Principals of cue integration for accurate position estimates across contexts
Abstract
A central function of the brain is to create internal representations of stimuli and experiences from theoutside world. A classic example of this function is our ability to determine our current position in physical spaceor an imagined, mental space. In a physical space, such as a familiar parking lot, we can localize our currentposition by combining information from sensory landmarks (e.g. familiar buildings or street light locations) withself-motion cues (e.g. our current movement speed or heading direction). If asked later to recall the location ofour car in the same parking lot, we can also mentally determine our position along an imagined spatial trajectoryto our car???s location, despite the absence of actual self-motion inputs. This mental spatial trajectory can alsoincorporate non-spatial, semantic cues. A classic example is the mnemonic device of a ???memory palace???, in whichan individual memorizes items (e.g. a list of words) by visualizing the items along a mental spatial trajectorythrough an imagined spatial environment (e.g. a house).Recent research suggests that the same neural substrates support navigation through both physical andmental space. Neurons in the medial entorhinal cortex encode many of the components needed to generate aneural map of physical space, including: an agent???s current position, heading direction and running speed, as wellas multiple sensory landmark features. Work across rodents, as well as non-human and human primates, has nowdemonstrated that these same entorhinal neurons also encode landmark-like features and an agent???s currentposition in mental space. This suggests that entorhinal neural responses reflect navigation through a behaviorallyrelevant stimulus space regardless of the coordinate frame of that space. Importantly, at a high level, navigationthrough physical space and navigation through a mental, imagined space share core features. In both cases,multiple cues must be integrated to form a unified percept of position. In physical navigation, self-motion cues(e.g. optic flow) and landmark cues (e.g. environmental boundaries) must be integrated to generate a positionestimate within a given spatial environment. In mental or imagined navigation, the same type of cues must beintegrated to generate an estimate of position within the relevant feature space. In the case of mental navigation,self-motion cues are replaced by information regarding abstract movement through the relevant feature space,such as time elapsed, and landmark cues are replaced by semantic concepts, such as discrete components of thetask (e.g. the start of the behavioral task or reaching a task-relevant goal). However, for both physical and mentalnavigation, the algorithms for how these different cues combine to generate MEC neural responses capable ofsupporting position estimates remains incompletely understood.We hypothesize that, at a fundamental level, a unified algorithm determines the integration principles ofcues used to generate position estimates during both spatial and mental navigation, with the flexible nature ofMEC coding reflective of different inputs rather than different underlying computations. Here, I propose to testhis hypothesis using a highly interdisciplinary approach. Using in vivo electrophysiology, we will first monitorthe simultaneous activity of tens to hundreds of MEC neurons as rodents perform navigational tasks in virtualreality or goal and memory directed foraging tasks (Task 1). We will then apply new machine learning approachesand statistical analyses to reveal how MEC encodes navigational and cognitive variables at both the level ofsingle-neurons and as a neural population (Subtask 1.2 and Subtask 2.2). Our findings from these experimentswill generate the foundation for new network-level computational models (Subtask 2.2). These models will testour hypothesis that different sensory inputs give rise to dynamic MEC neural patterns, while the underlyingalgorithms generating these
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 27, 2018
- Source ID
- N000141812690
Entities
People
- Lisa M. Giocomo
Organizations
- Office of Naval Research
- Stanford University
- United States Navy