11 PhD Positions in the Field of Artificial Intelligence and Machine Learning
KU Leuven ESAT - STADIUS, Stadius Centre for Dynamical Systems, Signal Processing and Data Analytics
Belgium

11 PHD POSITIONS IN THE FIELD OF ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING



This project is hosted by the STADIUS Center for Dynamical Systems, Signal Processing, and Data Analytics group of the Department of Electrical Engineering (ESAT) at KU Leuven. STADIUS is an academic research center, with a research focus on mathematical engineering, where mathematical tools from numerical linear and multilinear algebra, statistics and optimization are engineered for applications of dynamical systems and control, signal processing, data modeling and analytics.

 

Project

11 PhD positions will be funded by the impulse program on Artificial Intelligence (AI) Research, which is an initiative of the Flemish government to position Flanders as an international leader in the field of AI. This AI-impulse program will focus on three main pillars: strategic basic research; technology transfer & industrial applications; and supporting activities (awareness, training, ethics…). To shape the strategic basic research, a team consisting of representation of industry associations, all five Flemish Universities and all four Strategic Research Centres have worked together to define a limited number of AI Grand Challenges that are internationally recognized and for which Flanders has key R&D and domain specific strengths. 

 

These PhD positions will be active in the first grand challenge of this impulse program, called Data Science: Hybrid, Automated, Trusted, and Actionable. This grand challenge will deal with the following tasks, which will constitute the objectives of these PhD positions.

 

!! Important: Send your CV directly to the contact person(s) mentioned at the topic for which you want to apply. 

 

Data quality assessment and enhancement: Develop novel supervised and unsupervised machine learning techniques to track data quality across time (data inconsistencies, noise, artifacts, …), which can be used to decide for each data epoch to either remove it, enhance it, or keep it as is. Furthermore, we will design algorithms to automatically enhance / replace / inpaint data by exploiting the ‘surrounding’ high-quality data to denoise or replace the low-quality data based on subspace methods, graph inference and interpolation, or multi-linear low-rank models. 

contact: Prof. dr. ir. Alexander Bertrand, mail: alexander.bertrand@esat.kuleuven.be

 

Semantic segmentation: AI systems often have to cope with an overwhelming amount of input data, while only a fraction of it is relevant for further processing. Therefore automated data decimation processes are required, which discard irrelevant samples (data points), remove redundancy by fusing multiple input sources/features, and perform a segmentation of time-series into higher-level objects. In this task, we will automate these processes as much as possible, with a focus on automatically splitting (multi-modal) time series or images into segments of variable sizes, within which the statistics are homogeneous across each segment. 

contact: Prof. dr. ir. Alexander Bertrand, mail: alexander.bertrand@esat.kuleuven.be

 

Global matrix/tensor models for structured data: Numerical linear and multilinear algebra have long been a key component of advanced techniques in domains ranging from signal processing to machine learning. Today they continue to enable the development of methods for modeling structured data. Such models are typically global, in that they represent models for the full data set or the data distribution. The work in this task will be divided into tensor-based and linear-algebraic based techniques (including kernel-based methods). (this topic has 3 positions)

contact: 

Prof. dr. ir. Lieven De Lathauwer, tel.: +32-56-24 60 62, mail: lieven.delathauwer@esat.kuleuven.be 

Prof. dr. ir. Johan Suykens, tel.: +32 16 32 18 02, mail: johan.suykens@esat.kuleuven.be

 

Model fusion: At a system level, multiple predictive models are often integrated to support a decision.  These models may originate from multiple inputs, outputs, scales or contexts. This task focuses on the integration of these various heterogeneous types of knowledge into a general framework optimally fusing expert-driven models as well as data-driven machine (deep) learning models. The aim is to share as much as possible common properties among inputs/outputs in order to build up knowledge on common layers.

contact: 

Prof. dr. ir. Sabine Van Huffel, tel.: +32 16 32 17 03, mail: sabine.vanhuffel@kuleuven.be 

Prof. dr. ir. Johan Suykens, tel.: +32 16 32 18 02, mail: johan.suykens@esat.kuleuven.be

 

Interpretability towards the end user: Introducing more AI into decision support systems often leads to black boxes, yielding no insight into the reasons of their decisions. In medical and industrial settings, where trust and accountability are important issues, such system should preferably be interpretable. Therefore, the focus of this task will be on the interpretability and understandability of decisions towards the end user, on turning data and knowledge into models that are easy to understand and readily provide insights to the end-user, possibly also providing ‘intelligent’ modules that elaborate on and clarify the reasons why certain decisions are made. To this goal, user friendly and intuitive graphical user interfaces are mandatory. Another type of interpretability to consider is providing actionable advice suggesting the end-user which actions are expected to improve outcome. A second type is to involve the end user or domain expert to provide event times for unlabeled or censored observations that are queried by the system (active learning). In the multi-target setting, queries can take the form of (instance, event) pairs instead of just instances, allowing a finer grained query specification.

contact: 

Prof. dr. ir. Sabine Van Huffel, tel.: +32 16 32 17 03, mail: sabine.vanhuffel@kuleuven.be 

Prof. dr. ir. Johan Suykens, tel.: +32 16 32 18 02, mail: johan.suykens@esat.kuleuven.be

 

'Length of Stay' Hospital Decision Support System: In this PhD we will develop an automated AI system that predicts the expected Length of Stay (LoS) in a hospital bed for patients, starting from the central database of our academic hospital UZ Leuven (1000s of beds, more than 500 000 consultations per year). Can a patient be treated at home, or should (s)he be hospitalized ? If so, when a patient is kept too long, costs for the hospital and waiting queues for other patients increase, but when a patient leaves prematurely, it might happen that (s)he has to come back, possibly with health complications, and hence financial impact. So there is an 'optimal' LoS, which we assume to be predictable. This is a challenging problem, where one has to take into account the pathology for which the patient is treated, the required care trajectory and the individual patient profile and the overall demands on the system (capacity, queues, etc.). The end result is a fully automated, user-friendly software module in which we run efficient forecasting algorithms, several hundreds of times per day. 

