Mars 2022

Retour à la vue des calendrier
Jeudi 17 Mars
Heure: 12:30 - 13:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: A Tailored Benders Decomposition Approach for Last-mile Delivery with Autonomous Robots
Description: Ivana Ljubic This work addresses an operational problem of a logistics service provider that consists of finding an optimal route for a vehicle carrying customer parcels from a central depot to selected facilities, from where autonomous devices like robots are launched to perform last-mile deliveries. The objective is to minimize a tardiness indicator based on the customer delivery deadlines. We provide a better understanding of how three major tardiness indicators can be used to improve the quality of service by minimizing the maximum tardiness, the total tardiness, or the number of late deliveries. We study the problem complexity, devise a unifying Mixed Integer Programming formulation and propose an efficient branch-and-Benders-cut scheme to deal with instances of realistic size. Numerical results show that this novel Benders approach with a tailored combinatorial algorithm for generating Benders cuts largely outperforms all other alternatives. In our managerial study, we vary the number of available facilities, the coverage radius of autonomous robots and their speed, to assess their impact on the quality of service and environmental costs. Joint work with: L. Alfandari and M.M. de Silva
Lundi 28 Mars
Heure: 13:00 - 14:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Abstractive Summarization Evaluation: Overview and Reflections
Description: Yanzhu Guo The topic of summarization evaluation has recently received a surge of attention due to the rapid development of abstractive summarization systems. We conduct a survey of the state-of-the-art evaluation metrics along with relevant datasets and visualization systems. We also touch upon the statistical deficiencies in current meta-evaluation approaches such as the problematic choice of scoring range, the lack of paired evaluation as well as the prevalence of underpowered tests. Finally, we show experimental results proving the unreliability of human-annotated ground-truth reference summaries and thus argue for reference-free metrics as a more promising future direction.