Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2018 / Cataloged from PDF version of thesis. Page 140 blank. / Includes bibliographical references (pages 134-139). / The strategies infants and young children use to understand the world around them provide unique insight into the structure of human cognition. However, developmental research is subject to heavy pragmatic constraints on recruiting large numbers of participants, bringing families back for repeat sessions, and working with special populations or diverse samples. These constraints limit the types of questions that can be addressed in the lab as well as the quality of evidence that can be obtained. In this dissertation, I present a new platform, "Lookit," that allows researchers to conduct developmental experiments online via asynchronous webcam-recorded sessions, with the aim of expanding the set of questions that we can effectively answer. I first present the results of a series of empirical studies conducted in the laboratory to assess difficulty faced by infants in integrating information across visual hemifields (Chapter 2), as an illustration of the creative workarounds in study design necessary to accommodate the difficulty of participant recruitment. The rest of this work concerns the development of the online platform, from designing the prototype (Chapter 3) and initial proof-of-concept studies (Chapter 4) to the demonstration of an interface for researchers to specify and manage their studies on a collaborative platform (Chapter 5). I show that we are able to reliably collect and code dependent measures including looking times, preferential looking, and verbal responses on Lookit; to work with more representative samples than in the lab; and to flexibly implement a wide variety of study designs of interest to developmental researchers. / by Kimberly M. Scott. / Ph. D. / Ph.D. Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

Thesis: Ph. D. in Neuroscience, Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, May, 2020 / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 213-232). / Whether it is speaking to one another, or nailing a tennis serve, humans can perform an incredible range of behaviors, most of which are learned. How do we and other animals learn complicated sequential behaviors, and once they are learned how are they executed? This thesis is an investigation into the neural basis of the two modes of behavior that occur at the beginning and end of learning a motor skill: the initially highly variable exploratory behavior, and the ultimately stereotyped skilled performance. To understand the start and end points of learned motor behaviors, I present two studies, each on the premotor activity of ensembles of neurons that underlie song production in zebra finches. Executing learned motor behaviors requires animals to produce precisely timed motor sequences. / While cortical motor regions traditionally have been viewed as encoding features of motor gestures (Georgopoulos et al., 1982), more recent studies have suggested that motor regions may have intrinsic dynamics to pattern the production of motor gestures (Churchland et al., 2012). A similar debate has arisen in songbirds. Adult birdsong requires the premotor nucleus HVC (used as a proper noun), in which projection neurons burst sparsely at stereotyped times in the song. It has been hypothesized that projection neuron bursts, as a population, form a continuous sequence, while a different model of HVC function proposes that HVC activity is tightly organized around motor gestures. Using a large dataset of HVC neurons recorded in singing birds, we test several predictions of these models. We find that projection neuron bursts in adult birds are continuously and nearly uniformly distributed throughout song. / Another model posits that LMAN may act as an excitable media producing locally propagating waves of activity, and predicts that all nearby pairs of neurons would be highly correlated. To test these models and to understand how LMAN actively generates behavioral variability, we built a miniature lightweight microdrive to simultaneously record from multiple neurons, as well as a lightweight endoscope to perform functional calcium imaging of ensembles of LMAN neurons. With these new technologies, we observed the simultaneous activity of pairs of single units in singing juvenile and adult birds. We find that most pairs of neurons with small separation (<250 μm) are completely uncorrelated, which is incompatible with the wave model. However, a small subset of pairs have strikingly large correlations, with correlation coefficients of up to 0.81. Intriguingly, these correlated pairs of neurons can be separated by up to 400 μm. / The existence of such highly correlated neurons within LMAN is inconsistent with LMAN being a simple balanced excitatory-inhibitory network with uniformly random connectivity. These results suggest that new models of variability generation are required to explain how LMAN generates exploratory behavioral variability. / by Galen Lynch. / Ph. D. in Neuroscience / Ph.D.inNeuroscience Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Thesis: Ph. D. in Neuroscience, Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, June, 2019 / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references. / There is an unbreakable link between shape and function. In biology, the architecture of cells, tissues and organisms, that have evolved adapting to the world around them, translate into specific functional outcomes. Self-organization is an adaptive, non-linear and dynamic process, where diverse ordered patterns emerge from an initially disordered and noisy state through local interactions between the elements of a system. This can lead to the fascinating biological diversity and functional