Research Tim Verdonck

Abstract

Due to technological advances, the available amount of data has increased exponentially over the last decade. The field of data science (DS) has followed this growth as it provides an indispensable tool for translating data into insight and knowledge. Where DS was traditionally concerned with learning associations in data, it has become clear in recent times that causal relations often provide a deeper understanding of the data and a stronger tool in many practical applications. One of the established approaches to causal modeling is to use a directed acyclical graph (DAG) to represent the causal relations. These DAGs have to be learned based on observed data. Many of the SOTA techniques for DAG learning are very sensitive to anomalies, and yield unreliable results in their presence. We aim to develop methods for DAG learning that remain efficient and reliable under contamination of the data. The project starts by building a solid foundation for the concepts of robustness in DAG learning. Building upon these foundations, we will then proceed to build a general robust DAG learning methodology. The project envisions three different but complementary approaches to the development of robust DAG learning methods. The developed methodology will be evaluated theoretically and empirically, and tested in a variety of real world cases.

Researcher(s)

• Promoter: Verdonck Tim
• Co-promoter: Latré Steven
• Fellow: Leyder Sarah

Research team(s)

• Applied mathematics

Project type(s)

• Research Project

Abstract

Process optimization is of crucial importance to businesses. Because of the involved complexity, businesses typically rely on domain experts for operating business processes. In this proposal, the aim is to develop a data-driven approach by leveraging recent advances in causal machine learning (CML) to support decision-making. To this end, Process Optimization with Sequential Individual Treatment Effects (POSITE) is introduced. POSITE is a powerful, reliable and flexible methodology that is capable of learning models to accurately predict causal effects in complex, sequential decision-making processes using causal machine learning. The resulting causal models are interpretable and robust, as such ensuring reliability and usability. The methodology is embedded within a cost-sensitive, constrained and stochastic decision-making framework to guide process operators in optimization process outcomes. The envisioned solution is an actionable and innovative approach toward process optimization that opens various directions for future fundamental and applied research.

Researcher(s)

• Promoter: Verdonck Tim

Research team(s)

• Applied mathematics

Project type(s)

• Research Project

Abstract

Conventional climate change mitigation alone will not be able to stabilise atmospheric CO2 concentrations at a level compatible with the 2°C warming limit of the Paris Agreement. Safe and scalable negative emission technologies (NETs), which actively remove CO2 from the atmosphere and ensure long-term carbon (C) sequestration, will be needed. Fast progress in NET-development is needed, if NETs are to serve as a risk-hedging mechanism for unexpected geopolitical events and for the transgression of tipping points in the Earth system. Still, no NETs are even on the verge of achieving a substantial contribution to the climate crisis in a sustainable, energy-efficient and cost-effective manner. BAM! develops 'super bio-accelerated mineral weathering' (BAM) as a radical, innovative solution to the NET challenge. While enhanced silicate weathering (ESW) was put forward as a potential NET earlier, we argue that current research focus on either 1/ ex natura carbonation or 2/ slow in natura ecosystem-based ESW, hampers the potential of the technology to provide a substantial contribution to negative emissions within the next two decades. BAM! focuses on an unparalleled reactor effort to maximize biotic weathering stimulation at low resource inputs, and implementation of an automated, rapidlearning process that allows to fast-adopt and improve on critical weathering rate breakthroughs. The direct transformational impact of BAM! lies in its ambition to develop a NET that serves as a climate risk hedging tool on the short term (within 10-20 years). BAM! builds on the natural powers that have triggered dramatic changes in the Earth's weathering environment, embedding them into a novel, reactor-based technology. The ambitious end-result is the development of an indispensable environmental remediation solution, that transforms large industrial CO2 emitters into no-net CO2 emitters.

