We develop a statistical stopping rule for iterative sampling-based algorithms for stochastic programmes, such as stochastic dual dynamic programming (SDDP) and sample average approximation (SAA). We make use of recent advances in sequential stopping theory based on e-values.
joint project with Nick Koning
- Speaker
- Date
- Friday 18 Jul 2025, 12:15 - 13:00
- Type
- Seminar
- Room
- ET-14
- Building
- E Building
We train supervised machine learning models on historical bookings with perfect information to compute a request fit that quantifies each new request’s expected impact on the overall route. This request fit predictor is combined with either a heuristic or an event-based solver approach to accept or reject requests in real-time. Tested on real on-demand bus data, this hybrid framework increases served passengers by 16% compared to the current system.
The second study represents ongoing research into user behaviour, and here we report preliminary ideas and results. In the current workflow, users enter origin, destination, and a preferred pickup time, and the system then offers the nearest available slots. We introduce an “offer-only” interface: users provide only origin and destination, and the system suggests optimized time slots.
We want to evaluate this in online experiments to test two behavioural effects: (1) removing the pickup-time choice to measure endowment-driven flexibility, and (2) varying the number and deviation of proposed slots to assess assortment sensitivity. We give empirical evidence into users’ sensitivity to offered time slots under the current system, and want to apply our “offer-only” experimental findings to the same dataset to quantify operational improvements.
Coordinators
See also
- More information
Lunch will be provided (vegetarian option included).
For more information please contact the Secretariat Econometrics at eb-secr@ese.eur.nl
