Three PhD positions in the Data Mining group, TU/e
Three PhD positions in the Data Mining group, TU/e
You cannot apply for this job anymore (deadline was 27 Nov 2019).
Browse the current job offers or choose an item in the top navigation above.
Job description
The Data Mining group at Eindhoven University of Technology (TU/e) is searching for candidates for three fully-funded 4-years PhD positions. All three positions involve fundamental scientific research: the primary goal is to enrich the portfolio of available data mining techniques with new innovative methods. Each position involves a specific project partner in industry, ensuring that your research will have direct impact in the real world. Hence, these positions offer the best of both worlds: theoretical research with direct practical impact.
The scientific framework in which the positions are grounded is Exceptional Model Mining, a data mining framework that strives to find subgroups within a dataset that are interesting. Subgroups are only deemed interesting if they satisfy two conditions. On the one hand, they must be interpretable: we must be able to succinctly describe the definition of a subgroup, so that the knowledge that they represent becomes actionable. On the other hand, they must be exceptional: they must display some kind of behavior that sets them apart from overall behavior. The scientific challenges revolve around how to efficiently search for subgroups, and how to express exceptional behavior such that the subgroups we find are meaningful.
TU/e is a dynamic, research-intensive university in the heart of Europe. TU/e is consistently ranked within the top-100 positions in several world rankings for its research and quality of education. The Data Mining group consists of 5 professors and 17 PhD students and postdocs. The research results of the group appear in top-tier conferences and journals including but not limited to IJCAI, AAAI, ICDM, SDM, KDD, ECMLPKDD, DAMI, MACH, and Nature Communications. The group has close collaboration with several leading companies, organizations and hospitals. The advertised positions are in collaboration with the Dutch National Police (Position A), Philips Research and the ZGT hospital (Position B), and Rabobank (Position C).
All 3 successful candidates will be based in the Data Mining group, where are opportunities to collaborate with other PhDs and to co-supervise MSc students relevant to PhD their research. The successful candidates are expected to be willing to spend some of their time at the Dutch National Police, Philips Research, and Rabobank, correspondingly. The research will be concluded with a PhD thesis. You will be supervised by dr. Wouter Duivesteijn and prof. dr. Mykola Pechenizkiy.
Position A. Exceptional behavior profiling / EMM in operational field labs at Dutch National Police.
The outlet center in Roermond is a hot target for pickpockets. Using a plethora of sensor data, the Dutch National Police is trying to detect the pickpockets before they strike. Since criminals can reasonably be assumed to behave differently from the overall population, it stands to reason that Exceptional Model Mining will be beneficial here.
A specific research challenge for this position lies in the following. The police possesses data characterizing how cars navigate the road network. If this data consists of 50% commuter traffic and 50% tourist traffic, the overall behavior is an average of these two populations. Hence, no single car will display what EMM typically designates as "the overall behavior". But then, if we strive to find subgroups behaving exceptionally, this behavior must be exceptional compared to what, exactly? Both theoretical solutions to this problem, as well as solutions that involve police experts in the loop, are to be explored.
Position B. EDIC - Exceptional and Deep Intelligent Coach.
Our increased life expectancy goes hand in hand with an increased number of years living with chronic conditions. The EDIC projects strives to develop an intelligent digital coach for deploying a healthy lifestyle for people suffering from such conditions. A key component of this coach is informed and interpretable complex event detection and model adaptation on heterogeneous and noisy data streams capturing patients' medial condition, lifestyle, and behavior. Your task is to perform the research necessary for fulfilling this component, within the wider EDIC consortium; this involves researchers from both TU/e and the University of Twente, the ZGT hospital driving the use cases (diabetes and obesity), and the companies Philips, Holst, and RRD who deliver the technology and know-how to build the physical coaching system.
A specific research challenge for this position concerns extending Exceptional Model Mining beyond traditional flat-table data. Traditionally, one assumes that all data is given, stems from one homogeneous probability distribution, and all observations are independent. The heterogeneous and noisy data streams generated in this project satisfy none of those traditional assumptions. A large conceptual step must be made within EMM to enable handling such data: how can one characterize exceptional behavior, if one can never oversee all behavior, and if one must compare across dependent observations?
Position C. Explainable and accountable KYC Predictive Analytics at Rabobank.
Analyzing customer transaction data is a longstanding form of data mining at banks. However, a bank interacts with customers along multiple channels: semi-structured text documents such as contracts, completely free-form text documents such as helpdesk chat logs, and the very clearly structured transaction data can combine to deliver a more complete image of the wants and needs of a customer. Rabobank strives to know their customer better, and Exceptional Model Mining plays a key role therein.
A specific research challenge for this position lies in extracting more knowledge on how a complex black-box classifier works. European legislation compels any company using automated systems for making decisions substantially impacting a customer, to deliver insights to that customer on how the decision was made. With the rise of complex classification models in the era of deep learning, this puts the state of the art in classification at odds with the demands on a company under the law. With Exceptional Model Mining, combining inherent interpretability with complex exceptional behavior, we can deliver deeper insights in how the classifier arrived at the decision. This combines fulfilling a company requirement under European law, with fascinating deeper insight in classifier behavior for scientists.
The scientific framework in which the positions are grounded is Exceptional Model Mining, a data mining framework that strives to find subgroups within a dataset that are interesting. Subgroups are only deemed interesting if they satisfy two conditions. On the one hand, they must be interpretable: we must be able to succinctly describe the definition of a subgroup, so that the knowledge that they represent becomes actionable. On the other hand, they must be exceptional: they must display some kind of behavior that sets them apart from overall behavior. The scientific challenges revolve around how to efficiently search for subgroups, and how to express exceptional behavior such that the subgroups we find are meaningful.
