| Details: |
|
Date：
|
24/04/2025 (Thursday)
|
|
Time：
|
15:00-17:30
|
|
Venue：
|
LT1, Wui Chi Building, Main Campus
|
|
Student：
|
Yide Yu
|
|
Topic：
|
Design and Development of Graph-based Solvers for Partially Observable Markov Decision Processes in Control Systems
|
|
Abstract： |
This thesis explores graph-based solvers for Partially Observable Markov Decision Processes (POMDPs), addressing critical challenges in decision-making under uncertainty. POMDPs serve as a foundational framework in artificial intelligence, enabling sequential decision-making in environments where states are only partially observable. However, existing solvers face limitations in scalability, computational efficiency, and adaptability to complex, dynamic environments. To overcome these challenges, this research proposes a novel graph-theoretical framework that integrates Markov Decision Process Graph Representations (MDPG) and automata theory to enhance POMDP solvers. Key contributions include the development of the State-Observation-Gap (SOG) metric to quantify observation uncertainty, the Curiosity-driven Algorithm based on Graph for Exploration (CAGE) algorithm series to balance exploration and exploitation, and the biomimetic BIOMAP algorithm to transform partially observable problems into fully observable ones. These innovations are complemented by the Partial Observable Model for Anesthesia Control (POMA-C) framework for precise anesthesia control under incomplete observations and the Wildlife Analysis and Population Ecology Tool (WAPET) for ecological simulations, demonstrating the versatility of graph-based methods across diverse applications. Experimental results show that the proposed approaches significantly improve decision-making efficiency, robustness, and scalability in partially observable environments. This research bridges theoretical advancements and practical implementations, offering a comprehensive framework for solving POMDPs and opening new avenues for future exploration in reinforcement learning and real-world decision-making scenarios. |
|
Enquiry：
|
fca@mpu.edu.mo
|
|
|---|
