Talks and presentations

STEMO: Early Spatio-Temporal Forecasting with Multi-Objective Optimization

25 August 2024 – 29 August 2024

KDD Conference Presentation, KDD 2024, Barcelona, Spain

Paper Summary This work introduces STEMO, a spatio-temporal forecasting framework designed for early-stage prediction under limited or evolving observations. The model integrates spatial dependencies and temporal dynamics while explicitly optimizing multiple objectives, such as predictive accuracy and robustness.

The key contribution lies in addressing the trade-offs inherent in early forecasting scenarios, where incomplete signals and dynamic environments challenge conventional models. By incorporating structured spatial representations and multi-objective learning, STEMO improves generalization and stability in real-world forecasting tasks.

Long-term Fairness in Ride-Hailing Platforms

9 September 2024 – 13 September 2024

ECML PKDD Conference Presentation, ECML PKDD 2024, VILNIUS, LITHUANIA

Paper Summary This paper investigates long-term fairness mechanisms in ride-hailing platforms, focusing on balancing efficiency with equitable treatment of drivers over time. Traditional allocation and pricing systems often optimize short-term revenue or matching efficiency, potentially leading to cumulative inequities.

The study formulates fairness as a long-horizon optimization problem and proposes algorithmic mechanisms that regulate allocation policies to mitigate persistent disparities. The framework provides theoretical insights and empirical evaluation, demonstrating how fairness-aware policies can improve equity without severely degrading system efficiency.