Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2014. / This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 197-200). / This work is concerned with mRNA processing in mammalian cells and proceeds in two parts. In the first part, I introduce a computational framework for inferring the abundances of mRNA isoforms using high-throughput RNA sequencing data. This framework was applied to study the targets of the ubiquitous splicing factor hnRNP H in human cells. In the second part, I describe an experimental study of the Musashi (hnRNP-like) family of RNA-binding proteins in stem cells and cancer cells, which incorporates computational analyses that rely heavily on the framework developed in part one. In sum, this work provides a computational framework of general use in global analyses of RNA processing and its protein regulators, as well as functional insights into a family of poorly understood RNA-binding proteins. Several related analyses and techniques developed as part of the thesis are described in Appendix A-C. Appendix A describes a study of activity-dependent gene expression and mRNA processing in the mouse olfactory bulb. It uses computational techniques developed in part one of the thesis. Appendix B describes a technique for quantitative visualization of alternative splicing from RNA sequencing data and its integration into a genome browser. Appendix C describes a method for clonal analysis of neural stem cell growth and differentiation in culture using live imaging and `microdot' plates, developed as part of the work presented in part one of the thesis. / by Yarden Katz. / Ph. D.

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2015. / This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 167-176). / This thesis develops probabilistic programming as a productive metaphor for understanding cognition, both with respect to mental representations and the manipulation of such representations. In the first half of the thesis, I demonstrate the representational power of probabilistic programs in the domains of concept learning and social reasoning. I provide examples of richly structured concepts, defined in terms of systems of relations, subparts, and recursive embeddings, that are naturally expressed as programs and show initial experimental evidence that they match human generalization patterns. I then proceed to models of reasoning about reasoning, a domain where the expressive power of probabilistic programs is necessary to formalize our intuitive domain understanding due to the fact that, unlike previous formalisms, probabilistic programs allow conditioning to be represented in a model, not just applied to a model. I illustrate this insight with programs that model nested reasoning in game theory, artificial intelligence, and linguistics. In the second half, I develop three inference algorithms with the dual intent of showing how to efficiently compute the marginal distributions defined by probabilistic programs, and providing building blocks for process-level accounts of human cognition. First, I describe a Dynamic Programming algorithm for computing the marginal distribution of discrete probabilistic programs by compiling to systems of equations and show that it can make inference in models of "reasoning about reasoning" tractable by merging and reusing subcomputations. Second, I introduce the setting of amortized inference and show how learning inverse models lets us leverage samples generated by other inference algorithms to compile probabilistic models into fast recognition functions. Third, I develop a generic approach to coarse-to-fine inference in probabilistic programs and provide evidence that it can speed up inference in models with large state spaces that have appropriate hierarchical structure. Finally, I substantiate the claim that probabilistic programming is a productive metaphor by outlining new research questions that have been opened up by this line of investigation. / by Andreas Stuhlmüller. / Ph. D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 1996. / Includes bibliographical references (leaves 138-150). / by Francesca Gandolfo. / Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, February 2011. / Cataloged from PDF version of thesis. / Includes bibliographical references (p. 97-102). / Successful interaction with the external world requires choosing appropriate actions in the context of available choices. Such decisions require the evaluation of the reward magnitude, or value, associated with each potential action. Delineating the neural circuits involved in this process remains an important goal in systems neuroscience. However, little is known about the neural circuits that compute, or represent, low level primary reward signals. We have combined quantitative psychophysical measures of subjective reward magnitude elicited by rewarding electrical brain stimulation, fMRI as a readout of whole-brain neural activity, and local inactivation of brain structures, to identify the neural representation of subjective reward magnitude. We find that multiple brain regions are activated by rewarding brain stimulation, but only two brain regions, the nucleus accumbens and the central and basolateral nucleus of the amygdala, exhibit patterns of activity levels that track the reward magnitude measured psychophysically, suggesting a role in the neural representation of reward magnitude. Furthermore, pharmacological silencing of the ventral tegmental area (VTA) disrupts reward-tracking behavior and increases stimulus-dependent activity in the nucleus accumbens and amygdala. Together these data suggest that ascending and descending pathways combine to produce a signal that ultimately guides behavior and is subject to modulation by VTA inputs. / by Ana Marcia Fiallos. / Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2011. / Cataloged from PDF version of thesis. / Includes bibliographical references (p. 117-128). / Every object in the world has a physical size which is intrinsic to how we interact with it: we pick up small objects like coins with our fingers, we throw footballs and swing tennis rackets, we orient our body to bigger objects like chairs and tables and we navigate with respect to landmarks like fountains and buildings. Here I argue that the size of objects in the world is a basic property of object representation with both behavioral and neural consequences. Specifically, I suggest that objects have a canonical visual size based on their real-world size (Chapter 2), and that we automatically access real-world size information when we recognize an object (Chapter 3). Further, I present evidence that there are neural consequences of realworld size for the large-scale organization of object knowledge in ventral visual cortex (Chapter 4). Specifically, there are regions with differential selectivity for big and small objects, that span from along the dorsal and lateral surfaces of occipito-temporal cortex in a mirrored organization. Finally, I suggest that the empirical findings can be coherently explained by thinking about the experience of an observer situated in a three-dimensional world. This work provides testable predictions about retinal size biases in visual experience, and an approach in which to understand the neural representation of any object in the world. / by Talia Konkle. / Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2003. / Includes bibliographical references. / Face recognition is one of the most important problems our visual system must solve. Here I used magnetoencephalography (MEG) in an effort to characterize the sequence of cognitive and neural processes underlying this remarkable ability. This work is designed to answer several questions. First, how long does it take for the human visual system to recognize a stimulus as a face? Second, what are the stages of processing in face perception? Finally, what is the nature of representations extracted at each of these stages? MEG provides an ideal tool for addressing these questions, as its high temporal resolution enables us to separately measure perceptual operations that may occur only a few tens of milliseconds apart from each other. Yet, unlike single-unit recording, it can be used in normal human subjects. Three new findings about human face recognition will be reported in this thesis. First, a face stimulus begins to be categorized as a face within 100 ms after stimulus onset in humans, substantially faster than previously thought. Second, face recognition occurs in two distinct stages: an initial stage at which the stimulus is categorized as a face, and a stage that occurs 70 ms later at which the individual identity of the face is extracted. Finally, the representations extracted at these two stages differ not only in specificity, but also in the aspects of a face represented at each stage. / by Jia Liu. / Ph.D.