Researcher(s)

• Promoter: Latré Steven
• Co-promoter: Oramas Mogrovejo José Antonio
• Co-promoter: Verdonck Tim

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

Just like many industries today, the accountancy sector is also confronted with disruptive digitalization. This digitization means that accountants are expected to provide more and more proactive services, where the focus used to be on executive and compliance-related work. With our project we want to help accountants to fulfill these new expectations. By applying advanced statistical methods and machine learning techniques, we want to focus strongly on following two research topics. First of all, we want to test and develop different methods to discover anomalies ​​in accounting data. This helps the accountant to automate standard checks, but also to discover potential opportunities. Secondly, we want to test and develop robust and interpretable cash flow forecasting models. In both areas we are convinced that there is still enormous potential to create added value for the accountant. The collaboration with Boltzmann provides the ideal context for this project due to the presence of a rich, ever-expanding dataset, combined with professional expertise in various areas within the framework project team.

Researcher(s)

• Promoter: Verdonck Tim

Research team(s)

• Applied mathematics

Project type(s)

• Research Project

Abstract

In this IOF POC, we aim to bring together the expertise in sports science and machine learning to develop intelligent tools for coaching endurance sports. These tools will assist the coach in tracking the fitness level of the athletes and provide early warning of any potential issues within the physiological data. By leveraging the power of machine learning, we hope to create a more efficient and effective coaching process that can help athletes reach their full potential. Additionally, by integrating sports science knowledge, we aim to ensure that the tools we develop are grounded in the latest research and understanding of how the body responds to endurance training.

Researcher(s)

• Promoter: Verdonck Tim
• Co-promoter: Latré Steven

Research team(s)

• Applied mathematics

Project type(s)

• Research Project

Abstract

The project targets (i) a novel, virtual and distributed data ecosystem for food delivery to physical shops that becomes hyper responsive and efficient thanks to (ii) more accurate forecasting and personalization models that use enhanced AI and scheduling technologies to (iii) optimize the end-to-end logistics from farmers to Distribution Centers (DCs) to stores and customers.

Researcher(s)

• Promoter: Latré Steven
• Co-promoter: Hellinckx Peter
• Co-promoter: Verdonck Tim

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

The Generalized Linear Model (GLM) is a very popular and flexible class of regression models that generalizes ordinary linear regression by allowing for example non-normal response variables. Logistic regression, which is widely used for binary classification, and Poisson regression, often used to model count data, both belong to this class. The parameters are typically estimated using maximum likelihood, but this very often leads to various problems when analyzing real data from practice. Firstly, outliers in the data may heavily influence classical methods, yielding unreliable results. Secondly, estimation and interpretability becomes very difficult or impossible when the number of variables becomes very high. Thirdly, real data often display a more complex dispersion behavior than expected under the GLM model. To solve these issues, sparse and robust estimation methods that model simultaneously the mean and the dispersion behavior in the context of GLMs will be developed. Their mathematical properties will be thoroughly investigated. The newly proposed methods should also be computationally efficient such that modern large datasets can be analyzed easily. Open-access user-friendly software will be provided.

Researcher(s)

• Promoter: Verdonck Tim
• Fellow: Tubex Lise

Research team(s)

• Applied mathematics

Project type(s)

• Research Project

Abstract

Climate change will affect Arctic ecosystems more than any other ecosystem worldwide, with temperature increases expected up to 4-6°C. While this is threatening the integrity and biodiversity of the ecosystems in itself, the larger ecosystem feedbacks triggered by this change are even more worrisome. During millions of years, atmospheric carbon has been stored in the Arctic soils. With warming, the carbon can rapidly escape the soils in the form of CO2 and (even worse) the strong greenhouse agent CH4. Despite decades of research, scientists still struggle to unveil the scale of this carbon exchange, and especially how it will interact with climate change. An overarching question remains: how much carbon will potentially escape the Arctic in the future climate, and how will this affect climate change? FutureArctic embeds this research challenge directly in an inter-sectoral training initiative for early stage researchers, that aims to form "ecosystem-of-things" scientists and engineers at the ForHot site. The FORHOT site in Iceland offers a geothermally controlled soil temperature warming gradient, to study how Arctic ecosystem processes are affected by temperature increases as expected through climate change.

Researcher(s)

• Promoter: Latré Steven
• Co-promoter: Verdonck Tim

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project