TU/e is a dynamic, research-intensive university in the heart of Europe. TU/e is consistently ranked within the top-100 positions in several world rankings for its research and quality of education. The Data Mining group consists of 5 professors and 17 PhD students and postdocs. The research results of the group appear in top-tier conferences and journals including but not limited to IJCAI, AAAI, ICDM, SDM, KDD, ECMLPKDD, DAMI, MACH, and Nature Communications. The group has close collaboration with several leading companies, organizations and hospitals. The advertised positions are in collaboration with the Dutch National Police (Position A), Philips Research and the ZGT hospital (Position B), and Rabobank (Position C).
All 3 successful candidates will be based in the Data Mining group, where are opportunities to collaborate with other PhDs and to co-supervise MSc students relevant to PhD their research. The successful candidates are expected to be willing to spend some of their time at the Dutch National Police, Philips Research, and Rabobank, correspondingly. The research will be concluded with a PhD thesis. You will be supervised by dr. Wouter Duivesteijn and prof. dr. Mykola Pechenizkiy.
Position A. Exceptional behavior profiling / EMM in operational field labs at Dutch National Police.
The outlet center in Roermond is a hot target for pickpockets. Using a plethora of sensor data, the Dutch National Police is trying to detect the pickpockets before they strike. Since criminals can reasonably be assumed to behave differently from the overall population, it stands to reason that Exceptional Model Mining will be beneficial here.
A specific research challenge for this position lies in the following. The police possesses data characterizing how cars navigate the road network. If this data consists of 50% commuter traffic and 50% tourist traffic, the overall behavior is an average of these two populations. Hence, no single car will display what EMM typically designates as "the overall behavior". But then, if we strive to find subgroups behaving exceptionally, this behavior must be exceptional compared to what, exactly? Both theoretical solutions to this problem, as well as solutions that involve police experts in the loop, are to be explored.
Position B. EDIC - Exceptional and Deep Intelligent Coach.
Our increased life expectancy goes hand in hand with an increased number of years living with chronic conditions. The EDIC projects strives to develop an intelligent digital coach for deploying a healthy lifestyle for people suffering from such conditions. A key component of this coach is informed and interpretable complex event detection and model adaptation on heterogeneous and noisy data streams capturing patients' medial condition, lifestyle, and behavior. Your task is to perform the research necessary for fulfilling this component, within the wider EDIC consortium; this involves researchers from both TU/e and the University of Twente, the ZGT hospital driving the use cases (diabetes and obesity), and the companies Philips, Holst, and RRD who deliver the technology and know-how to build the physical coaching system.
A specific research challenge for this position concerns extending Exceptional Model Mining beyond traditional flat-table data. Traditionally, one assumes that all data is given, stems from one homogeneous probability distribution, and all observations are independent. The heterogeneous and noisy data streams generated in this project satisfy none of those traditional assumptions. A large conceptual step must be made within EMM to enable handling such data: how can one characterize exceptional behavior, if one can never oversee all behavior, and if one must compare across dependent observations?
Position C. Explainable and accountable KYC Predictive Analytics at Rabobank.
Analyzing customer transaction data is a longstanding form of data mining at banks. However, a bank interacts with customers along multiple channels: semi-structured text documents such as contracts, completely free-form text documents such as helpdesk chat logs, and the very clearly structured transaction data can combine to deliver a more complete image of the wants and needs of a customer. Rabobank strives to know their customer better, and Exceptional Model Mining plays a key role therein.
A specific research challenge for this position lies in extracting more knowledge on how a complex black-box classifier works. European legislation compels any company using automated systems for making decisions substantially impacting a customer, to deliver insights to that customer on how the decision was made. With the rise of complex classification models in the era of deep learning, this puts the state of the art in classification at odds with the demands on a company under the law. With Exceptional Model Mining, combining inherent interpretability with complex exceptional behavior, we can deliver deeper insights in how the classifier arrived at the decision. This combines fulfilling a company requirement under European law, with fascinating deeper insight in classifier behavior for scientists.
Specifications
- max. 38 hours per week
- Eindhoven View on Google Maps
:fill(white)/logos/tue-en-wide_acOLN2b.png)
Eindhoven University of Technology (TU/e)
Requirements
We are looking for candidates that meet the following requirements:
- a solid background in Computer Science with specialization in data mining, machine learning, deep leaning;
- a strong interest in data science research with focus on local pattern mining, exceptional model mining or related techniques;
- data mining software development skills at least in one language, e.g. R, Python, Java;
- good communication skills in English, both in speaking and in writing;
- capability and willingness to work both independently and in a team of data scientists and interact with domain experts; being highly motivated, rigorous, and disciplined;
- being enthusiastic about working on the challenging use cases;
- experience in research and a publication record will be considered additional advantages.
Conditions of employment
- A full time temporary appointment for a period of 4 years, with an intermediate evaluation after 9 months;
- A gross salary of €2.325 per month in the first year increasing up to €2.972 in the fourth year;
- Tight collaboration of academia with industry with access to real data and domain expertise.
- Strong collaboration ties with several research groups in Europe and worldwide.
- Healthy travel funding for presenting your work at the leading conferences and going on visiting research.
- Support for your personal development and career planning including courses, summer schools, conference visits etc.;
- A broad package of fringe benefits, e.g. excellent technical infrastructure, child daycare and sports facilities, extra holiday allowance (8%, May), and end-of-year bonus (8.3%, December).