complexity in such systems. Unwavering storms on the surface of Jupiter, patterns on the wing of a butterfly, a regenerating planarian eye, development of a neuronal circuit in the human brain can all be studied systematically using the conceptual tools derived from the field of self-organization. Here, I sought to address a central, but understudied, problem in animal regeneration: How do regenerative progenitors organize into complex replacement structures in the context of adult anatomy? I used the planarians as a system for studying regenerative progenitors and focused on eye regeneration to elucidate the mechanisms. I found that self-organization has a major role in determining the behavior of regenerative progenitors. This work revealed three properties that govern regenerative progenitor behavior, and these three properties in concert explain many previously mysterious aspects of how regeneration works: (i) self-organization, (ii) an extrinsic migratory target for progenitors, and (iii) a broad progenitor specification zone that allows progenitors to be targeted into self-organizing systems even if they are transiently in incorrect locations during the process of regeneration. These components yield a model with broad explanatory and predictive power. As an example, we were able to generate wild-type animals with 3, 4, or 5 eyes instead of 2 by simple manipulations of the system using the model developed. Remarkably, the extra eyes were stably maintained throughout the life of the animal, resulting in wild-type animals with an alternative and stable anatomical state. This model prominently incorporates self-organizing principles, which have been little explored in regeneration. The new conceptual model with broad explanatory power allowed us to address some of the fundamental previous mysteries of regeneration. / by Kutay Deniz Atabay. / Ph. D. in Neuroscience / Ph.D.inNeuroscience Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Thesis: Ph. D. in Computation, Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, June, 2019 / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 155-167). / Neuronal connectivity can be reconstructed from a 3D electron microscopy (EM) image of a brain volume. A challenging and important subproblem is the segmentation of the image into neurons. For the past decade, convolutional networks have been used for 3D reconstruction of neurons from EM brain images. In this thesis, we develop a set of deep learning algorithms based on convolutional nets for automated reconstruction of neurons, with particular focus on highly anisotropic images of brain tissue acquired by serial section EM (ssEM). In the first part of the thesis, we propose a recursively trained hybrid 2D-3D convolutional net architecture, and demonstrate the feasibility of exploiting 3D context to further improve boundary detection accuracy despite the high anisotropy of ssEM images. In the following parts, we propose two techniques for training convolutional nets that can substantially improve boundary detection accuracy. First, we introduce novel forms of training data augmentation based on simulation of known types of image defects such as misalignments, missing sections, and out-of-focus sections. Second, we add the auxiliary task of predicting affinities between nonneighboring voxels, reflecting the structured nature of neuronal boundary detection. We demonstrate the effectiveness of the proposed techniques on large-scale ssEM images acquired from the mouse primary visual cortex. Lastly, we take a radical departure from simple boundary detection by exploring an alternative approach to object-centered representation, that is, learning dense voxel embeddings via deep metric learning. Convolutional nets are trained to generate dense voxel embeddings by assigning similar vectors to voxels within the same objects and well-separated vectors to voxels from different objects. Our proposed method achieves state-of-the-art accuracy with substantial improvements on very thin objects. / by Kisuk Lee. / Ph. D. in Computation / Ph.D.inComputation Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, February, 2020 / Manuscript. / Includes bibliographical references (pages 227-238). / This work utilizes theoretical approaches to answer the question: which functions grid and place cells perform that directly lead to their own emergence? To answer such a question, an approach that goes beyond a simple modelling is necessary given the fact that there could be circuit solutions other than grid or place cells that better perform these functions. With this reasoning, I adopted a systematic guideline that aims for an optimization principle attempting to find the optimal solution for performing the hypothesized functions while reproducing the correct phenomenology. Within the optimization principle framework, I applied both recurrent neural network (RNN) training and coding-theoretic approaches to set up appropriate optimization problems for testing a given function hypotheses. The descriptive function hypotheses: 1) Grid cells exist for having a high-capacity and robust path-integrating code and 2) Place cells exist for having a sequentially-learnable and highly-separable path-integrating code were adopted. The non-converging performance in training an RNN to perform a hard navigation task suggests that the attractor dynamics forbids a network to simultaneously possess online learnability and high coding capacity. Because of this dynamical constraint in learning, a grid cell circuit has to be hardwired through some developmental process and cannot be easily modified by an experience-based synaptic rule without compromising its capacity. On the contrary, a place cell circuit being able to continually learn a novel environment inevitably have a mere linear