Abstract: Sensors are needed to detect and quantify environmental phenomena. In an ideal situation, one would have unlimited sensor capacity to monitor everything needed in order to have the full picture and make accurate decisions in our problem (i.e. full observation setting). However, most of the times sensors are limited (and cost to operate them) and so they need to be carefully positioned in in a partially observable environment in order to maximise their usefulness, i.e. without having the full picture on how the world looks like. This problem is common in a continuous of applications, ranging from adversarial settings like scheduling checkpoints at airports and protecting wildlife from poaching efforts as well as non strategic settings like prescribing how to observe the universe with a limited number of sensors (i.e. telescopes) to maximise number of events detected.
The talk will present how to gather information from an unknown partially observable environment through a sensor network in order to maximise the sum events detected in the long run. The focus will be on the applications rather than the techniques and will use two examples: and adversarial patrolling with IoT devices and a non-adversarial setting of observing events in a 2D abstraction of the universe.
Speaker: Enrique Munoz de Cote, Director of Multi-Agent Science, PROWLER.io
Bio: Enrique is Director of multiagent science at Prowler.io with more than 60 academic papers published in top-tier journals and conferences in AI, including a book on multi agent learning. His interests lie in the intersection between machine learning, AI and economic theory. He has been recognised in numerous occasions, including a best paper at UAI and likes testing his team's ideas on computer science competitions. He is a member of the board of directors of the Association for Trading Agent Research (ATAR) and several program committee boards in machine learning and artificial intelligence. His team at Prowler is working on controlling network problems such as traffic flows and supply-chain inventory optimisation and designing technology to port mechanism design concepts into a data driven world.