Kechen Zhang - US grants

Many neurons in the auditory system respond to sounds nonlinearly; that is, its response to two sounds played simultaneously differs from the sum of its responses to each sound played alone. Nonlinearities are necessary for many computational functions, but unlike nonlinear models that allow closed-form solutions, nonlinear models are often too hard to characterize in practice. To make nonlinear models tractable, this project will combine single-unit recording in awake marmoset monkey with automated online stimulus design by parallel computing. The goal of this stimulus design is not to maximize the firing rate of a neuron, but to extract the most information about the global stimulus-response relationship. Optimal sounds will be designed "on the fly" according to a neuron's response history, with the help of a fast parallel computer whose running time is compatible with the single-unit recording experiment. The proposed research is expected to produce practical and widely applicable methods for characterizing nonlinear sensory neurons. The auditory system is an ideal system for this type of online experiment because sound space is of lower dimensions and allows faster computations. The methods developed here are expected to generalize to nonlinear problems in other sensory modalities.

Theory and algorithm development will focus on generating sound stimuli which can either most accurately estimate a given model, or maximally distinguish competing models. Nonlinear models with various degrees of complexity, including neural network models, will be used simultaneously, and contrasted against one another in the automated experiment. The model-based sound design method will be used to characterize complex response properties of neurons in auditory cortex and inferior colliculus of awake marmoset monkey, a vocal primate. This project focuses on the auditory cortex because studies of its pronounced nonlinearities may potentially benefit most from the new method. For comparison the same method will also be applied to the inferior colliculus, the inputs to which are better known, allowing more realistic hierarchical models to be developed. The models obtained from this method should provide a concise summary of the global stimulus-response relationship of a neuron that generalizes across all types of stimuli. Neural network models may also help extract additional information about the connectivity between different neuronal types, thus providing a link between the stimulus-response function and the structure of the underlying neural circuits.

The PIs and Co-PIs of grants supported through the NSF-NIH Collaborative Research in Computational Neuroscience (CRCNS) program meet annually. This will be the sixth meeting of CRCNS investigators. The meeting brings together a broad spectrum of computational neuroscience researchers supported by the program, and includes poster presentations, talks and plenary lectures. The meeting is scheduled for June 6-8, 2010 and will be held at Johns Hopkins University.

DESCRIPTION (provided by applicant): For as long as it has been possible to measure electrical activity in the nervous system, it has been known that the brain produces oscillatory rhythms. Some rhythms are generated during sleep, others during waking; certain patterns of oscillatory brain activity occur in all healthy people, while other patterns only occur in disease states such as epilepsy, clinical depression, or schizophrenia. Many different brain rhythms have been identified and characterized, and yet almost nothing is known about their function. We know that the brain oscillates, but we do not know why. Over the past few years, discoveries have been made that provide tantalizing new clues for answering this question, by suggesting that neural oscillations are very much like threads that the brain weaves together to create the fabric of memory and perception. In rats, one particular kind of oscillation referred to as theta rhythm is very predominant in the hippocampus and entorhinal cortex, brain areas that play a critical role in learning and memory. It is becoming increasingly clear that theta oscillations (in the frequency band of 4-12 Hz) are building blocks from which the hippocampus and entorhinal cortex can construct memory representations. The studies proposed here will combine neurophysiological recording experiments with computational modeling studies to investigate how the rat brain uses theta oscillations to form memories of familiar locations in space. Neurons called place cells and grid cells become active whenever a rat visits certain familiar locations, and these neurons are strongly synchronized by theta oscillations. Proposed computational modeling studies will investigate how place cells and grid cells use theta oscillations to encode spatial memories, and will seek to decipher the structure of the biological neural networks that perform this task. Proposed neurophysiology studies will attempt to show for the first time that neural oscillators in subcortical regions store memory representations using a phase code, and will examine how the cerebral cortex interacts with subcortical oscillators to read out these memory representations. Pharmacological inactivation studies will be conducted to demonstrate how memory processing breaks down when neural oscillators are disrupted, which may help to explain the causes of memory impairment in humans who suffer from amnesic syndrome in conjunction with disorders like Alzheimer's disease, schizophrenia, depression, anxiety disorders, and post-traumatic stress. By elucidating how memories are formed from theta oscillations in spatial memory circuits, the research proposed here will provide groundbreaking new insights into the fundamental role that neural oscillations play in normal memory processes. This work may in the future make it possible to diagnose and treat brain diseases and mental disorders that currently are not well understood, but which may prove to have roots in dysfunction of the neural oscillators that provide the basic building blocks for memory and perception.

DESCRIPTION (provided by applicant): Our project will be focused on sensory representations in the auditory system with two overall objectives. First, we will adopt an experimental strategy that calls for optimal probing of the auditory system. We will determine how complex sound stimuli at different spatial locations are represented at multiple stages of the auditory pathway in awake marmoset monkeys by improving an optimal design approach recently made feasible by the collaborative research by the same team. This method works online, during neurophysiological recording, by generating sound stimuli that maximize the information gained about a hierarchical neural network model. A model attained from the online experiment provides an accurate description of complex auditory responses to rich sound stimuli in three dimensional space, and at the same time we also obtain a plausible network explanation of how complex response properties in inferior colliculus and auditory cortex might arise from combining multiple sound localization cues at lower levels in the auditory pathway. Second, we will start with the hypothesis that neural populations are optimized for real world situations, and we will determine whether the rich variety and heterogeneity of neuronal response properties measured in experiments can be explained by the hypothesis that they actually form an efficient population optimized for natural sounds. The expected outcome is a principled computational explanation for the complexity of neuronal populations for sound localization, including all the diversity and variability associated with various functional cell types. Intellectual merit: Understanding the relationship between neural activity and the stimuli from the external world is one of the basic goals in systems neuroscience. Our collaborative research may help resolve a longstanding problem in auditory neuroscience concerning how neuronal responses in auditory cortex and inferior colliculus represent space over the full 360? range of azimuths and elevations. Our approach is very general and should apply readily to other sensory modalities as well. In particular, the optimal design method can obtain a global picture of the stimulus-response landscape as fast as possible, and this speedy feature is especially valuable for working with awake and behaving animals, and even humans. Our results may extend to many disciplines whenever one needs to efficiently probe a parameterized input-output system, or to optimize a population of sensors. For example, the ideas and the methods developed here could be used to guide practical neuromorphic engineering designs or to optimize populations of artificial sensors, which may find many practical applications. Our results will also provide a solid conceptual basis and a set of modeling tools to qualify abnormal or diseased states of the auditory system at both the cortical and subcortical levels for processing of complex sound signals. This progress could lead to the development of viable therapies for various neurological disorders, and it could improve the development of neural prostheses by better parsing complex auditory scenes with the help of sound localization cues. Broader impact: This project contributes to teaching at several levels. It will provide research training to two graduate students and the results of these studies are expected to constitute the bulk of their Ph.D. theses. In addition, a postdoctoral fellow will receive research training in collaborative research in computational neuroscience. Efforts will be made to include participation from students and researchers over a wide demographic, and the positions will be broadly advertised to ensure that qualified underrepresented candidates are aware of the opportunities. This research will also be integrated into our educational efforts by incorporating results into courses developed for both graduate and undergraduate students at Johns Hopkins University. The training, educational and outreach components will directly affect a large number of students and other groups outside of the university by exposing them to open problems and interdisciplinary research methods in neurophysiology and computational neuroscience. The PI and co-PIs are committed to advancement of women and underrepresented minorities in research and education. Furthermore, the proposed work will strengthen our infrastructure for further studies by pioneering new recording techniques that are based on closed-loop automated stimulus design. The research will be disseminated in many venues, including national and international meetings and peer reviewed journals. When applicable, our results will be disseminated in popular, non-technical literature as well. All software associated with this work will be made freely available on the internet, and appropriate subsets of the data collected for this project will be made available on the CRCNS-funded data sharing facility.

DESCRIPTION (provided by applicant): The proposed work investigates the representation of information about sound in the central nucleus of the inferior colliculus (CNIC). This structure receives a complex array of inputs containing different kinds of information about the sounds coming into the ears. As many as 20-30 separate processing centers in the brainstem contribute inputs to the CNIC. These include information about what the sound is, who or what produced it, and where the source is located. All of this information must pass through the CNIC on the way to the cortex, so the way in which the neurons of the CNIC assemble these diverse representations is essential to understanding auditory processing. Three investigations are proposed. First, a new statistical method will be applied to construct generic models of the receptive fields of neurons in CNIC. These models will allow understanding of the way the identity of sounds and their information content is represented. Second, the way in which the three cues for sound localization are combined in neurons of the CNIC will be studied. This will provide insights into the way in which neurons encode multiple aspects of stimuli, a common problem in all parts of the brain. Third, the circuits that amplify dynamic features of sound will be studied. These elements allow us to deal with complex acoustic environments with multiple sound sources, especially with the problem of separating sources that produce overlapping sounds.

The striking spatial correlates of hippocampal place cells and grid cells have provided unique insights into how the brain constructs internal, cognitive representations of the environment and uses these representations to guide behavior. These spatially selective cells are influenced by both self-motion signals and by external sensory landmarks. Self-motion signals provide the basis for a path integration computation, in which the hippocampal system tracks the animal's location by integrating its movement vector (speed and direction) over time to continuously update a position signal on an internal cognitive map. To prevent accumulation of error, it is crucial that this endogenous spatial representation be anchored by stable, external sensory cues, such as individual landmarks and environmental boundaries. Accurate path integration requires that an internal representation of position be updated in precise agreement with the animal's displacement in the world. What if the relation between position calculated by self-motion cues and position defined by landmark cues is altered, e.g. during development (slow time scale) or due to injury (fast time scale)? Does the animal recalibrate the internal gain between representations of its movement and the updating of the representation of its position in the brain? We hypothesize that this gain must be learned by reference to visual feedback. We constructed an augmented reality system that allows precise, closed-loop control of the visual environment as rats move through physical space and provide evidence that the path integration system can indeed be recalibrated. We propose a collaborative research program to investigate plasticity of the path integration gain at multiple neural levels using combined theoretical, engineering, and experimental approaches. We will combine mathematical analysis, biologically inspired attractor network theory, and principles derived from engineering to develop the first models of how the path integration system dynamically recalibrates itself in response to sensory feedback. We will perform recordings from the hippocampus and medial entorhinal cortex to provide data to constrain and test these models. The combined expertise of the Pl and Co­ Investigators in electrophysiological recordings of the hippocampal system, engineering, and mathematical neuroscience will propel the theory forward to explain the network dynamics and functional implications of this ethologically critical form of neural plasticity.

This project brings together theories of brain functions and principles of robotic swarm control to develop smarter swarms and to better understand the neural processes underlying spatial representations, navigation, and planning. Our world is constantly changing, and mammals have evolved the cognitive ability to plan new paths or new strategies as needed. By contrast, autonomous robots are less robust, and often have difficulty operating in complex, changing environments. This research project is grounded in the idea that individual robots in a group can be thought of analogously to neurons in an animal's brain, which interact with one another to form dynamic patterns that collectively signal locations in space and time relative to brain rhythms. This distribution of information across space and time will enable a new paradigm of swarm control, in which swarms automatically adapt to changes in the world in the same way that a rat knows which detour to take around an unexpected obstacle. Unmanned robots are rapidly becoming a crucial technology for commercial, military, and scientific endeavors throughout the nation and across the globe. Critical future applications such as disaster relief and search & rescue will require intelligent spatial coordination among many robots spread over large geographical areas. This project will advance neural swarming as a control paradigm for this next generation of technological development. Additionally, this project will drive an extensive science, technology, engineering, and mathematics education program to bring the concepts of spatial intelligence, hippocampal information processing, and swarm control to high school students to improve literacy in neuroscience and robotics.

The project's goal is to build a unified framework for self-organized, bottom-up control of spatial task planning that synergistically advances theoretical neuroscience and swarm control paradigms. In the project's brain-to-swarm metaphor, neurons are autonomous agents, spikes are agent-based phase signals, and emergent circuit activity is emergent swarm behavior. The approach targets neural computations in hippocampal circuits and related systems that may contribute to online dynamic replanning. The research thrusts comprise data-driven dynamical network and point-process models of neural activity sequences, mathematical analysis of swarming dynamics using matrix manifolds, and autonomous systems simulations in realistic virtual environments. The project will advance understanding of emergent hippocampal dynamics and autonomous methods for dynamic replanning, motivating new research in distributed control. The project's framework may enable mass-scalability for large, agile swarms of simple robotic agents.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

The hippocampus is a critical structure for learning and memory in the mammalian brain. During active exploration, hippocampal circuits support the activity of place cells that track an animal?s position as it moves. The spatial representations of place cells in a particular environment collectively form a spatial map, the function of which remains largely unclear. During periods of rest, the hippocampus switches to a network state characterized by stochastic and highly variable activity patterns. A prominent feature of this irregular activity is recurring strong bursts of excitation, called sharp waves, that recruit large proportions of hippocampal neurons and sweep across the circuits of the hippocampus. Sequentially ordered activation of place cells during sharp waves is considered to serve a critical role in memory consolidation. In addition to replaying recently experienced routes, sharp wave sequences can follow unexplored paths through space and can be biased to goal destinations according to reward value. This project will investigate an explicit theoretical framework for flexible sequence generation in service of prospective route planning for navigation. The theory rests on the observation that the same network that initiates sharp waves, the hippocampal CA3 region, also prominently carries a distinct lower-frequency band of gamma oscillation. One study has shown that the slow gamma rhythm modulated spiking activity during sharp wave sequences, such that the phase of gamma with the strongest activity corresponded to periods when the decoded sequence would dwell, or ?hover?, at discrete spatial locations. Informed by recording data from that study, this project will develop realistic hippocampal network models to study how sharp waves and slow gamma oscillations might emerge simultaneously. On the basis of that physiological model, spatial activity will be embedded into the synaptic weights of the network to assess whether sharp waves and slow gamma interact to support gamma-locked attractor dynamics, and how that interaction depends quantitatively on the synaptic modifications for spatial learning. In particular, focus will be placed on a puzzling experimental demonstration that sharp wave sequence recall slows down during learning. The long-term objective is to fully evaluate whether these emergent phenomena support a unified framework for the function of spatial maps in navigation on the basis of enabling flexible generation of novel sequences. A clear theoretical understanding of these key hippocampal phenomena in realistic networks will elucidate their role in memory, planning, and disease states.