Thesis (S.B.)--Massachusetts Institute of Technology, Dept. of Physics; and, (S.B.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2004. / Includes bibliographical references (p. 21-22). / It has been assumed that Ca2+ influx of different duration and amplitude would generate different level of potentiation. The conventional protocols of generating LTP have been 1. tetanic stimulation of presynaptic cell, 2. theta burst stimulation of presynaptic cell, and 3. correlated stimulation of pre- and post-synaptic cells. However, the effects of different Ca2+ influx can not be precisely dissected with the conventional protocols for the following defects: 1. the protocols do not discriminate between pre- and post-synaptic side plasticity, 2. the protocols observe synaptic plasticity between two cells which involve multiple synapses with heterogeneous properties, 3. precise control and measurement of the amount of Ca2+ influx are not possible in the protocols. In the present experiment, we perfused glutamate directly on to a single postsynaptic site, depolarized the postsynaptic intracellular potential to a controlled voltage for a controlled duration of time, thus controlling the opening of postsynaptic NMDA receptors and Ca2+ influx. By using this method, we found 1. that modification of synaptic strength has a bell-shaped dependency to the amount of Ca2+ influx, 2. that weak Ca2+ current through desensitized NMDA receptors sustained for a long period of time (160 ms) generates LTD, 3. evidence that phosphorylation of AMPAR leads to insertion of AMPAR. / by Dongsung Huh. / S.B.

Thesis (Ph. D. in Computational Neuroscience)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2003. / Includes bibliographical references (leaves 157-163). / The design and use of expert systems for medical diagnosis remains an attractive goal. One such system, the Quick Medical Reference, Decision Theoretic (QMR-DT), is based on a Bayesian network. This very large-scale network models the appearance and manifestation of disease and has approximately 600 unobservable nodes and 4000 observable nodes that represent, respectively, the presence and measurable manifestation of disease in a patient. Exact inference of posterior distributions over the disease nodes is extremely intractable using generic algorithms. Inference can be made much more efficient by exploiting the QMR-DT's unique structure. Indeed, tailor-made inference algorithms for the QMR-DT efficiently generate exact disease posterior marginals for some diagnostic problems and accurate approximate posteriors for others. In this thesis, I identify a risk with using the QMR-DT disease posteriors for medical diagnosis. Specifically, I show that patients and physicians conspire to preferentially report findings that suggest the presence of disease. Because the QMR-DT does not contain an explicit model of this reporting bias, its disease posteriors may not be useful for diagnosis. Correcting these posteriors requires augmenting the QMR-DT with additional variables and dependencies that model the diagnostic procedure. I introduce the diagnostic QMR-DT (dQMR-DT), a Bayesian network containing both the QMR-DT and a simple model of the diagnostic procedure. Using diagnostic problems sampled from the dQMR-DT, I show the danger of doing diagnosis using disease posteriors from the unaugmented QMR-DT. / (cont.) I introduce a new class of approximate inference methods, based on feed-forward neural networks, for both the QMR-DT and the dQMR-DT. I show that these methods, recognition models, generate accurate approximate posteriors on the QMR-DT, on the dQMR-DT, and on a version of the dQMR-DT specified only indirectly through a set of presolved diagnostic problems. / by Quaid Donald Jozef Morris. / Ph.D.in Computational Neuroscience

