5 PhD scholarships at the Center for Basic Machine Learning Research in Life Science (MLLS), University of Copenhagen and the Technical University of Denmark

The Center for Basic Machine Learning Research in Life Science (MLLS) gathers machine learning researchers from the Departments of Biology and Computer Science, University of Copenhagen (UCPH) and DTU Compute, the Technical University of Denmark (DTU). It is funded by the Novo Nordisk Foundation (NNF) through their Data Science Initiative. The Center aims to conduct world class research in machine learning methodology that simultaneously addresses important problems for applications in life science and has a wider relevance for machine learning.

The PhD scholarships will start on November 1st 2021 or as soon as possible thereafter. Enrollment and employment will be at one of the three involved departments with joint co-supervision across the departments.  

Description of the scientific environment

More information about MLLS can be found here . All PhD students and postdocs of the participating PIs will take part in joint Center group meetings, seminar series and workshops. PhD candidates will have the opportunity to spend time within both machine learning and life science research groups as well as participate in activities within ELLIS and the NNF Data Science Academy.

Project description

The topics of the PhD projects will fall under the following topics:

• Deep Generative Models for Manifold Learning. In this project we will explore deep generative models inspired by manifold learning. One direction is to develop a loss function similar to UMAP/tSNE that aims at preserving neighborhood properties of the representation to obtain more interpretable representations. Another direction is to further develop methods for efficient training of these models and compare them to variational autoencoders and generative adversarial networks.
  Supervisors: Anders Krogh and Søren Hauberg
• Learning and leveraging protein representations Learned representations of protein sequences can be estimated from the large available databases, and promise to play a central role in the analysis of Biological data in the coming years. As the underlying models grow in complexity we often find that the representations become increasingly difficult to reason about. For example, in protein design there is interest in using Bayesian optimization to select protein candidates that are suitable for a specific task. But when the representation is difficult to interpret, it is hard to design suitable priors and acquisition functions for the optimization. Also, contemporary models often assume conditional independence between output dimensions, which force all correlations to be captured by the abstract representations, leading to unnecessary complex representations. This project consists of two parts: 1) Equip learned representations with a geometric structure that allows us to better understand the representation space, 2) Design protein representations which explicitly encode the covariance between the individual residues. 3) Leverage such representations to build better strategies for geometric Bayesian optimization.

Supervisors: Wouter Boomsma & Søren Hauberg

Relevant literature: https://arxiv.org/abs/2012.02679

• New methods for sequence decoding Recently, pre-trained language models like BERT have become the workhorse encoder for sequence learning tasks. This PhD project will work on improving decoding in language tasks by using more advanced likelihood functions that explicitly handle dependencies between positions in the sequence. The topics include variational approximate inference methods for high dimensional conditional random field state space models, probabilistic insertion and deletion models amenable to exact dynamic programming inference. The projects will include pure methods development relevant for both NLP and biological sequence and the development of large scale probabilistic protein language models as an alternative to masked language models such as BERT.
  Supervisors: Ole Winther and Wouter Boomsma.
• Equivariant neural networks for graph-valued predictions. While most current work in graph deep learning focuses on graph classification on benchmark datasets, life science brings far more complex graph learning tasks. One class of problems which require far more faithful representations, is problems that aim to predict a graph. This includes basic problems such as autoencoders or generative models of graphs, but also more complex graph regression problems. This PhD project will utilize permutation equivariant neural networks to create graph-to-graph prediction models with rich representational capacities. The ideal candidate for this PhD project has a strong mathematical background, is experienced in practical implementation of deep learning models, and loves seeing math come to life in better, more efficient algorithms.

Supervisors: Aasa Feragen and Ole Winther

• Quantifying uncertainty in segmentation of anatomical networks from biomedical images. Most graph deep learning development is currently driven by the existence of benchmark databases that allow easy access to pre-extracted graphs with associated classification problems. As a result, the graph generation process and its associated errors and uncertainties is generally ignored, and most existing methods assume that graphs are deterministic and error free. However, when graphs are extracted from noisy data, that noise will inevitably lead to errors in the derived graphs. This project aims to quantify uncertainty in anatomical networks extracted from biomedical images, and to derive graph deep learning models that are capable of correctly treating that uncertainty in downstream analysis. The ideal candidate for this PhD project has experience with Bayesian deep learning and enjoys hands-on work with real data.

Supervisors: Aasa Feragen and Ole Winther

Candidates should indicate what topics have their priority and motivate their choices.

Principal supervisors areProfessor Ole Winther, Biology, UCPH/DTU, [email protected] ,  Aasa Feragen, [email protected] , DTU Compute, Anders Krogh, Computer Science, UCPH, [email protected] , Jes Frellsen, DTU, [email protected] , Søren Hauberg, DTU, [email protected] and Wouter Boomsma, Computer Science, UCPH, [email protected] . The principal supervisors will carry out the assessment of the candidates.

The PhD programme

Qualifications needed for the PhD programme
To be eligible for the regular PhD programme, you must have completed a degree programme, equivalent to a Danish master’s degree (180 ECTS/3 FTE BSc + 120 ECTS/2 FTE MSc) related to the subject area of the project, e.g. biology with specialization in molecular biology, biochemistry, molecular biomedicine. For information of eligibility of completed programmes, see General assessments for specific countries and Assessment database .

Terms of employment in the regular programme
Employment as PhD fellow is full time and for maximum 3 years.
Employment is conditional upon your successful enrolment as a PhD student at the PhD School at the Faculty of SCIENCE, University of Copenhagen. This requires submission and acceptance of an application for the specific project formulated by the applicant.
The terms of employment and salary are in accordance to the agreement between the Ministry of Finance and The Danish Confederation of Professional Associations on Academics in the State (AC). The position is covered by the Protocol on Job Structure.
Salary range starts at DKK 27,788.69 per month incl. PhD supplement (1 April 2021 level).

The employment area will be the University of Copenhagen or at the Technical University of Denmark

Responsibilities and tasks in the PhD programme

• Carry through an independent research project under supervision
• Complete PhD courses corresponding to approx. 30 ECTS/ ½ FTE
• Carry out dissemination and teaching activities
• Participate in active research environments, including a stay at another research institution, preferably abroad
• Teaching and knowledge dissemination activities
• Write scientific papers aimed at high-impact journals
• Write and defend a PhD thesis on the basis of your project
  

We are looking for the following qualifications:

• The grade point average achieved
• Professional qualifications relevant to the PhD project
• Relevant publications
• Relevant work experience
• Other relevant professional activities
• Language skills

The master’s degree should preferably be in computational science and engineering (CSE), computer science, statistics, pure/applied mathematics, physics, bioinformatics or equivalent academic qualifications. Preference will be given to candidates who are excited about basic research, have an interest in the mathematical foundations of machine learning and have experience with mathematical modelling and the programming and tools (e.g., PyTorch/TensorFlow) needed to apply these models to real data. Furthermore, good command of the English language is essential.

Application and Assessment Procedure

 Your application including all attachments must be in English and submitted electronically by clicking APPLY NOW below.

Please include: 

The University wishes our staff to reflect the diversity of society and thus welcomes applications from all qualified candidates regardless of personal background.

The deadline for applications is 31st of August 2021, 23:59 GMT +2.

After the expiry of the deadline for applications, the authorized recruitment manager selects applicants for assessment on the advice of the Interview Committee. Afterwards an assessment committee will be appointed to evaluate the selected applications. The applicants will be notified of the composition of the committee and the final selection of a successful candidate will be made by the Head of Department, based on the recommendations of the assessment committee and the interview committee.

All applications are managed through KU’s job portal. Relevant evaluation committees consisting of members from all partner universities will consider the applications in the evaluation process.

All relevant application material will be shared between DTU Compute, DIKU KU and BIO KU.

The main criterion for selection will be the research potential of the applicant and the above mentioned skills. The successful candidate will then be requested to formally apply for enrolment as a PhD student at the PhD school of Science DTU tilføjer her

You can read more about the recruitment process a https://employment.ku.dk/faculty/recruitment-process/ .

Academic approval and Enrolment The main criterion for selection will be the research potential of the applicant and the above mentioned skills. The successful candidate will then be requested to formally apply for enrolment as a PhD student at either the PhD school of: KU:You can read more about the recruitment process here: https://employment.ku.dk/faculty/recruitmentprocess/. General information about PhD programmes at SCIENCE is available at https://www.science.ku.dk/phd . DTU: For information about our enrolment requirements and the general planning of the PhD study programme, please see the DTU PhD Guide.

Questions

For specific information about the PhD scholarship, please contact

KU:
Principal supervisor Professor Ole Winther, Department of Biology, [email protected]

Associate professor Wouter Boomsma, Department of Computer Science, [email protected]

Professor Anders Krogh, Department of Computer Science, [email protected]

DTU:
Professor Aasa Feragen, DTU Compute, [email protected]
Associate Professor Jes Frellsen, DTU Compute, [email protected]
Professor Søren Hauberg, DTU Compute, [email protected]

General information about PhD programmes at SCIENCE is available at https://www.science.ku.dk/phd .

Technology for people
DTU develops technology for people. With our international elite research and study programmes, we are helping to create a better world and to solve the global challenges formulated in the UN’s 17 Sustainable Development Goals. Hans Christian Ørsted founded DTU in 1829 with a clear vision to develop and create value using science and engineering to benefit society. That vision lives on today. DTU has 12,000 students and 6,000 employees. We work in an international atmosphere and have an inclusive, evolving, and informal working environment. Our main campus is in Kgs. Lyngby north of Copenhagen and we have campuses in Roskilde and Ballerup and in Sisimiut in Greenland.

Part of the International Alliance of Research Universities (IARU), and among Europe’s top-ranking universities, the University of Copenhagen promotes research and teaching of the highest international standard. Rich in tradition and modern in outlook, the University gives students and staff the opportunity to cultivate their talent in an ambitious and informal environment. An effective organisation – with good working conditions and a collaborative work culture – creates the ideal framework for a successful academic career.