contact: Prof. dr. ir. Bart De Moor, tel.: +32 16 32 17 09, mail: bart.demoor@kuleuven.be 

 

Smart Grid Decision Support System: Because of increasing awareness on and demands concerning climate change, efficient monitoring and control of a nation- and even continent-wide power grid system is a must. In this PhD, we start from the (massive big) data set of one of the Belgian power grid operators, to be fused with other data sets (like the daily weather past records and future forecast), first to model the power grid, and next to determine optimal, cost-efficient and reliable modi of operation, in which we model and predict demand and supply, safeguard the reliability and allow for optimal scheduling of maintenance. This PhD will be about big data, designing smart and fast algorithms (some of which have to calculate in real time) and ultimately implementing all results in a user-friendly operator module. 

contact: Prof. dr. ir. Bart De Moor, tel.: +32 16 32 17 09, mail: bart.demoor@kuleuven.be 

 

Privacy-preserving AI: There is a strong need for AI systems that enable the building of machine learning models across multiple partners holding sensitive personal data (e.g., medical or genomic data) or competitively valuable data (e.g., pharmaceutical data). Such systems must enable individuals whose personal data is processed or cooperating parties that want to retain control over their data to cooperate on their own terms. Privacy-preserving (PP) AI integrates machine learning and cryptographic techniques to ensure that even the coordinating node of a PP AI system does not obtain information about the underlying data from contributing parties beyond pre-agreed results. Such federated AI systems need to be developed to tackle all key learning tasks, such as association and hypothesis testing, meta-analysis, clustering, classification, and regression. Moreover, novel schemes for all key machine learning tasks need to be developed to leverage state-of-the-art Multi-Party Computation (MPC) and (Somewhat) (Fully) Homomorphic Encryption (SFHE) to provide scalable privacy-preserving machine learning methods. Core applications involve the PP analysis of genomic data, of real-world clinical data, and of pharmaceutical drug discovery data.

contact: Prof. dr. ir. Yves Moreau, tel.: +32 16 32 86 45, mail: Yves.Moreau@esat.kuleuven.be 

 

Deep matrix factorization: We have previously developed a highly scalable method for incomplete matrix factorization with side information (Macau). This method provided a key idea of learning of latent representations comparable to, but different from, deep learning autoencoders. We aim at combining ideas of matrix factorizations and deep learning to develop new flexible and scalable deep learning architectures and apply them in different areas of genomic data, real-world clinical data, and drug discovery.

contact: Prof. dr. ir. Yves Moreau, tel.: +32 16 32 86 45, mail: Yves.Moreau@esat.kuleuven.be 

 

Profile

You should have a master's degree in information technology, biomedical engineering, electrical or mathematical engineering, artificial intelligence, or a similar degree with an equivalent academic level. A genuine interest in signal processing and machine learning should motivate your application. The candidate should have strong social abilities allowing an active participation to the multidisciplinary network, fruitful exchanges with other students and researchers, and an excellent integration in the team of your research group. 
 

Offer

Duration: 36 Months (extendable with maximally 12 months to finish the PhD)

 

The selected candidate will be able to take advantage of the unique set-up of the AI impulse program, encompassing all five Flemish Universities. Furthermore, concrete proof-of-concepts will be developed on a yearly basis from which knowledge and technology can be transferred to the business community. It is the aim that the demonstrations of the proof-of-concepts serve as the entry point for the Technology Transfer and Industrial Applications part of the larger Flemish impulse program to tailor the results of the research work towards the specific needs of companies. With this in mind, the PhD candidates involved in this AI impulse program will become independent researchers with improved career prospects in both the academic and non-academic sectors, and will contribute to Flanders’s competitiveness and attractiveness in the field of AI.

 

The work will be performed within the research division STADIUS ('Stadius Centre for Dynamical Systems, Signal Processing, and Data Analytics') at the Electrical Engineering Department (ESAT) at KU Leuven, Europe’s most innovative university (Reuters, 2018). STADIUS's major research objective is to contribute to the development of improved digital (control and signal processing) systems that incorporate advanced mathematical modeling techniques as a crucial new ingredient.

 

You can apply for this job no later than September 01, 2019

Interested?

Please contact and send your CV immediately to the contact person responsible for the topic you are interested in.

You can apply for this job no later than September 01, 2019 via the online application tool

KU Leuven seeks to foster an environment where all talents can flourish, regardless of gender, age, cultural background, nationality or impairments. If you have any questions relating to accessibility or support, please contact us at diversiteit.HR@kuleuven.be.


If you apply for this position please say you saw it on Computeroxy

Apply

All Jobs

FACEBOOK
TWITTER
LINKEDIN
GOOGLE
https://computeroxy.com/11-phd-positions-in-the-field-of-artificial-intelligence-and-machine-learning,i7923.html">

Harvard University Academic Positions

Kuwait University Current Faculty Openings

Osaka University Academic Opportunities

Purdue University Job Postings for Faculty Positions

Texas Tech University Faculty Openings

Tsinghua University Job Postings

University of Cambridge Job Openings

University of Geneva Faculty Opportunities

University of New South Wales Job Openings

University of Nottingham Research Positions

University of Oslo Academic Jobs

University of Saskatchewan Faculty Positions

University of Southampton Research Vacancies

University of Tokyo Current Academic Vacancies

University of Toronto Open Faculty Positions