geometry on microstructured surfaces. The technique performs metrology on molten metal droplets deposited onto microstructured surfaces and then frozen. Unlike other techniques, this visualization technique can be used on large areas of curved and opaque microstructured surfaces to determine contact line. This dissertation also presents measurements and models for how curvature and flexing of microstructured polymers affects hydrophobicity. Increasing curvature of microstructured surfaces leads to decreased slide angle for liquid droplets suspended on the surface asperities. For a surface with regularly spaced asperities, as curvature becomes more positive, droplets suspended on the tops of asperities are suspended on fewer asperities. Curvature affects superhydrophobicity because microscopic curvature changes solid-liquid interaction, pitch is altered, and curvature changes the shape of the three phase contact line. This dissertation presents a model of droplet interactions with curved microstructured surfaces that can be used to design microstructure geometries that maintain the suspension of a droplet when curved surfaces are covered with microstructured polymers. Controlling droplet dynamics could improve microfluidic devices and the shedding of liquids from expensive equipment, preventing corrosion and detrimental performance. This dissertation demonstrates redirection of dynamic droplet spray with anisotropic microstructures. Superhydrophobic microstructured surfaces can be economically fabricated using metal embossing masters, so this dissertation describes casting-based microfabrication of metal microstructures and nanostructures. Low melting temperature metal was cast into flexible silicone molds which were themselves cast from microfabricated silicon templates. The flexibility of the silicone mold permits casting of curved surfaces, which this dissertation demonstrates by fabricating a cylindrical metal roller with microstructures. The metal microstructures can be in turn used as a reusable molding tool. This dissertation also describes an industrial investment casting process to produce aluminum molds having integrated microstructures. Unlike conventional micromolding tools, the aluminum mold was large and had complex curved surfaces. The aluminum was cast into curved microstructured ceramic molds which were themselves cast from curved microstructured rubber. Many structures were successfully cast into the aluminum with excellent replication fidelity, including circular, square, and triangular holes. This dissertation demonstrates molding of large, curved surfaces having surface microstructures using the aluminum mold. This work contributes a more full understanding of the phenomenon of superhydrophobicity and techniques for the economic fabrication of superhydrophobic microstructures.

Beyond hope: Rhetorics of mobility, possibility, and literacy

When writing teachers enter the classroom, they often bring with them a deep faith in the power of literacy to rectify social inequalities and improve their students’ social and economic standing. It is this faith—this hope for change—that draws some writing teachers to locations of social and economic hardship. I am interested in how teachers and theorists construct their own narratives of social mobility, possibility, and literacy. My dissertation analyzes the production and expression of beliefs about literacy in the narratives of a diverse group of writing teachers and theorists, from those beginning their careers to those who are published and widely read. The central questions guiding this study are: How do teachers’ and theorists’ narratives of becoming literate intersect with literacy theories? and How do such literacy narratives intersect with beliefs in the power of literacy to improve individuals’ lives socially, economically, and personally? I contend that the professional literature needs to address more fully how teachers’ and theorists’ personal histories with literacy shape what they see as possible (and desirable) for students, especially those from marginalized communities. A central focus of the dissertation is on how teachers and theorists attempt to resolve a paradox they are likely to encounter in narratives about literacy. On one hand, they are immersed in a popular culture that cherishes narrative links between literacy and economic advancement (and, further, between such advancement and a “good life”). On the other hand, in professional discourse and in teacher preparation courses, they are likely to encounter narratives that complicate an assumed causal relationship between literacy and economic progress. Understanding, through literacy narratives, how teachers and theorists chart a practical path through or around this paradox can be beneficial to literacy education in three ways. First, it can offer direction in professional development and teacher education, addressing how teachers negotiate the boundaries between personal experience, theory, and pedagogy. Second, it can help teachers create spaces wherein students can explore the impact of paradoxical views about the role of literacy on their own lives. Finally, it can offer direction in public policy discourse, extending awareness of what we want—and need—from English language arts education in the twenty-first century. To explore these issues, I draw on case studies and ethnographic observation as well as narrative inquiry into teachers’ and theorists’ published literacy narratives. I situate my findings within three interrelated frames: 1) the narratives of new teachers, 2) the published works of literacy educators and theorists, and 3) my own literacy narrative. My first chapter, “Beyond Hope,” explores the tenuous connections between hope and critique in literacy studies and provides a methodological overview