Data Science Research Center HIDA and DSRC@BGU Summer Exchange

The purpose of this DSRC@BGU and HIDA Summer Exchange is to bring together data science talent from both institutions to work together on predefined research projects. These projects should advance the participant's skillset in the application and, or research or, information and data science methods. Principal investigators (PI) shall welcome the other institution's participants in their labs and to their teams, and ensure that they have the necessary support and resources to succeed in their projects. The participating researchers shall share credit for the results of the project. This hands-on experience at each other's institutions shall serve as the foundation for ongoing scientific collaboration between Helmholtz centers and BGU.

The exchange program is open to Bachelor and Master students, as well as to doctoral researchers and postdocs at BGU. At Helmholtz, the program is open to doctoral researchers and postdocs.

HiDA Logo

About HIDA

HIDA offers training in Information and Data Science to doctoral researchers and postdocs. As a central forum and innovation hub, HIDA facilitates the interdisciplinary exchange of Data Science expertise between all of the Helmholtz Centers and research fields. HIDA connects six regional Helmholtz Information and Data Science Schools and ensures exchange between them. What's more, HIDA cooperates with top-level universities and research institutions nationally and internationally to further knowledge exchange in Data Science methods and skills within the Helmholtz Association and beyond.

Visit HIDA website »

What can participating students expect?

* Interesting and challenging data science problems that expand participants' skillsets and knowledge of methodologies

* Experienced scientists that mentor and support their professional development in data science methods and their application in the six Helmholtz domains

DSRC@BGU will cover living expenses for their participants' six-week stay at the HIDA lab. In return, HIDA finances Helmholtz participants' stays at BGU).

Registration Form for Students »

Questions? Feedback? Get in touch with Racheli Eldin (eldin@bgu.ac.il)

Projects

Show All

Hide All

Project 1 - Effect of air pollution and heat release in cities on urban clouds and the urban climate

Host name

Julia Fuchs

Link to host lab

http://www.imk-asf.kit.edu/english/1372.php

http://www.imk-asf.kit.edu/english/MSA.php

http://www.imk-asf.kit.edu/english/198.php

Group description:

Our research group is analyzing satellite observations of various processes in the atmosphere and on the Earth's surface at European and African sites. We are not only using satellite data for diverse applications in atmospheric and environmental sciences, but also validate data and develop algorithms. Within our research fields we focus on the interactions of low clouds and the land surface as well as urban air pollution using various statistical methods including machine learning techniques.

What is the data science project's research question?

How is air pollution and heat release in cities driving urban cloud modifications and the urban climate?

What data will be worked on?

Data containing information on cloud properties and meteorological condition from geostationary and polar-orbiting satellites and ground-based measurements of air pollution.

What tasks will this project involve?

After a short introduction on the background of the research project and satellite data in general, appropriate data science methods are discussed together. A readily prepared data set will be further preprocessed for the use in the data analysis project. In this project, amongst others machine learning algorithms will be applied with the goal to predict cloud properties over urban areas using air pollution and meteorological parameters. At this point, the expertise of the participant will be of great value for the research project. The discussion and interpretation of the modeled cloud and air quality parameters follows.

What makes this project interesting to work on?

The participant will be able to follow the working methods of an environmental scientist, actively working in the field of satellite remote sensing for many years, getting hands-on experience in a highly relevant and cutting-edge research field. This project is part of fundamental climate research that has important implications for the quality of life in cities around the globe today and in the future. Within the research visit the participant is given the opportunity to expand knowledge in environmental science, freely shape the project through her or his data science expertise and own initiative.

What skills are necessary for this project?

Data analytics / statistics, Data mining / Machine learning, Visualization, Geographic Information Systems

What infrastructure, programs will be used for this project?

Linux-based servers, python; further open-source software can be installed

What will be the project's outcome?

Contribution to research paper, Develop, improve and publish algorithms in environmental science

Which part of the project could be worked on remotely?

The theoretical background, the introduction to the topic as well as suggestions for further reading will be given remotely.

The setup of a working python environment including open-source programming tools and software on the students local working laptop will be prepared and setup in close exchange together with the student.

The code scripting, the preprocessing of a light-weighted test data set and parts of the training, validation and statistical analysis of the machine learning model will be done remotely.

How would you work together remotely?

We would meet on a regular basis using Zoom and MS-Teams to ensure a close collaboration. Research papers, theoretical introductions, code algorithms and test data sets will be transferred online using E-mail, Github and WeTransfer. Work on a scientifc publication will be continued after the stay using Overleaf.

Which part of the project would be worked on at the research center?

During the stay at the KIT Research Center the tested and prepared code will be transferred to a large remote-sensing data set. Results of the machine learning model will be analyzed and interpreted followed by an informal presentation and discussion during a meeting of the Satellite Climatology group at the KIT. Together with the student the main outcomes will be prepared for a scientific publication that can be continued remotely after the stay.

Additional information

Is the data open source?	Yes
What level of experience is necessary for this project?	Master level
Would the participant be working as part of a team or individually?	Team

Urban clouds - Understanding clouds over urban areas | Julia Fuchs - Satellite Climatology

Project 2 - A prediction model for disease status or molecular phenotype incorporating genetic and environmental effects in Major Depressive Disorder

Host name

Na Cai

Link to host lab

https://www.pioneercampus.org/about-us/principal-investigators/na-cai/index.html

Group description:

My research group focuses on understanding the etiology of neuropsychiatric conditions, particularly Major Depressive Disorder (MDD). MDD is a common and complex disorder with both genetic and environmental factors, many of which are shared with other conditions. The current challenge is to discover those factors specific and causal to MDD (or its subtypes) and how they interact with each other, using high-dimensional genetic, molecular, phenotypic and environmental data.

What is the data science project's research question?

The development of a prediction model for disease status or molecular phenotype incorporating genetic and environmental effects.

What data will be worked on?

Large scale population cohorts such as the UKBiobank and in-house disease cohorts will be available for investigations into complex diseases such as MDD, and gene expression datasets such as the GTEx Consortium will be used for investigations into molecular phenotypes.

What tasks will this project involve?

The data scientist will first familiarise themselves with genetic data and methods in quantitative genetics, including association studies, gene-environmental interaction analyses, estimation of genetic architecture of a complex trait, and in particular prediction methods using polygenic risk scores (PRS). They will then build on current methods for trait prediction and formulate a prediction model incorporating environmental factors. Their work will involve computationally implementing the model, testing it on both simulated and real datasets, and reporting their results in both writing and oral presentations. Strong statistical and computational abilities, in addition to interest for applications of statistical modelling to biomedical questions, are desired. This work is complementary to on-going work in the lab, and will enable new directions of investigation.

What makes this project interesting to work on?

Current prediction models do not incorporate the environmental factors or their effects on how genetic factors influence a trait, therefore leading to the low transferability of genetic predictions between different environmental contexts. Reasons for this include the lack of large-scale datasets profiling environmental factors such as lifestyle measures, and the lack of a statistical model incorporating their effects in predictions. The former is increasingly mitigated with large population based cohorts, while the latter remains a challenge. This project works on this challenge, and is a timely and useful addition to the current technologies and will bring advance to precision and personalised medicine. Further, the data scientist involved in this work will be introduced to the latest technologies and unsolved problems in quantitative genetics, a field rapidly transformed by the availability of big data and which presents many opportunities for data scientists.

What skills are necessary for this project?

Data analytics / statistics, Scientific computation, Computational models, Computer simulations, Visualization, High-performance computing

What infrastructure, programs will be used for this project?

High performance computing in Unix environment; scientific programming in statistical or compiled language for data analysis and visualisation (knowledge of Python necessary); statistical methods in quantitative genetics (in particular the linear mixed-model library LIMIX)

What will be the project's outcome?