capacity. These results imply that the functional separation of grid and place cell systems observed in the brain could be a result of an unavoidable dynamical constraint from their underlying RNNs. Lastly, a fundamental principle called the tuning-learnability correspondence was uncovered in pursuit of a sequentially learnable neural implementation for place cells. It explains that the seemingly incidental existence of conjunctive tuning property is in fact caused by a necessary metastable attractor dynamics for having sequential learnability rather than by another functional need attached to a particular tuning property. In addition, from the unique property of metastable attractor dynamics, I also predicted that the biased place field propensity recently observed in CA1 sub-region should originate from CA3 due to an inevitable biased activation in the RNN as a side effect of such a dynamical property. In sum, both this principle and the subsequent prediction thus provide a new perspective that contradicts the conventional wisdom which often assumed that a certain nonspatial tuning property exists for performing a relevant task. / by Tzuhsuan Ma. / Ph. D. / Ph. D. Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, February, 2020 / Manuscript. / Includes bibliographical references (pages 81-89). / The highest-performing natural language processing models generally solve language tasks by deriving statistical regularities of sequences of arbitrary tokens supplied as training data. Humans have a much richer notion of language, however. For one thing, they understand that language refers to objects aid actions in the real world, which enables them to use language to efficiently transmit instructions on how to accomplish goals. For another, they learn to focus their attention on only those spans of text important for accomplishing the task at hand. ăIn this thesis, we attempt to improve machine models of language by taking inspiration from these aspects of human language. The first half of this thesis concerns understanding instructional "how-to" language, such as "Add remaining flour. Then mix." The meaning is ambiguous without context: Add how much flour to what? Mix what, using what tools, until when? We show how to successfully parse this language by maintaining a distribution over the state of a theoretical kitchen as the instructions are parsed. We also show how to aid interpretation if videos of the task are also available by training a joint vision-language model with over 300,000 Youtube videos on how to cook. The second half discusses taking advantage of people's ability to focus on important parts of a passage in a multiple-choice reading comprehension task to enhance the performance of an automatic question-answering system. We record the gaze location of hundreds of subjects as they read and answer questions about newspaper articles. We then train a state-of-the-art transformer model to predict human attention as well correct answers and find this leads to a substantial boost in performance over merely training the model to predicting correct answers. / by Jonathan Matthew Malmaud. / Ph. D. / Ph. D. Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, February, 2020 / Manuscript. Vita. / Includes bibliographical references. / A subset of children with autism spectrum disorder (ASD) exhibit temporary but significant improvement of their behavioral symptoms during episodes of fever. Fever is an increase body temperature that is typically the product of the immune response mounted in order to fight infection. Investigation into the curious connection between fever and the manifestation of ASD behavioral phenotypes is at its infancy and therefore the mechanisms underlying this phenomenon are unknown. Here, we attempt to understand the neural and molecular mechanisms mediating this connection. The etiology of ASD is heterogeneous and is composed of both environmental and genetic risk factors. Therefore, we compared an environmental model of neurodevelopmental disorders in which mice were exposed to maternal immune activation (MIA) during embryogenesis with mice genetically deficient for Cntnap2, Fmr1, and Shank3. We found that activation of the immune system using lipopolysaccharide (LPS) was sufficient to rescue behavioral deficits within the MIA model, but not the monogenic mutant model mice. Behavioral rescue was correlated with reduced hyperactivation in the primary somatosensory cortex dysgranular zone (S1DZ), a region that has been previously shown to be tightly linked to MIA behavioral phenotypes. Consistent with the selective effects of LPS within MIA offspring, this reduction in hyperactivation was unique to MIA offspring. Differences in response to LPS could be explained by reduced IL-17a induction in monogenic mutant mice compared with MIA offspring. Circumventing this difference by directly administering IL-17a into the brain was sufficient to promote sociability in MIA offspring as well as monogenic mutant mice. IL-17a signaling is shown to be critical for the LPS-induced behavioral rescue and reduction in hyperactivity. Genetic removal of its cognate receptor, IL-17Ra selectively within the S1DZ, similarly prevented the ability of LPS to rescue MIA sociability deficits. These data support a model in which IL-17a signaling within the S1DZ mediates the behavioral and neurophysiological effects of immune activation in the MIA model. / by Michael Douglas Reed. / Ph. D. / Ph. D. Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

