CAIML #34
CAIML #34 happened on January 21, 2025, at congstar.
Agenda
18h30 Open Doors
19h00 Welcome & Intro
19h15 Jonas Deichmann (Data Scientist at congstar): Within the Horizon: Forecasts for Service Hotlines
Join us for a discussion on how machine learning can improve call center operations. We’ll cover:
- Forecasting Techniques: Using models like XGBoost to predict contact volumes.
- Operational Efficiency: Enhancing agent planning with accurate forecasts.
- Cost Management: Reducing costs through precise forecasting.
We’ll also discuss the role of explainable AI in ensuring transparency.
19h50 Bonald Ziyue LI (Assistant Professor at University of Cologne): One Fits All? Towards a Generalist Agent for Smart Mobility based on Meta Reinforcement Learning
Smart mobility and transportation are essential components of modern smart cities. Real-time decision-making in smart transportation systems presents significant spatial-temporal complexity. Reinforcement learning (RL)-based agents have emerged as promising solutions for automating such decisions, applied in areas such as autonomous driving, route planning, and traffic signal control—integrating vehicles, humans, and infrastructure into cohesive systems. However, most current RL models are environment-specific, meaning a model trained in one city (e.g., Cologne) is typically effective only within the same environment. This lack of scalability and generalization poses a critical challenge. This talk explores the potential to develop a generalist agent capable of overcoming this limitation—one that is trained in Cologne but deployable in Berlin. Using traffic signal control as a case study, we will discuss innovative approaches to training, model design, and knowledge sharing that enable the development of such adaptable, general-purpose agents.
20h20 Networking with food and drinks provided by congstar