Banner
Home      Log In      Contacts      FAQs      INSTICC Portal
 
Documents

Keynote Lectures

Is SimDec Truly a Revelatory Approach for Global Sensitivity Analysis or is it Turtles All the Way Down?
Julian Scott Yeomans, York University, Canada

Autonomous Digital Twins for Optimal Control of Discrete Event Systems
Andrea Matta, Politecnico di Milano, Italy

Modeling and Simulation of Dense Crowds Dynamics at the Intersection of Agent-based and Deep-Learning Models: Predict and Understand
Benoit Gaudou, University Toulouse 1 Capitole, France

 

Is SimDec Truly a Revelatory Approach for Global Sensitivity Analysis or is it Turtles All the Way Down?

Julian Scott Yeomans
York University
Canada
 

Brief Bio

Julian is a Professor of Operations Management & Information Systems and the Program Director for both the Master of Management in Artificial Intelligence and the Master of Business Analytics at the Schulich School of Business, York University, Toronto. In collaboration with Mariia Kozlova (LUT University), Julian has created SimDec, which has been used as a technique for global sensitivity analysis. SimDec combines visual uncertainty analytics with an innovative computational method for identifying and quantifying the influence of factor impacts. Recent application studies have examined small modular nuclear reactors, 3D printing in construction, agricultural food-water-energy systems, aviation electrification, healthcare, and superconducting magnets at CERN. To promote the widest adoption and penetration of SimDec as possible, a downloadable free-of-charge electronic book, together with open-source computer code in Python, Julia, R, and Matlab and a “no-code-required” web dashboard, have been made freely available. For a “low-tech” overview of SimDec you can read Julian’s recent interview in the Schulich Research Newsletter.


Abstract
SimDec (“simulation decomposition”) is a recently developed analytical approach that enables a visualizable analysis of impacts and interactions within data. Such visualizations can be easily understood and interpreted by all users regardless of technical background. While straightforward and elegant, SimDec enhances explanatory capabilities by visually “teasing out” inherent cause-and-effect relationships, while also uncovering counter-intuitive behaviours. Recent studies have indicated that SimDec might be considered the pre-eminent technique for conducting applied, “real world” global sensitivity analysis. Could such research revelations truly herald the second coming or is SimDec simply esoteric rot – nothing but turtles all the way down? You be the judge.



 

 

Autonomous Digital Twins for Optimal Control of Discrete Event Systems

Andrea Matta
Politecnico di Milano
Italy
 

Brief Bio
Andrea Matta is Full Professor of Manufacturing at Department of Mechanical Engineering of Politecnico di Milano where he develops his teaching and research activities since 1998. He was Distinguished Professor at the School of Mechanical Engineering of Shanghai Jiao Tong University from 2014 to 2016. He was visiting scholar at Ecole Centrale Paris, University of California at Berkeley, and Tongji University. His research area includes analysis, design and management of manufacturing and health care systems. He is Editor in Chief of Flexible Services and Manufacturing Journal since 2017. He was awarded with the Shanghai One Thousand Talent and Eastern Scholar.


Abstract
With the coming of the Industry 4.0 wave, digital representations of production systems havebeen promoted from marginal to central. Digital twins are not simply conceived as simulation models of their physical counterparts for offline what-if analysis, differently they are developed as self-adaptable and empowered decision-makers timely aligned with the dynamics of the real system. Enriched by these new features, digital twins are widely recognized as the key enablers for the implementation of the smart manufacturing paradigm. Despite this new role, there are significant barriers to the adoption of the digital twin concept in industrial applications. The creation and continuous update of digital twin models is still a challenge because of the high skills required to use the simulation applications available in the market, the long development times, and their difficult integration with optimization and artificial intelligence packages. The frequent changes manufacturing systems encounter in their life cycle boost these issues. This talk describes data-driven approaches for generating, synchronizing, and validating multi-perspective models for digital twins of discrete event systems from sensor data.



 

 

Modeling and Simulation of Dense Crowds Dynamics at the Intersection of Agent-based and Deep-Learning Models: Predict and Understand

Benoit Gaudou
University Toulouse 1 Capitole
France
 

Brief Bio
Benoit Gaudou is a Professor at Université Toulouse Capitole in Computer Science and a member of the Research Institute in Computer Science of Toulouse (IRIT). He get his Ph.D. in Artificial Intelligence from the University of Toulouse in 2008, focusing on the modeling and formalization of social attitudes in modal logic and their application to agent communication languages. His research now focuses on agent-based modeling and simulation of complex socio-environmental systems, with a particular interest in integrating cognitive agents into models, building large-scale models integrating multi-disciplinary dynamics and combining various modeling approaches.
Since 2010, he is actively involved in the development and training of the open-source GAMA platform (https://gama-platform.org/), an agent-based modeling and simulation platform.


Abstract
Trustworthy models for the dynamics of dense crowds are crucial for the prediction of pedestrian flows and the management of large crowds, but also from a fundamental perspective, to understand the roots that they share with active matter but also the pedestrian specifics. However, current models suffer from some severe deficiencies, especially at high density and for real and large-scale situations. In this keynote, I will discuss limitations of existing approaches and discuss various improvments to tackle these challenges through a triple approach combining agent-based, physics-based and data-driven modeling. Finally, I will introduce and discuss the current research trends combining agent-based models and prediction-oriented modeling approaches (and in particular Machine Learning and Large Language Models).



footer