Thesis: S.M. in Neuroscience, Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, February, 2020 / Manuscript. / Includes bibliographical references (pages 49-57). / Activity-dependent changes in neuronal properties (neuronal plasticity) are critical for information processing and storage in the brain. It is well-established that protein synthesis is essential for both memory formation and the long-lasting changes in synaptic strength that accompany learning. However, it's still unclear when protein synthesis needs to occur relative to the experience to form durable memories, and the identities and roles of crucial proteins in these processes have not been elucidated. Neurogranin, a small protein that regulates calcium-dependent signaling, is poised to modulate both memory and synaptic plasticity. This thesis aims to provide insights into the molecular underpinnings mediating context memory formation in the hippocampus. By combining molecular, behavioral, pharmacological, and viral manipulations, we assessed the role of neurogranin in hippocampal memory formation and synaptic plasticity. We observed a rapid, activity-dependent upregulation of neurogranin mediated by FMRP. Neurogranin was found to be regulated by the adrenergic system, and our data suggested a role in the adrenergic-mediated enhancement in memory formation and a form of synaptic plasticity known as long-term potentiation. These findings strongly suggest that neurogranin plays an important role in regulating memory and synaptic plasticity. / by Sebastian Templet. / S.M. in Neuroscience / S.M. in Neuroscience Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, February, 2020 / Manuscript. / Includes bibliographical references (pages 139-149). / Human reasoning is complex, messy, and approximate, and as a result, has been subject to a millennia-long enterprise to extract principles that are simple, neat, and impeccable. This enterprise is incomplete; there are acts of reasoning that humans perform everyday, often effortlessly, that remain both poorly understood and beyond the capabilities of modern methods of artificial intelligence. Specifically, humans understand the causal structure of the world, and mentally manipulate it to imagine worlds that could have been but were not, and even worlds that could never exist in reality. This thesis investigates computational principles of imaginative reasoning; develops programming languages to express the knowledge upon which imaginative reasoning relies, and upon this foundation introduces practical algorithms of automatic inference. Concretely, we introduce probabilistic programming languages - which encode causal probabilistic models as programs - with two new forms of inference. The first is distributional inference, which means to reason with statistical information rather than observational data. This allows us for instance to address problems of algorithmic fairness, robustness, and perform parameter estimation using data about probabilities, expectations and other distributional properties. The second is causal inference, which allows us in complex simulation models to reason about counterfactual what-if scenarios, as well as causation, i.e., whether some event A is the cause of some other event B. To perform inference, we introduce a number of new algorithms. Unlike traditional methods, these modify the internal structure of the model or reinterpret how it is executed. We introduce parametric inversion, which inverts the causal structure to literally run programs in reverse from observations to causes, and predicate exchange, which relaxes Boolean operators to make inference more tractable. Collectively, these contributions shrink the gap between human and machine reasoning, as well as serve as practical tools for scientific modelling and inference. / by Zenna Tavares. / Ph. D. / Ph. D. Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences

Thesis: S.M. in Cognitive Science, Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2018 / Cataloged from PDF version of thesis. / Includes bibliographical references (pages 48-51). / This thesis investigates the first language acquisition of scalar inferences from Korean particle -cocha 'even.' Based on the fact that also evokes the same existential inference with even, and that also and even have the same focus scope in Korean, this thesis compares the acquisition of Korean -cocha 'even' with -to 'also' to provide a more elaborated explanation of scalar inference acquisition. Three experiments - one felicity judgement task, and two preference tasks - were conducted to answer the following research questions: i) when Korean-speaking children are able to make scalar inferences from -cocha 'even'; ii) whether Korean-speaking children are able to correctly assign the scope of -to 'also' and -cocha 'even' to the subject or the object; iii) which step of the even scalar inference process causes children's difficulty. As a result, it was found that Korean-speaking children are able to draw existential inferences at the age of 9 to 10, but still have difficulty in making scalar inferences from even. Next, Korean-speaking children had difficulty in correctly assigning the scope of also and even to the subject or the object even though Korean focus particles are not governed by the c-command rule. Additionally, presenting an alternative phrase facilitated children's process of scalar inferences, as the reference-set hypothesis predicts. Finally, children even at the age of 3 and 4 had the cognitive ability to arrange the elements of a set according to probability. In conclusion, children do not have the semantic ability to associate even with 'the lowest probability,' and syntactic ability to find what is focused by even. Consequently, children cannot create a set containing the focused phrase and alternative phrases although they already have the cognitive ability to compute probability and arrange the elements of the set in order of probability. / by Soyeon Kang. / S.M. in Cognitive Science / S.M.inCognitiveScience Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences