Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2018. / Cataloged from PDF version of thesis. / Includes bibliographical references (pages 199-211). / This thesis develops formal computational cognitive models of the social intelligence underlying human cooperation and morality. Human social intelligence is uniquely powerful. We collaborate with others to accomplish together what none of us could do on our own; we share the benefits of collaboration fairly and trust others to do the same. Even young children work and play collaboratively, guided by normative principles, and with a sophistication unparalleled in other animal species. Here, I seek to understand these everyday feats of social intelligence in computational terms. What are the cognitive representations and processes that underlie these abilities and what are their origins? How can we apply these cognitive principles to build machines that have the capacity to understand, learn from, and cooperate with people? The overarching formal framework of this thesis is the integration of individually rational, hierarchical Bayesian models of learning, together with socially rational multi-agent and game-theoretic models of cooperation. I use this framework to probe cognitive questions across three time-scales: evolutionary, developmental, and in the moment. First, I investigate the evolutionary origins of the cognitive structures that enable cooperation and support social learning. I then describe how these structures are used to learn social and moral knowledge rapidly during development, leading to the accumulation of knowledge over generations. Finally I show how this knowledge is used and generalized in the moment, across an infinitude of possible situations. This framework is applied to a variety of cognitively challenging social inferences: determining the intentions of others, distinguishing who is friend or foe, and inferring the reputation of others all from just a single observation of behavior. It also answers how these inferences enable fair and reciprocal cooperation, the computation of moral permissibility, and moral learning. This framework predicts and explains human judgment and behavior measured in large-scale multi-person experiments. Together, these results shine light on how the scale and scope of human social behavior is ultimately grounded in the sophistication of our social intelligence. / by Max Kleiman-Weiner. / Ph. D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2006. / Includes bibliographical references (p. 141-148). / This thesis studies two key properties of learning algorithms: their generalization ability and their stability with respect to perturbations. To analyze these properties, we focus on concentration inequalities and tools from empirical process theory. We obtain theoretical results and demonstrate their applications to machine learning. First, we show how various notions of stability upper- and lower-bound the bias and variance of several estimators of the expected performance for general learning algorithms. A weak stability condition is shown to be equivalent to consistency of empirical risk minimization. The second part of the thesis derives tight performance guarantees for greedy error minimization methods - a family of computationally tractable algorithms. In particular, we derive risk bounds for a greedy mixture density estimation procedure. We prove that, unlike what is suggested in the literature, the number of terms in the mixture is not a bias-variance trade-off for the performance. The third part of this thesis provides a solution to an open problem regarding the stability of Empirical Risk Minimization (ERM). This algorithm is of central importance in Learning Theory. / (cont.) By studying the suprema of the empirical process, we prove that ERM over Donsker classes of functions is stable in the L1 norm. Hence, as the number of samples grows, it becomes less and less likely that a perturbation of o(v/n) samples will result in a very different empirical minimizer. Asymptotic rates of this stability are proved under metric entropy assumptions on the function class. Through the use of a ratio limit inequality, we also prove stability of expected errors of empirical minimizers. Next, we investigate applications of the stability result. In particular, we focus on procedures that optimize an objective function, such as k-means and other clustering methods. We demonstrate that stability of clustering, just like stability of ERM, is closely related to the geometry of the class and the underlying measure. Furthermore, our result on stability of ERM delineates a phase transition between stability and instability of clustering methods. In the last chapter, we prove a generalization of the bounded-difference concentration inequality for almost-everywhere smooth functions. This result can be utilized to analyze algorithms which are almost always stable. Next, we prove a phase transition in the concentration of almost-everywhere smooth functions. Finally, a tight concentration of empirical errors of empirical minimizers is shown under an assumption on the underlying space. / by Alexander Rakhlin. / Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2007. / "June 2007." / Includes bibliographical references. / The mammalian superior colliculus (SC) coordinates visual, somatosensory, and auditory stimuli to guide animal behavior. The superficial layers (sSC) receive visual information via two major afferent projections: 1) A direct retinal projection and 2) an indirect projection from Layer V visual cortex. The retinal projection reaches the rat sSC by embryonic day 16, is topographic, and refines to form a high resolution map of visual space early in development, before eye-opening in rodents (-P12-P14). The cortical projection is delayed by about eight days, just reaching the sSC around P4, and does not complete its topographic refinement until around the time of eye-opening. These afferents compete for synaptic space during a time when patterns of spontaneous and evoked activity are rapidly changing. I have used the mouse sSC as a model system to test the role of new activity patterns due to the initial onset of visual experience after eye-opening in visual synaptic development. I have described the organization of retinal and cortical afferents and the laminar organization of the mouse sSC in Chapter 3. Previous work demonstrated eye-opening (EO) induces the appearance of dendritic PSD-95 and LTP in the sSC within 2-4 hours. / (cont.) I provide evidence that EO-induced PSD-95 trafficking is required for the stabilization of new synapses in vivo as a result of patterned visual experience after eye-opening. mEPSC frequency recorded in a vertical neuronal subtype of the mid-SGS increases at least three-fold after eye-opening, indicating a rapid synaptogenesis that does not occur in PSD95KO mice, or in age-matched littermates deprived of initial visual experience. A structural analysis of these neurons revealed caliber-specific patterns of spine and filopodia development that depend on EO and the projection from visual cortex. Between P11 and P13, dendrites post-synaptic to cortical axons undergo an EO-independent tripling of filopodial density and an EO-dependent maintenance of dendritic spine density. These data suggest that rapid vision-induced trafficking of PSD-95 enables long-term potentiation and stabilization of newly formed cortico-collicular synapses in response to patterned visual stimuli. Furthermore, these data suggest that cortical inputs are sensitive to pattern vision deprivation between P12 and P13, but retinal inputs are not. / by Marnie A. Phillips. / Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2009. / Cataloged from PDF version of thesis. / Includes bibliographical references. / Human observers are able to rapidly and accurately categorize natural scenes, but the representation mediating this feat is still unknown. Here we propose a framework of rapid scene categorization that does not segment a scene into objects and instead uses a vocabulary of global, ecological properties that describe spatial and functional aspects of scene space (such as navigability or mean depth). In Chapter 1, four experiments explore the human sensitivity to global properties for rapid scene categorization, as well as the computational sufficiency of these properties for predicting scene categories. Chapter 2 explores the time course of scene understanding, finding that global properties can be perceived with less image exposure than the computation of a scene's basic-level category. Finally, in Chapter 3, I explore aftereffects to adaptation to global properties, showing that repeated exposure to many global properties produces robust high-level aftereffects, thus providing evidence for the neural coding of these properties. Altogether, these results provide support for the hypothesis that rapid categorization of natural scenes may not be mediated primarily though objects and parts, but also through global properties of structure and affordance. / by Michelle R. Greene. / Ph.D.

Thesis (S.M.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2004. / Includes bibliographical references (leaves 60-66). / The Wilson-McNaughton tetrode preparation for awake, behaving rodents was adapted by a group of investigators for use in dorsolateral striatum. Measures were taken to improve the reliability of reaching the target area in the brain and the stability of the implanted tetrode drive over several weeks of recording. Novel methods were developed to confirm the dorsoventral level of tetrodes at intermediate stages of advancement during recording, and to reconstruct estimated directions and distances of recorded sources from tetrodes by post-hoc analysis. Alternative methods of source separation and data visualization were implemented. Additional refinements to improve unit separation within and across recording sessions are proposed. The resulting recording technique is expected to have considerable potential in clarifying behavioral and other functional correlates of systems of striatal anatomical compartmentalization. A set of experiments is proposed to investigate how dorsolateral striatal neuronal activity changes in correlation with learning of three stimulus-response tasks relative to three control tasks with similar sensory, motor and motivational aspects but different learning and memory requirements, and to localize task-responsive units with respect to striosomes and body part areas identified by neuronal responses to cutaneous stimulation/passive manipulation and anterograde anatomical tracers from primary motor cortex. Neuronal activity in the globus pallidus and substantia nigra pars reticulata, output areas of the basal ganglia, is also to be examined over the course of acquisition of the three stimulus-response tasks. / (cont.) The proposed investigations will begin to empirically anchor learning and memory functions of the striatum and basal ganglia to patterns of neuronal activity in the context of the intricate anatomical organization of these areas. / by Deepa Radhakrishna Iyengar. / S.M.

Thesis (Ph.D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2002. / Includes bibliographical references (p. 141-151). / This thesis is a study of dynamics and learning in recurrent neural networks. Many computations of neural systems are carried out through a network of a large number of neurons. With massive feedback connections among these neurons, a study of its dynamics is necessary in order to understand the network's function. In this thesis, I aim at studying several recurrent network models and relating the dynamics with the networks' computation. For this purpose, three systems are studied and analyzed in detail: The first one is a network model for direction selectivity; the second one is a generalized network of Winner-Take-All; the third one is a model for integration in head-direction systems. One distinctive feature of neural systems is the ability of learning. The other part of my thesis is on learning in biologically motivated neural networks. Specifically, I study how the spike-time-dependent synaptic plasticity helps to stabilize persistent neural activities in the ocular motor integrator. I study the connections between back-propagation and contrastive-Hebbian learning, and show how backpropagation could be equivalently implemented by contrastive-Hebbian learning in a layered network. I also propose a learning rule governing synaptic plasticity in a network of spiking neurons and compare it with recent experimental results on spike-time-dependent plasticity. / by Xiaohui Xie. / Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, February 2009. / Includes bibliographical references. / The superior colliculus (SC) is a multi-layered midbrain structure responsible for multimodal integration and orienting behavior in mammals. The superficial layers of the SC (sSC) receive direct visual input from retinal ganglion cells (RGC) as well as input from pyramidal cells in layer V of the ipsilateral visual cortex (VC). The retinal input is refined well before eye opening (EO) and RGC axons arborize topographically to form an appropriate map of visual space. The projection from VC is still broad and unrefined at the time of EO, however. In both sSC and VC, physiological and biochemical evidence indicate considerable synaptic refinement in response to EO, which occurs naturally at the end of the second postnatal week. These studies use anterograde filling of corticocollicular axons in combination with controlled eyelid opening and reclosing paradigms to compare the corticocollicular projections of age-matched eye-opened and eye-sutured littermates. Reconstructions of individual corticocollicular axons in rat pups and statistically sampled arborization patterns across the colliculus at set times before and after controlled eye-lid opening, show that the onset of pattern vision is critical for the establishment of registration between the cortical and collicular maps of visual space. Moreover, if pattern vision is delayed by prolonging eye-lid closure the cortical projection withdraws to single axon cylinders. A latent plasticity remains, however; the corticocollicular axons can reestablish topologically appropriate arborization if eye opening occurs within at least a week of its normal occurrence. / by Julie R. Goldberg. / Ph.D.

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2017. / Cataloged from PDF version of thesis. / Includes bibliographical references (pages 119-132). / The richness of visual experience far exceeds our ability to remember what we have seen. However, it is unclear what neural mechanisms give rise to these limits to visual short-term memory capacity. Here, we measured neural activity in a change localization task, in which monkeys viewed two displays of multiple colored squares separated by a brief delay, and made a saccade to the square that changed color between displays. In chapter 2, we examine local field potentials in the lateral intraparietal area (LIP), frontal eye field, and lateral prefrontal cortex (PFC). At stimulus encoding, lower frequency oscillations decreased in power in proportion to the total number of stimuli presented, while higher frequency oscillations increased in power in proportion to the number of stimuli contralateral to the recording site. During the delay, lower frequency power instead increased with the number of contralateral stimuli, while higher frequency power was not modulated. We interpret these findings in terms of roles for low- and high-frequency oscillations in changing and maintaining cognitive state. In chapter 3, we compare spiking activity between LIP, PFC, and inferotemporal cortex (IT). Although the task required that the animal remember stimulus colors, activity in LIP and PFC primarily reflected the stimulus positions, while activity in IT primarily reflected color. In PFC, color information increased with the number of stimuli presented, while in IT, color information remained constant or decreased. Thus, IT was more strongly capacity-limited than PFC. Color selectivity during the delay was weak in all regions. However, in IT, activity at test stimulus presentation reflected the difference in square colors between the sample and test displays, while in PFC, activity primarily reflected the location of the changed square. Selectivity to these attributes was stronger on correct trials than incorrect trials. Our findings suggest a possible role for passive processes in IT in visual short-term memory. / by Simon Kornblith. / Ph. D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 1991. / Includes bibliographical references. / by Young Ha Kwon. / Ph.D.

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Brain and Cognitive Sciences, 2016. / Cataloged from PDF version of thesis. / Includes bibliographical references. / Rett Syndrome (RTT) is a pervasive, X-linked neurodevelopmental disorder that predominantly affects girls. The clinical patient features of RTT are most commonly reported to emerge between the ages of 6-18 months and as such, RTT has largely been considered to be a postnatal disorder. The vast majority of cases are caused by sporadic mutations in the gene encoding methyl CpG-binding protein 2 (MeCP2), which is expressed in the brain during prenatal neurogenesis and continuously throughout adulthood. MeCP2 is a pleiotropic gene that functions as a complex, high-level transcriptional modulator. It both regulates and is regulated by coding genes and non-coding RNAs including microRNAs (miRNAs). The effects of MeCP2 are mediated by diverse signaling, transcriptional, and epigenetic mechanisms. Whereas the postnatal effects of MeCP2 have been widely studied, pre-symptomatic stages of RTT have yet to be thoroughly investigated. Recent evidence from our lab among others suggests a role for MeCP2 during prenatal neurogenesis that may contribute to the neuropathology observed later in life. We sought to characterize the course of neurogenesis in MeCP2-deficient human neurons with the use of induced pluripotent stem cells (iPSCs) derived from RTT patient skin samples. We generated a variety of monolayer and 3D neuronal models and found that RTT phenotypes are present at the earliest stages of brain development including neuroepithelial expansion, neural progenitor migration and differentiation, and later stages of membrane and synaptic physiological development. We established a link between MeCP2 and key microRNAs that are misregulated in RTT and lie upstream of signalling pathways that contribute to aberrant neuronal maturation in the absence of MeCP2. We have uncovered novel roles of MeCP2 in human neurogenesis. Whereas the processes that comprise early neural development were previously considered irrelevant to RTT pathology, the deficits we observed in neuronal differentiation, migration, and maturation are a crucial component to the larger picture of RTT pathogenesis and provide additional insight into the emergence of RTT patient symptoms. / by Danielle A. Feldman. / Ph. D.