Please use this identifier to cite or link to this item:
http://arks.princeton.edu/ark:/88435/dsp01bk128f12k
Title: | Geometric and Dynamic Neural Codes of Sensory, Memory and Prediction |
Authors: | Libby, Alexandra |
Advisors: | Buschman, Timothy J |
Contributors: | Neuroscience Department |
Keywords: | dynamics learning memory oscillation rotation sequence |
Subjects: | Neurosciences Artificial intelligence Cognitive psychology |
Issue Date: | 2022 |
Publisher: | Princeton, NJ : Princeton University |
Abstract: | Our brains use memories and predictions to improve how we process information and make decisions. When we interact with our environment, we are receiving sensory inputs, storing memories, making predictions and choosing behaviors. Thus, the brain needs mechanisms to represent, organize and combine all these forms of information. This thesis focuses on the neural dynamics that represent and combine sensory and memory information, which are essential components of the sequential and predictive neural codes that support behavior. In Chapter 2, we use data from mouse auditory cortex to show how neural population codes support sequential sensory processing, by facilitating predictive coding, while also storing short-term memories. We show that specific geometries in the population code can align sensory representations to make predictions, while others separate sensory and memory representations to avoid interference. Using experimental evidence and computational modeling, we show that the brain’s uses an efficient, rotational mechanism to orthogonalize the sensory and memory representations, which keeps them separate and avoids overwriting. In Chapter 3, we study how the brain controls the neural dynamics of sensory, memory and sequential representations across time. With an inhibition stabilized firing rate model, we show that oscillations allow the brain to sample among ‘dynamical regimes’ or to switch across externally sensory driven dynamics and internally memory generated dynamics, such as predictions or sequences. Additionally, we show that coherent oscillations can couple dynamics across networks. For example, two networks that show pattern completion and associative dynamics when alone, produce sequential dynamics when coupled. Taken together these projects illustrate two mechanisms that control neural dynamics. First, the geometry of neural population responses can align or separate information. Second, oscillations in inhibition can combine multiple dynamics to facilitate behaviorally relevant computations. |
URI: | http://arks.princeton.edu/ark:/88435/dsp01bk128f12k |
Alternate format: | The Mudd Manuscript Library retains one bound copy of each dissertation. Search for these copies in the library's main catalog: catalog.princeton.edu |
Type of Material: | Academic dissertations (Ph.D.) |
Language: | en |
Appears in Collections: | Neuroscience |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
Libby_princeton_0181D_14199.pdf | 7.95 MB | Adobe PDF | View/Download |
Items in Dataspace are protected by copyright, with all rights reserved, unless otherwise indicated.