Speaker: Prof.YU Jiayuan Time: Thursday 14:00 p.m., 18th January Place: A722 SINANO Topic 1: Privacy-preserving resource allocation An unknown number of agents consume a common good resource in a repeated fashion. We show that if the agents control their consumptions according to a certain policy, then these consumptions converge to a socially efficient allocation. This is achieved without communicating any of the utility functions of the agents, and with very limited broadcast signals from a central authority. We present one such policy for divisible goods and one for indivisible goods. Topic 2: Machine learning for decision-support systems This talk presents the challenges and approaches for developing two decision-support systems. The INSIGHT (insight-ict.eu) system aims to reduce road congestion and improve emergency response by building a decision-support system for control-room operators. This system, deployed at Dublin city council and a German federal agency, analyses large amounts of real-time data from heterogeneous sources (dedicated sensors and social media), crowdsourcing tasks to smartphone users, and outputs warnings. The Indus.ai system aims to improve the safety, efficiency, and accountability of large-scale construction projects. This system uses computer vision and is currently deployed in constructions sites of new subway stations in Toronto.
|