BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260523T215601EDT-5728PluWmR@132.216.98.100
DTSTAMP:20260524T015601Z
DESCRIPTION:Abstract\n\nIntelligent Transportation Systems (ITSs) represent
  an integration of advanced technologies into transportation infrastructur
 e to enhance efficiency\, safety\, and sustainability. Recent advancements
  in Artificial Intelligence (AI) have boosted the development of ITSs\, en
 hancing urban mobility\, reducing congestion\, and improving sustainabilit
 y. However\, AI-enhanced ITSs also face significant challenges in achievin
 g real-world scalability due to the high sample complexity of data-driven 
 learning algorithms\, e.g.\, machine learning (ML) and reinforcement learn
 ing (RL)\, which require vast amounts of data for training and adaptation.
  This thesis explores multiple strategies to improve sample efficiency in 
 two critical real-world ITS domains: charging load forecasting of electric
  vehicle (EV) charging stations and traffic signal control (TSC).\n\nTo ad
 dress data scarcity in EV charging stations\, we propose MetaProbformer\, 
 a Transformer-based meta-learning approach for probabilistic load forecast
 ing. It enables rapid adaptation to new stations with limited historical d
 ata through meta-training across diverse datasets. The method achieves rob
 ust performance across unseen scenarios\, reducing reliance on extensive t
 raining samples on target charging stations while maintaining forecasting 
 performance.\n\nFor TSC\, we present two model-based and one offline-to-on
 line RL methods: ModelLight\, FM2Light\, and DTLight. ModelLight is a mode
 l-based meta-RL framework for optimizing single signalized intersections. 
 In this framework\, world models capturing the dynamics of signalized inte
 rsections are acquired and employed to generate imaginary trajectories wit
 hin an optimization-based meta-learning approach. It enables traffic signa
 l controllers to quickly adapt to varying traffic patterns with minimal ad
 ditional training\, improving both sample efficiency and generalization ab
 ility. For multi-intersection TSC scenarios\, we propose FM2Light\, a mode
 l-based multi-agent RL framework that enhances sample efficiency and intro
 duces fairness constraints to ensure equitable traffic flow distribution\,
  addressing both congestion reduction and social equity in urban networks.
  FM2Light employs an ensemble of global world models and a fairness-refine
 d reward structure to enhance both sample efficiency and intersection-leve
 l fairness in large-scale deployments. Additionally\, we propose DTLight\,
  a lightweight Transformer-based traffic signal controller that leverages 
 offline historical data to pre-train control policies\, followed by online
  fine-tuning with real-time feedback. By doing so\, we aim to take advanta
 ge of the high sample efficiency of offline RL while incorporating valuabl
 e feedback from the online environment. This hybrid approach minimizes the
  need for costly online exploration while adapting dynamically to changing
  traffic conditions.\n\nCollectively\, these methods in our thesis aim to 
 provide practical\, sample-efficient solutions for ITS applications\, redu
 cing the need for large datasets and expensive real-world experimentation.
  The proposed approaches aim to improve forecasting accuracy and traffic m
 anagement\, contributing to more intelligent\, responsive\, and sustainabl
 e urban transportation systems.\n
DTSTART:20251111T140000Z
DTEND:20251111T160000Z
LOCATION:Room 603\, McConnell Engineering Building\, CA\, QC\, Montreal\, H
 3A 0E9\, 3480 rue University
SUMMARY:PhD defence of Xingshuai Huang – Sample-Efficient Algorithms for In
 telligent Transportation Systems
URL:/ece/channels/event/phd-defence-xingshuai-huang-sa
 mple-efficient-algorithms-intelligent-transportation-systems-368689
END:VEVENT
END:VCALENDAR