Ila R. Fiete - US grants

This project aims to uncover the existence of a qualitatively better class of analog error-correcting codes than previously known in the brain, show how such codes can be used and decoded, and develop the theory for quantifying the performance of such codes.

Information theory was introduced into neuroscience relatively early, and the theory of efficient (source) coding has been widely embraced in the sensory neurosciences. However, the second branch of information theory, which deals with the maximally parsimonious addition of redundancy to recover signal from noise, has curiously not made inroads in neuroscience. Shannon's channel coding theorem revealed the existence of codes that make possible error correction at efficiencies previously thought impossible.

The investigator's central hypothesis is that the brain routinely employs such error correcting codes and the machinery required to decode and work with them. The hypothesis is motivated by a recent analysis of the grid cell code for animal location by the investigator and colleagues, showing it has unprecedented error-correction properties compared to known population codes in the brain (Sreenivasan & Fiete, 2011). The investigator proposes to: 1) Develop definitions and constraints for analog neural codes, to apply the channel coding framework to neural codes and thus characterize their "goodness" on error-correction. 2) Identify high-level coding properties that enable strong error-correction, and search for these properties in observed but poorly understood neural codes. At the same time, explore strong theoretical error-correcting codes that the brain may plausibly implement. 3) Model plausible neural mechanisms for decoding such codes. Decoding is inference, so this question can be more generally thought of as exploring neural mechanisms for hierarchical inference.

This project is computational and theoretical, and also involves close collaboration with neurophysiologists, to apply quantification techniques to neural data and work with experiments to inform the theories and test predictions.

The goal of this project is to combine computational modeling with behavioral and neuroimaging studies to characterize the mechanisms of navigational abilities in humans and understand how they decline with age. The PIs will focus on an important navigational circuit in mammals, which consists of the hippocampus and associated areas, and includes grid cells of the entorhinal cortex as well as place cells. Place cells have highly location-specific responses, turning on at one location in an environment and firing little elsewhere; grid cells by contrast fire at multiple locations within an environment, with periodically separated activity blobs in a striking triangular lattice pattern. Studies in rodents have detailed the properties of grid and place cells, and led to neural network models whose additional predictions have often been borne out by single-unit neuron recordings. However, much less is known about grid cells and place cells in humans, and the nature of interactions between different parts of the navigation circuit remains unclear, in rodents and humans.

In this project, the PIs bring to bear virtual-reality-based behavioral experiments, ultra-high-resolution fMRI recordings during virtual navigation, and neural network modeling, to better understand the circuit for spatial navigation in humans. The PIs plan a three-pronged approach to these questions. The first is to characterize phenomenologically the characteristic errors made by humans, through navigation environments with and without accurate external landmark cues, and under other externally varying conditions, in aged and non-aged subjects. The second is to employ neural network models of grid cells, to model the network parameters that could give rise to the observed deficits, and in turn test the predictions of these models with the neuroimaging experiments. The experimental setup will permit systematic variation in the fidelity of external sensory cues, to probe the relative contributions of the complementary computations of dead-reckoning (path integration) versus landmark-based navigation, and uncover their potential neural substrates in humans. The results will help to develop models of how parallel streams of spatial information are combined and processed across brain areas to aid in navigation. The third component is to develop accurate algorithms for extracting spatial information from high-resolution fMRI data from regions and sub-regions of the entorhinal-hippocampal complex. The aim is to map the distribution of location information across areas and learn where it is most compromised in old age.

This award is being co-funded by NSF's Office of the Director, International Science and Engineering. A companion project is being funded by the German Ministry of Education and Research (BMBF).

