ESE Seminar: Dissertation Defense Lijun Zhang
Friday, January 21, 2022 1:30 PM to 3:30 PM
About this Event
Modern neural recording methodologies, including multi-electrode and optical recordings, allow us to monitor the large population of neurons with high temporal resolution. Such recordings provide rich datasets that are expected to understand better how information about the external world is internally represented and how these representations are altered over time. Achieving this goal requires the development of novel pattern recognition methods and/or the application of existing statistical methods in novel ways to gain insights into basic neural computational principles. In this dissertation, I will take this data-driven approach to dissect the role of short-term memory in olfactory signal processing in two relatively simple models of the olfactory system: fruit fly (Drosophila melanogaster) and locust (Schistocerca americana).
First, I will focus on understanding how odor representations within a single stimulus exposure are refined across different populations of neurons (faster dynamics; on the order seconds) in the early olfactory circuits. Using light-sheet imaging datasets from transgenic flies expressing calcium indicators in select populations of neurons, I will reveal how odor representations are decorrelated over time in different neural populations. Further, I will examine how this computation is altered by short-term memory in this neural circuitry. Next, I will examine how neural representations for odorants at an ensemble level are altered across different exposures (slower dynamics; on the order of tens of seconds to minutes). I will examine the role of this short-term adaptation in altering neural representations for odor identity and intensity. Lastly, I will present approaches to help achieve robustness against both extrinsic and intrinsic perturbations of odor-evoked neural responses. I will conclude with a Boolean neural network inspired by the insect olfactory system and compare its performance against other state-of-the-art methods on standard machine learning benchmark datasets. In sum, this work will provide insights into how short-term plasticity alters sensory neural representations and their computational significance.
Event Details
See Who Is Interested
User Activity
No recent activity
About this Event
Modern neural recording methodologies, including multi-electrode and optical recordings, allow us to monitor the large population of neurons with high temporal resolution. Such recordings provide rich datasets that are expected to understand better how information about the external world is internally represented and how these representations are altered over time. Achieving this goal requires the development of novel pattern recognition methods and/or the application of existing statistical methods in novel ways to gain insights into basic neural computational principles. In this dissertation, I will take this data-driven approach to dissect the role of short-term memory in olfactory signal processing in two relatively simple models of the olfactory system: fruit fly (Drosophila melanogaster) and locust (Schistocerca americana).
First, I will focus on understanding how odor representations within a single stimulus exposure are refined across different populations of neurons (faster dynamics; on the order seconds) in the early olfactory circuits. Using light-sheet imaging datasets from transgenic flies expressing calcium indicators in select populations of neurons, I will reveal how odor representations are decorrelated over time in different neural populations. Further, I will examine how this computation is altered by short-term memory in this neural circuitry. Next, I will examine how neural representations for odorants at an ensemble level are altered across different exposures (slower dynamics; on the order of tens of seconds to minutes). I will examine the role of this short-term adaptation in altering neural representations for odor identity and intensity. Lastly, I will present approaches to help achieve robustness against both extrinsic and intrinsic perturbations of odor-evoked neural responses. I will conclude with a Boolean neural network inspired by the insect olfactory system and compare its performance against other state-of-the-art methods on standard machine learning benchmark datasets. In sum, this work will provide insights into how short-term plasticity alters sensory neural representations and their computational significance.