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2016. / Cataloged from PDF version of thesis. / Includes bibliographical references (pages 113-132). / Alzheimer's disease (AD) causes a devastating loss of memory and cognition for which there is no cure. Without effective treatments that slow or reverse the course of the disease, the rapidly aging population will require astronomical investment from society to care for the increasing numbers of AD patients. Additionally, the financial and emotional burden on families of affected individuals will be profound. Traditional approaches to the study of AD use either biochemical assays to probe cellular pathophysiology or non-invasive imaging platforms to investigate brain-wide network alterations. Though decades of research using these tools have advanced the field significantly, our increased understanding of AD has not led to successful interventions. One reason for this impediment may be that the tools used in neither approach can achieve the spatial and temporal precision necessary to study the consequences of molecular insults across the brain over time. In this thesis, I capitalize on recent advances in tissue processing technologies to gain a network-level perspective on the molecular and cellular progression of AD. First, I present optimized methods for in situ proteomic phenotyping of large-volume tissue specimens. Then, I use the techniques to map amyloid-beta (A[beta]) aggregates at the whole-brain scale across disease stages in a mouse model of AD. The spatially-unbiased, temporally-precise map demonstrates hierarchical susceptibility of increasingly large, memory-related brain networks to A[beta] deposition. Importantly, the 4D nature of the map reveals that subcortical nodes and white matter tracts of the Papez memory circuit exhibit unique, early vulnerability to A[beta] aggregates. Finally, using large-volume labeling approaches, I confirm the molecular findings by showing disease-specific A[beta] aggregation in human samples from the early hub regions. Together, this data unites desperate observations of network-level deficits and identifies critical locations of early A[beta] deposition in the brain. By linking molecular and network observations, I begin to provide biological explanations for the clinical manifestation of AD. This perspective can guide earlier patient identification and refine experimental approaches to developing cognitively efficacious treatments. These discoveries emphasize the necessity of multi-level investigations in neuroscience research and highlight the potential impacts of tools that enable researchers to bridge the gap. / by Rebecca Gail Canter. / Ph. D.

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2018. / This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 159-177). / Song learning circuitry is thought to operate using a unique representation of each moment within each song syllable. Distinct timestamps for each moment in the song have been observed in the premotor cortical nucleus HVC, where neurons burst in sparse sequences. However, such sparse sequences are not present in very young birds, which sing highly variable syllables of random lengths. Furthermore, young birds learn by imitating a tutor song, and it was previously unclear precisely how the experience of hearing a tutor might shape auditory, motor, and evaluation pathways in the songbird brain. My thesis presents a framework for how these pathways may assemble during early learning, using simple neural mechanisms. I start with a neural network model for how premotor sequences may grow and split. This model predicts that the sequence-generating nucleus HVC would receive rhythmically patterned training inputs. I found such a signal when I recorded neurons that project to HVC. When juvenile birds sing, these neurons burst at the beginning of each syllable, and when the birds listen to a tutor, neurons burst at the rhythm of the tutor's song. Bursts marking the beginning of every tutor syllable could seed chains of sequential activity in HVC that could be used to generate the bird's own song imitation. I next used functional calcium imaging to characterize HVC sequences before and after tutor exposure. Analysis of these datasets led us to develop a new method for unsupervised detection of neural sequences. Using this method, I was able to observe neural sequences even prior to tutor exposure. Some of these sequences could be tracked as new syllables emerged after tutor exposure, and some sequences appeared to become coupled to the new syllables. In light of my new data, I expand on previous models of song learning to form a detailed hypothesis for how simple neural processes may perform song learning from start to finish. / by Emily Lambert Mackevicius. / Ph. D.