• Login
• Admin

Jiang, Jiamu (2024) Energy dependent reinforcement learning based on the neuronal mechanisms of the olfactory processing in mushroom body. PhD thesis, University of Nottingham.

Preview

PDF (Jiang, Jiamu, 20196784, corrections) (Thesis - as examined) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader Available under Licence Creative Commons Attribution. Download (6MB) | Preview

Abstract

The metabolic energy is crucial for neural processing related to learning, which modulates computational capabilities, neuronal quantities, synaptic connections, and long-term memory formation. From the evolutionary perspective, these neural processes have shaped organisms' adaptive responses to environmental stimuli and enhanced survival chances. Studies have highlighted that associative conditioning extends the lifespan across various species. Also, insects modulate memory types based on ecological determinants.

This thesis concentrates on the olfactory nervous system in Drosophila's Mushroom Body (MB), as a structure paralleling the mammalian brain hippocampus, presenting as a model organism for unraveling memory formation intricacies, given its genetic accessibility and well studied olfactory processing.

This research posits that energy constraints might represent evolutionary adaptations promoting survival and learning efficiency. By dissecting the fruit fly's learning processes—specifically regarding metabolic energy—this study aims to assess potential lifespan extensions via energy modulation during learning and to gauge the efficacy of learning under energy constraints.

We identified three adaptive reinforcement learning variations, each influenced by the energy dynamics observed in fruit flies. The first variation underscores the capability of energy-driven memory pathway regulation to augment the fruit fly's lifespan, particularly when synergized with dopamine regulation. The subsequent variation reveals that the strategy of depressing synapses linked to undesired actions demonstrates high efficiency in synaptic adjustments across both aversive and appetitive conditioning contexts. The final variation applies the energy-adaptive methods to the conventional multi-armed bandit algorithms, such as the Upper Confidence Bound (UCB) and Bayesian-based Thompson Sampling (TS), and emphasizes the capacity of energy-adaptive methods to prolong the agents' lifespan without significant sacrifice in regret.

In summary, the thesis delineates the integral role of energy dynamics in shaping and optimizing learning processes and behaviors, drawing inspiration from the olfactory learning in the MB. These findings contribute to our understanding of the nature of energy, learning, and survival.

Item Type:	Thesis (University of Nottingham only) (PhD)
Supervisors:	van Rossum, Mark
Keywords:	Reinforcement Learning, Synaptic Plasticity, Survival Analysis, Drosophila
Subjects:	Q Science > QL Zoology
Faculties/Schools:	UK Campuses > Faculty of Science > School of Mathematical Sciences
Item ID:	78727
Depositing User:	Jiang, Jiamu
Date Deposited:	13 Dec 2024 04:40
Last Modified:	13 Dec 2024 04:40
URI:	https://eprints.nottingham.ac.uk/id/eprint/78727

Actions (Archive Staff Only)

Edit View