When Simulations Get People Wrong
Julie Dugdale, Laboratoire d'Informatique de Grenoble, France, France
Interpolating Multidimensional Sparse and Dense Reduced Data
Ryszard Kozera, Warsaw University of Life Sciences, Poland, Poland
Available Soon
Navonil Mustafee, University of Exeter, United Kingdom, United Kingdom
Brief Bio
Julie Dugdale is a Full Professor of Computer Science at the University Grenoble Alps and a member of the Grenoble Informatics Lab (LIG), the largest Computer Science research lab in France. She has been working in the area of simulation for over 30 years and published widely in the area. She is involved in several European and National projects where her involvement focuses on modelling and simulating human behaviour. From 2010 to 2020 she was the head of the HAwAI (Human Aware Artificial Intelligence) Research Team at LIG. She has now formed another research team called SocSIM-K for Social Simulation and Knowledge. She is an ex-President of the ISCRAM (Information Systems for Crisis Response and Management) Association and she obtained the ISCRAM Distinguished Service Award in 2010. She is now an Honorary Member of ISCRAM. Julie Dugdale is also the ex Vice-Chair of the IFIP (International Federation for Information Processing) Work Group on Information Technology for Disaster Risk Reduction (WG5.15). She is currently a member of the ELLIS pan-European AI network of excellence. She has worked in many countries, for example, she was a Full Adjunct Professor (20% post) at CIEM (Centre for Integrated Emergency Management) at the University of Agder in Norway. During a CNRS sabbatical she also worked as an invited researcher at CSIRO (The Commonwealth Scientific and Industrial Research Organisation), Melbourne, Australia where she worked on human behaviour in bushfires.
Her research concerns modelling aspects of human behaviour at the cognitive, work and societal levels using an agent-based approach. Often, but not always, her agent-based models are used for simulations. Broadly, her work falls into the domain of Agent-based Social Simulation (ABSS). Following a background in artificial intelligence, she is primarily interested in cognition and interaction. Specifically, modelling the cognitive activities of human behaviour, the cognitive supports in our work environment and how groups of people interact in order to accomplish a task. She has applied her work to many areas, but she is particularly
interested in emergency rescue, crisis management, energy management, and pedestrian mobility in cities.
Abstract
Human behaviour in crisis and emergency situations, particularly during evacuation, can affect our survival and the survival of others. Models and simulators have been proven to be useful tools to explore evacuation scenarios in crisis situations, for example, in the case of forest fires, earthquakes, and flash floods, etc. However, they often lack a realistic focus on human behaviours. Indeed, they tend to model what citizens are supposed to do, rather than what they actually do. Incorporated behaviours often assume that people will evacuate immediately, know the location (and shortest path) to a safe area, and move autonomously to it without regard for their family or friends. As we have seen, in reality, this is not the case. Instead, human behaviours are influenced by a multitude of interconnected factors such as, social ties, attachment to objects or places, our own knowledge (e.g. training received), previous experience with the crisis, etc. This talk will explore aspects affecting our behaviours in crisis situations and, drawing up some of our recent works I will show how they may be modelled and simulated using an agent-based approach.
Brief Bio
Prof. Ryszard KOZERA is currently a Director of The Institute of Information Technology at Warsaw University
of Life Sciences - SGGW in Poland. He obtained a Habilitation Degree (DSc) from Silesian University of
Technology (Poland) in 2006, a PhD from Flinders University of South Australia (Australia) in 1991 and a MSc
Degree from Warsaw University (Poland) in 1985. Selected previous research positions include: Assoc. Prof. at
The University of Western Australia, Perth, Australia 1991-2008 (current adjunct Assoc. Prof.) and Prof. at
Warsaw University of Technology, Poland 2009-2012. In addition, he was also awarded three times Alexander
von Humboldt Research Fellowships (Technical University of Berlin 1996-97; Christian Albrechts University of
Kiel, 2000, 2004) in Germany. His research interests are: optimization, interpolation and approximation, computer
vision and image analysis, numerical analysis, neural computation and networks, artificial intelligence, partial
differential equations and in general applied mathematics and modeling and simulation of processes in physics,
engineering, biology, medical sciences, computer science and agriculture. He published over 160 scientific papers
in international journals, research monographs and conference proceedings. He organized numerous international
conferences, workshops and sessions among all in Australia, Germany, Greece, Mexico, New Zealand and Poland.
He delivered over 70 talks and conference presentations world-wide. Prof. R. Kozera served also as co-editor of
conference proceedings, research monographs, journal, conference and book reviewer as well as he has also won
international grants and rewards. He was the assessor of the EU postdoctoral projects (Marie-Curie Fellowships).
He supervised over 85 BSc, Honours or MSc students and 4 PhD conferred students in Australia and Poland. He
delivered lectures on over 30 different topics at the undergraduate, postgraduate and PhD levels in computer
science (computer vision, artificial intelligence, neural computation) and applied mathematics (numerical
analysis, interpolation, optimization and partial differential equations). Full publication list can be found under:
https://bw.sggw.edu.pl/
Abstract
The problem of fitting a given ordered sample of data points in arbitrary Euclidean
space is discussed for both dense and sparse data. The corresponding knots are
assumed to be unknown and must first somehow be found. Various recipes
replacing the unknown knots for different interpolation schemes are addressed and
analyzed. For sparse data the latter leads to a highly non-linear multivariate
optimization task, equally non-trivial for theoretical analysis and for derivation of a
computationally efficient numerical scheme. Differtent numerical algorithms
including Leap-Frog are compared. In case of dense data the issue of convergence
of the interpolant to the unknown curve (together with the resulting order rate and
its sharpness) is discussed. Illustrative synthetic and real examples supplement the
analysis in question.