2024 Miami Section Invited Announcement: Adaptation from Parallelized Reinforcement Learning Agents for Complex Systems Under Uncertainties
Artificial Intelligence (AI) is one of the most exciting developments in the last decade. At the same time, Reinforcement Learning (RL) has emerged as a very successful paradigm for a variety of machine learning tasks. RL method usually finds an action policy that maximizes total cumulative reward over a given timeframe. With a combination of (deep) neural networks, it could handle complex features and make predictions about future states and rewards. However, existing RL - based optimization approaches are hampered by the generalization capacity and the increasing computational burden to accommodate the uncertainties. Some RL methods often analyze with some ideal assumptions without considering the abrupt changes and are hard to transfer to dynamic environments. We have designed a new paradigm of parallelized reinforcement learning (PRL) methods based on probabilistic events to handle the uncertainties. Several local (virtual) learning agents are employed to interact with pertinent environments in a distributed manner and report outcomes to the global agent. The advantage estimate functions of learning agents are designed with a backward sweep to transfer the outcomes to the value function updating process.
Date and Time
Location
Hosts
Registration
- Date: 09 May 2024
- Time: 11:00 AM to 12:00 PM
- All times are (UTC-04:00) Eastern Time (US & Canada)
-
Add Event to Calendar
Speakers
Dr. Zhen Ni of Florida Atlantic University
Adaptation from Parallelized Reinforcement Learning Agents for Complex Systems Under Uncertainties
Artificial Intelligence (AI) is one of the most exciting developments in the last decade. At the same time, Reinforcement Learning (RL) has emerged as a very successful paradigm for a variety of machine learning tasks. RL method usually finds an action policy that maximizes total cumulative reward over a given timeframe. With a combination of (deep) neural networks, it could handle complex features and make predictions about future states and rewards. However, existing RL - based optimization approaches are hampered by the generalization capacity and the increasing computational burden to accommodate the uncertainties. Some RL methods often analyze with some ideal assumptions without considering the abrupt changes and are hard to transfer to dynamic environments. We have designed a new paradigm of parallelized reinforcement learning (PRL) methods based on probabilistic events to handle the uncertainties. Several local (virtual) learning agents are employed to interact with pertinent environments in a distributed manner and report outcomes to the global agent. The advantage estimate functions of learning agents are designed with a backward sweep to transfer the outcomes to the value function updating process.
Biography:
