[Job] Ph.D. and postdoc positions - ML foundations for structured problems at the University of Luxembourg

[louk...@gmail.com]

We are looking for a Ph.D. student and a postdoctoral researcher to join the new machine learning group at the University of Luxembourg. 

The positions will involve doing basic research on one or more of the following topics:

• Neural algorithmic reasoning. This position is at the intersection of machine learning, combinatorial optimization, and theoretical CS. Relevant questions: How can we build neural networks that learn to solve discrete problems especially in the absence of supervision? Can we teach them to learn algorithms and to generalize in a strong sense, i.e., beyond the training distribution? Are there basic limits in what they can achieve? Relevant literature: Erdos goes neural, Machine Learning for Combinatorial Optimization
    
  
• Foundations of deep learning. The aim is to study the basic properties of neural networks focusing specifically on structured problems (e.g., ML on graphs, sets, images), paying particular attention to the inter-relation between expressivity, optimization, generalization & inductive bias. Relevant works: What training reveals about neural network complexity, What graph neural networks cannot learn, How neural networks extrapolate, geometric deep learning 
  

Ideal candidate profile. The ideal applicants are driven, they possess a good understanding of the state-of-the-art in deep learning, they have practical/demonstrable experience in building neural networks, and they have a good grasp of basic CS theory (e.g., discrete algorithms, optimization, linear algebra, probability). 

Knowledge of information theory, learning theory, or computational complexity will be considered a bonus. Evidence of the ability to publish in rank A* conferences/journals (such as NeurIPS, ICLR, ICML, JMLR, COLT, CVPR, STOC, FOCS, DISC, ...) will also be a strong asset -- especially for those applying to the postdoctoral position.  Applications from under-represented groups are encouraged to apply. 

What is offered.  The support and freedom to pursue basic research. A generous salary. Minimal obligations beyond research. A passionate and ambitious environment in an upcoming university. Opportunities for collaboration with groups at EPFL, MIT, and Imperial.  

How to apply. If you are interested, send the following to loukasan at gmail dot com:

1. Your CV. 
2. A motivation letter explaining some of the problems that you are excited about and why you think the position is a good fit for you. Take care to emphasize connections with the topics and works described above.  
3. The contact information of 2-3 people you have worked with that are willing to provide a reference letter. 
4. Up to three selected publications (especially for postdoc applicants).
5. A transcript of your grades (for those wanting to pursue a Ph.D.)

The reviewing of applications will start on the 15th of October and will continue until the positions are filled. The starting date will be no earlier than December 2021.