of the study. I argue that scholarship must move beyond a singular focus on either hope or critique in order to identify the transformative potential of literacy in particular circumstances. Analyzing literacy narratives provides a way of locating a critically informed sense of possibility. My second chapter, “Making Teachers, Making Literacy,” explores the intersection between teachers’ lives and the theories they study, based on qualitative analysis of a preservice course for secondary education English teachers. I examine how these preservice English teachers understood literacy, how their narratives of becoming literate and teaching English connected—and did not connect—with theoretical and pedagogical positions, and how these stories might inform their future work as practitioners. Centering primarily on preservice teachers who resisted Nancie Atwell’s pedagogy of possibility because they found it too good to be true, this research concentrates on moments of disjuncture, as expressed in class discussion and in one-on-one interviews, when literacy theories failed to align with aspiring teachers’ understandings of their own experiences and also with what they imagined as possible in disadvantaged educational settings. In my third and fourth chapters, I analyze the narratives of celebrated teachers and theorists who put forth an agenda that emphasizes possibilities through literacy, examining how they negotiate the relationship between their own literacy stories and literacy theories. Specifically, I investigate the narratives of three proponents of critical literacy: Mike Rose, Paulo Freire, and Myles Horton, all highly respected literacy teachers whose working-class backgrounds influenced their commitment to teaching in disenfranchised communities. In chapter 3, “Reading Lives on the Boundary,” I demonstrate how Mike Rose’s 1989 autobiographical text, Lives on the Boundary, juxtaposes rhetorics of mobility with critiques of such possibility. Through an analysis of work published in professional journals, I offer a reception history of Rose’s narrative, focusing specifically on how teachers have negotiated the tension between hope and critique. I follow this analysis with three case studies, drawn from a larger sampling, that inquire into the personal connections that writing teachers make with Lives on the Boundary. The teachers in this study, who provided written responses and participated in audio-recorded follow-up interviews, were asked to compare Rose’s story to their own stories, considering how their personal literacy histories influenced their teaching. My findings illustrate how a group of teachers and theorists have projected their own assessments of what literacy and higher education can and cannot accomplish onto this influential text. In my fourth chapter, “Horton and Freire’s Road as Literacy Narrative,” I concentrate on Myles Horton and Paulo Freire’s 1990 collaborative spoken book, We Make the Road by Walking. Central to my analysis are the educators’ stories about their formative years, including their own primary and secondary education experiences. I argue that We Make the Road by Walking demonstrates how theories of literacy cannot be divorced from personal histories. I begin by examining the spoken book as a literacy narrative that fuses personal and theoretical knowledge, focusing specifically on its authors’ ideas on theory. Drawing on Bakhtin’s notion of the chronotope—the intersection of time and space within narrative—I then explore the literacy narratives emerging from the production process of the book, in a video production about Horton and Freire’s meeting, and ultimately in the two men’s reflections on their childhood years (Dialogic). Interspersed with these accounts is archival material on the book’s editorial production that illustrates the value of increased dialogue between personal history and theories of literacy. My fifth chapter is both a reflective analysis and a qualitative study of my work at a men’s medium-high security prison in Illinois, where I conducted research and served as the instructor of an upper-level writing course, “Writing for a Change,” in the spring of 2009. Entitled “Doing Time with Literacy Narratives,” this chapter explores the complex ways in which literacy and incarceration are configured in students’ narratives as well as my own. With and against students’ stories, I juxtapose my own experiences with literacy, particularly in relation to being the son of an imprisoned father. In exploring the intersections between such stories, I demonstrate how literacy narratives can function as a heuristic for exploring beliefs about literacy between teachers and students both inside and outside of the prison-industrial complex. My conclusion pulls together the various themes that emerged in the three frames, from the making of new teachers to the published literacy narratives of teachers and theorists to my own literacy narrative. Writing teachers encounter considerable pressure to align their curricula with one or another theory of literacy, which has the effect of negating the authority of knowledge about literacy gleaned from experience as readers and writers. My dissertation contends that there is much to be gained by finding ways of articulating theories of literacy that encompass teachers’ knowledge of reading and writing as expressed in personal narratives of literacy. While powerful cultural rhetorics of upward social mobility often neutralize the critical potential of teachers’ own narratives of literacy—potential that has been documented by scholars in writing studies and allied disciplines—this is not always the case. The chapters in this dissertation offer evidence that hopeful and critical positions on the transformational possibilities of literacy are not mutually exclusive.