Happening @ Michigan

Happening@Michigan

The University of Michigan Events Calendar

Sponsors

No results

Keywords

No results

Types

No results

Search Results

Events

No results
Search events using: keywords, sponsors, locations or event type
When / Where
Free Presentation

Presented By: Michigan Robotics

Ask a Question About This Event

Learning Structured Reward Representations for Reinforcement Learning

PhD Defense, Mohamad Louai Shehab

Chart with text blocks: Learning Structured Reward Representations Applications Identification Anomaly Detection using mixture-IRL Chapter 4 Learning Reward Machines From Policies Chapter 6 Identifiability Learning Time-Varying Rewards with Priors Chapter 7 Linear Algebraic Characterization of Identifiability in IRL Chapter 5 Learning Reward Machines with Unknown Labels Chapter 8 Chart with text blocks: Learning Structured Reward Representations Applications Identification Anomaly Detection using mixture-IRL Chapter 4 Learning Reward Machines From Policies Chapter 6 Identifiability Learning Time-Varying Rewards with Priors Chapter 7 Linear Algebraic Characterization of Identifiability in IRL Chapter 5 Learning Reward Machines with Unknown Labels Chapter 8
Chart with text blocks: Learning Structured Reward Representations Applications Identification Anomaly Detection using mixture-IRL Chapter 4 Learning Reward Machines From Policies Chapter 6 Identifiability Learning Time-Varying Rewards with Priors Chapter 7 Linear Algebraic Characterization of Identifiability in IRL Chapter 5 Learning Reward Machines with Unknown Labels Chapter 8
Committee Chair: Necmiye Ozay

Abstract:
Learning from demonstrations is one of the most natural and powerful paradigms for acquiring intelligent behavior. Among its most prominent formulations is inverse reinforcement learning (IRL), which has achieved remarkable success in domains such as robotics and autonomous driving. Yet despite this promise, IRL remains fundamentally challenged by ill-posedness: the same observed behavior can often be explained by many different underlying objectives. If an agent moves from point A to point B, is it because it is attracted to B, repelled by A, or optimizing something more subtle entirely? From behavior alone, the answer is generally ambiguous.
This dissertation addresses several foundational questions at the heart of inverse reinforcement learning. First, it investigates which environments and dynamics are inherently suitable for learning from demonstrations. Second, it studies how "rationality priors" can be incorporated to regularize the ambiguity of IRL and improve identifiability. Finally, it develops methods for inferring latent memory structure directly from demonstrations, enabling classical IRL techniques to extend beyond Markovian assumptions to the richer setting of multi-stage tasks with history-dependent, non-Markovian rewards.

Robotics Atrium & on Zoom (Passcode: 19982001)
Chart with text blocks: Learning Structured Reward Representations Applications Identification Anomaly Detection using mixture-IRL Chapter 4 Learning Reward Machines From Policies Chapter 6 Identifiability Learning Time-Varying Rewards with Priors Chapter 7 Linear Algebraic Characterization of Identifiability in IRL Chapter 5 Learning Reward Machines with Unknown Labels Chapter 8 Chart with text blocks: Learning Structured Reward Representations Applications Identification Anomaly Detection using mixture-IRL Chapter 4 Learning Reward Machines From Policies Chapter 6 Identifiability Learning Time-Varying Rewards with Priors Chapter 7 Linear Algebraic Characterization of Identifiability in IRL Chapter 5 Learning Reward Machines with Unknown Labels Chapter 8
Chart with text blocks: Learning Structured Reward Representations Applications Identification Anomaly Detection using mixture-IRL Chapter 4 Learning Reward Machines From Policies Chapter 6 Identifiability Learning Time-Varying Rewards with Priors Chapter 7 Linear Algebraic Characterization of Identifiability in IRL Chapter 5 Learning Reward Machines with Unknown Labels Chapter 8

Related Links

Explore Similar Events

  •  Loading Similar Events...

Keywords


Back to Main Content
About Help Terms of Use Event Feeds Import Events Ask About the Calendar

Contact Us