How does one introduce decision support tools in transport logistics, in order to exchange information more easily with those who use them? That's the key question that Professor Vidal and his team will tackle over the next five years with the SCALE AI Research Chair in Data-Driven Supply Chains, recently awarded to the researcher

Over the course of his career, Professor Thibaut Vidal has acquired a solid expertise in the development of optimization and machine learning algorithms for supply chains. As part of the work carried out by his Chair, he seeks to, on one hand, simplify decision-support tools, and also to make them more secure and transparent.

"Decision support tools provide an answer, without users being able to question those decision-making algorhythms, in order to understand what led to the proposed solution," explains Professor Vidal. “It’s often beneficial to have more transparency and control over the decision-making process, to provide the user with more confidence, and also to guide methods diagnosis and improvement. "

The researcher cites the example of a request to get more financial credit. After providing a series of data to an algorithm, the user receives one of two responses: yes or no. By allowing the user to better understand the algorithm's decisions and to learn why they are denied or awarded credit, users will be able to better position themselves in order to increase their chances of obtaining a positive response in a subsequent assessment.

Very similar situations to the above arise in supply chains, when it comes to selecting a next destination for a driver or an ambulance, whether or not to accept a request or a proposed schedule, whether or not to carry out preventive maintenance, or even how to react when faced with an unforeseen event.

"In other words, the tool will not only make a decision, it will also have to answer the question "why"," explains Thibaut Vidal.

Interpretable machine learning

This type of question and analysis is already occurring in the field of machine learning, in an area of study called "interpretable machine learning." The Polytechnique Montréal team relies on this approach to improve tools that have become vital for many public and private organizations.

Take the case of a road transport company that manages around 1,000 trucks, and has to deliver products to 1,000 customers - all on the same day. A traffic jam forms in one place. A section of highway closes elsewhere. How can optimized routes be created in real-time, so as to simultaneously satisfy customers, ensure good working conditions for drivers, and guarantee company profitability? This is where decision-making tools come into play in the transport sector - and while we're already getting there in terms of advances, we can do better, urges Professor Vidal.

“With this type of optimization problem, above all else, the sheer number of possible solutions grows exponentially with an increase in data. Dealing with such problems requires very sophisticated resolution models, but those are all the more difficult to implement due to their complexity," he notes.

To overcome this problem, the researcher will focus on three areas of research. The first axis will focus on the development of mathematical optimization algorithms qualified as “classic.” The second axis will focus on interpretable machine learning algorithms, such as decision tree forests. The third axis will focus on creating a bridge between the first two axes, in order to take advantage of the efficiency of the first axis and the transparency of the second axis.

Three Chairs in Transport Logistics

The Research Chair led by Professor Vidal will benefit from a budget of $2 million over 5 years, to which SCALE AI and Polytechnique Montréal will contribute equally. This Chair is part of a group of three chairs specializing in transport logistics whose creation was announced today by SCALE AI. The other two chairs will be based at HEC Montréal and at the University of Toronto.
Funded by the federal government and the Government of Quebec, SCALE AI has nearly 120 industrial partners, research institutes, and other stakeholders in artificial intelligence.
The supercluster develops programs focused on supporting companies' investment projects that implement concrete applications in artificial intelligence, support the emergence of future flagships in the sector, and help spur on the development of a qualified workforce.

Learn more

Professor Thibaut Vidal expertise
SCALE AI website