KMSD is not only familiar with the three main classes of modelling and simulation :
♦ Multi-agent representation and interactions,
♦ Equation and physical and / or mathematical modelling of cyberphysical systems,
♦ Statistical models and learning from big data,...
but members of its network of experts were at the origin of advances in the state of the art
To mention a few :
♦ high-level tools making multi-agent modelling accessible to general-purpose users [Prof. Hervé Zwirn, working on complex systems ]
♦ multi-physical and multi-scale modelling [Prof. Jean-Michel Ghidaglia, working on mechanics of fluids]
♦ original methods of unsupervised statistical learning and scoring [Prof. Nicolas Vayatis, recommendations for marketing or finance applications]
♦ internationally recognised image processing algorithms [Prof. Jean-Michel Morel]
♦ a collective contribution to system modelling via the open source Modelica language.
Agents are people, groups of people, objects, or groups of objects. Their attributes and behaviours are modelled in relation to the context:
– birth -creation / death -disappearance,
– buy or sell of this or that,
– movement or not,
– increase or decrease of this or that,
– interaction, with description)…
System simulation that takes account of the environment directly provides results and indicators. The method is powerful but its field of application is limited to systems that can be represented in a relevant way.
The system or process is described by equations derived from mathematical and / or physical paradigms (including chemistry, biology, etc.).
A direct solution is most often out of reach, and simulation itself is a challenge. The quality and performance of simulation is a key factor in competitiveness and competitive differentiation for advanced industrial activities (multifluid dynamics, combustion, biology…).
System level simulation is now possible due to the latest high-level languages, such as Modelica.
The Machines, or algorithms, ‘learn’ from the data (often very large in size, hence the name ‘big data’), and can thus propose or make decisions as part of a process. Examples include:
– automotive : identifying fixed objects as environmental references to enable assisted driving,
– industry: estimate equipment reliability in real time and allow predictive maintenance and optimization of repair operations,
– logistic : predict all demands in a supply chain,
– industry : optimize production processes,
– finance : assess behaviour to better manage risks in different types of applications.
The aim is to put data or events into categories, described in qualitative or quantitative terms.
If these categories are predefined, learning is said to be supervised; if they are not learning is said unsupervised. When the description of categories evolves through learning, feedback is called enhanced learning.
Unsupervised learning, also called the predictive learning, is scientifically the most ambitious, and is thought to be a major factor in the future transformations of many activities.