AdONE Seminar: Sven Seuken (University of Zurich), Layla Martin (TUM)

Events

Prof. Sven Seuken (University of Zurich): "Machine Learning-powered Iterative Combinatorial Auctions" – In this talk, I will discuss how machine learning can be used to improve the design of market mechanisms. I will present one particular case study in detail: a machine learning-powered iterative combinatorial auction (CA). The main goal of integrating machine learning (ML) into the auction is to improve preference elicitation, which is a major challenge in large CAs. In contrast to prior work, our auction design uses "value queries" instead of prices to drive the auction. The ML algorithm is used to help the auction mechanism decide which value queries to ask in every iteration. While using ML inside an auction introduces new challenges, we demonstrate how we obtain an auction design that is individually rational, has good incentives, and is computationally tractable. Via simulations, we benchmark our new auction against the well-known combinatorial clock auction (CCA). Our results demonstrate that the ML-powered auction achieves higher allocative efficiency than the CCA, even with only a small number of value queries.

*Joint work with Gianluca Brero (University of Zurich) and Benjamin Lubin (Boston University).

Layla Martin(TUM, AdONE): "A Comparison of Different Carsharing Relocation Modes: Classification and Feature-Based Selection"

Rebalancing and relocation in carsharing systems with respect to optimal routing received some attention in recent literature. Most research assumes that workers relocate one vehicle, and then use a bike to continue to the next location, but other modes, including public transport, second drivers, and trucks, exist as well. We study the economic viability of these modes in different environments. We consider a model in which employees of the operators drive vehicles from cold spots to hot spots, and then use bike, car, or public transit to continue to the next cold spot. Alternatively, vehicles can be loaded onto a truck. We formulate the problem as a hybridization of those different modes. The Multi-Mode Carsharing Relocation Problem (M-CRP) is modeled as a flow-based Vehicle Routing Problem (VRP) with multiple synchronization constraints and a heterogeneous fleet. Synchronization occurs among vehicles (every node is visited by at most one vehicle), but also between workers and vehicles (workers require vehicles to reach another location and vice versa). For this model, we present a branch-and-cut algorithm with Variable MIP Neighborhood Descent. Managerially, we observe that selecting the proper mode and in particular integrating different modes allows operators to save substantial costs. We support operators in their decision which mode(s) to choose for their fleet in their given environment and give hints as to which features of the environment (fleet, city …) drive this choice.

Joint Work with Stefan Minner and M. Grazia Speranza

Date: Monday, July 8th , 2019 (starting at 14:30)

Location: Faculty Club (left), IAS Garching

◄ Back to: Events