? DESCRIPTION (provided by applicant): The system that controls smooth pursuit eye movements is one of the most accessible, promising systems for understanding how neural circuits transform sensory inputs into actions. Pursuit is a natural, ecologically- relevant behavir that allows primates to track moving objects of interest in the visual world. Now-classical analyses relating neural activity to behavior have already provided insights about the systems-level functions and computations of the pursuit circuit that perhaps exceed our understanding of all other voluntary behaviors in mammals. Despite these successes, there are large gaps in our understanding of the pursuit system. We seek a new level in understanding the circuit by measuring and manipulating the activity of large populations of identified neurons in the key sensory and prefrontal cortical areas. We intend to address how different neuronal types function during pursuit, how they implement systems-level computations within micro- and meso-circuits, and how control centers select the appropriate sensory data given cognitive factors. These questions, although stated in the context of the pursuit system, are instances of the more general problem of how population activity in large numbers of sensory neurons is parsed and converted into appropriate behaviors. The time is now ripe to take advantage of several technical and conceptual revolutions: Specifically, we propose to study the neural basis of pursuit eye movements in marmosets, in which cortical areas responsible for motion sensation and target selection are easily accessed and for which genomic toolboxes are being generated. We will investigate the functional circuitry at multiple scales using a combination of 2-photon calcium imaging and large-scale extracellular array electrophysiology, cell-type identification and optogenetic perturbation, and dynamical systems modeling.

Humans make rich inferences about the relationships between entities in the world from scarce information. For example, we can find a novel destination after seeing a few street numbers, or find a page in a dictionary by glancing at a few words in other pages. Theoretical considerations suggest that the brain makes such inferences by constructing internal models of the relationships in the environment (relationships between actions and states of the world), and by mentally simulating those models. However, the neural substrates and mechanisms of mental simulation are not understood. Our overarching goal is to integrate insights from theory and modeling with behavior and electrophysiology in awake, behaving monkeys to understand how mental simulation of internal models support relational inference. We will develop a behavioral task for monkeys in which they have ?navigate? mentally from one stimulus to another along a one-dimensional abstract space of discrete stimuli (i.e., a sequence of images). We will assess whether animals? behavioral characteristics exhibit hallmarks of mental simulation. We will then create a large library of neural network models to generate hypotheses for alternative computational strategies (including mental simulations) that the brain might employ for navigating abstract spaces. Next, we will record from candidate brain areas in the parietal and entorhinal cortex of monkeys, and analyze the data at single cell and population levels looking for signatures of mental simulation. Finally, we will adopt an iterative approach involving model-based data analyses and data-driven model revision with the ultimate goal of creating models that simultaneously succeed in performing task-relevant computations (i.e., behavior) and account for observed neural responses. Finally, we will validate our framework by evaluating the predictions of our models for both behavior and electrophysiology in new behavioral tasks.

Summary/Abstract, Project 3 Even in the same environment, an animal may make different decisions on different occasions, because its internal state, such as engagement in a task, interacts powerfully with external inputs to determine behavior. This proposal?s overarching goal is to understand how internal states influence decisions and to identify the underlying neural mechanisms. The team is part of the International Brain Laboratory (IBL), an established consortium that has developed a standardized mouse decision-making task and standardized methods for training, neural measurement, and data analysis, along with a working, scalable infrastructure for sharing data. The goal of Project 3 is to synthesize the findings of experimental Projects 1, 2, 4, and 5 into circuit-level mechanistic models of the IBL task. The task involves hierarchical, probabilistic decision-making through sensory evidence integration to make left-right decisions about where the stimulus is on the current trial, along with integration on a longer timescale to estimate the slowly varying left-right biases in where the stimuli are more likely to appear. Initial models not only will be trained to reproduce expert-level task performance, but also will include general biological constraints on neural dynamics and anatomical connectivity gradients. They will be analyzed for their learning dynamics, and for which parameters are the handles through which internal states exert their effects on circuit computation and dynamics. These models will yield predictions on multiple levels of abstraction: state-space predictions, network structure predictions, and anatomical predictions. The resulting models will be deployed in a tight loop with all experimental projects, to guide experimental design; serve as ground-truth testbeds for perturbative and causal connectivity analysis studies; and link statistical analysis results from data with mechanistic interpretations. The results of these experiment-model prediction comparisons will then be used to further refine and elaborate the models. Project 3 researchers will incorporate the experimentally derived neural activity data, causal connectivity by anatomical region data, and structural cell-type and connectivity data to further constrain the models. Finally, Project 3 will also generate highly simplified abstract neural circuit models, using novel methods of model compression to elucidate the general principles underlying hierarchical decision-making in the brain. All this work involves the use and de novo development of cutting-edge modeling, statistical, and data analysis tools. The work of Project 3 will thus deliver a mechanistic circuit- level understanding of this proposal?s overarching hypothesis that information flow and communication across brain regions during decision-making depends on internal state.