Contribution to research paper

Additional information
Is the data open source?	Publicly available data that is under controlled access
What level of experience is necessary for this project?	PhD
Would the participant be working as part of a team or individually?	Individually

Project 3 - Using NLP for a database of compounds relevant in materials science

Host name

Nicole Jung and Stefan Bräse

Link to host lab

https://www.ioc.kit.edu/braese/

Group description:

The Bräse group combines traditional chemical synthesis with the development of software to be used to solve challenges in chemistry research. In the IT development projects, we develop Open Source systems to build infrastructure for chemists in form of an electronic lab notebook, a repository for research data and several other systems that are necessary to support the scientists. Data that is gained via these projects is used for machine learning and other analytical projects.

What is the data science project's research question?

Can we establish a database for compounds relevant in materials science with NLP? Can this information be processed to a knowledge graph?

What data will be worked on?

Open Access databases and publications and Abstracts of Papers

What tasks will this project involve?

Open Access databases have to be identified to extract information that is suitable for the characterization and identification of chemical compounds. Documents and information have to be extracted to gain the relevant data on the structures, application and potential in catalysis and materials science. Several analytical tools have to be used for different sources of information, e.g. natural language processing but also some smaller algorithms have to be established. The KIT group will provide own developments for chemical compound extraction to facilitate the work.

What makes this project interesting to work on?

The work is interdisciplinary and challenging but the hosting group at the KIT is experienced in the processing of chemistry data. The host group is able to support the work with respect to the chemistry-related challenges and the development of models for the processing of data and the establishment of the necessary infrastructure.
The development of new models for the identification of new compounds relevant in catalysis and materials science is of high importance in many fields of science. The development and identification of suitable molecules to solve the challenges in the fields of energy generation and storage with high impact on environmental development is only one of them.

What skills are necessary for this project?

Data analytics / statistics, Scientific computation, Computational models, Data mining / Machine learning, Databases, high motivation

What infrastructure, programs will be used for this project?

We provide the access to all developments of the KIT development group in chemistry, including software to extract and process information (in particular software to store, extract and curate chemical information and structures). Common infrastructure including chemistry development kits is available. In addition, the guests may use the databases of the group to gain additional information.

What will be the project's outcome?

Contribution to research paper, Contribution to software development

Which part of the project could be worked on remotely?

Search for and in databases and suitable other sources for information on materials and their properties. Collection of most important applications of molecules and dependencies on the identified materials' properties. Studying NLP tools, extraction of data, and exemplary data analysis based on selected scientific documents. Setup database with the collected data.

How would you work together remotely?

The description of the project and teaching will happen via regular meetings of the project group via online meetings (1-2 meetings per week) and individual teaching and discussion via online meetings (as often as required).

Which part of the project would be worked on at the research center?

The detailed analysis of the stored and extracted data will be done at KIT. Work that includes chemical know how to correlate structures and information will also be done at KIT.

Additional information

Is the data open source?

Yes

What level of experience is necessary for this project?

Master level could be suitable as well, depends on the skills and motivation

Would the participant be working as part of a team or individually?

Team

Project 4 - Quantifying uncertainty in climate and weather models with machine learning tools

Host name

David Greenberg

Link to host lab

HZG Lab (own group as of February, 2020)

Group description:

We aim to understand and predict the behavior of earth's atmosphere, land and ocean. To do this, we develop software tools that combine the flexibility and power of machine learning with physical modeling based on sound scientific principles. In practice, this means scaling up cutting-edge techniques from statistics and deep learning to models and datasets that can describe the entire planet.

What is the data science project's research question?

How can we use machine learning tools to quantify uncertainty in climate and weather models?

These models exhibit complex, nonlinear and chaotic dynamics. Given physical assumptions and observational data, we often wish to predict a range of plausible future outcomes, or the range of model parameters consistent with data. This remains a challenging problem because the probability distributions describing these ranges are high dimensional, and often contain multiple non-Gaussian peaks.

A way forward is offered by modern neural density estimators, such as normalizing flows, variational autoencoders and generative adversarial networks. These tools have proven flexible and effective in capturing complex relations between many variables, but applying them to the models and datasets describing earth is challenging. One reason for this is that the number of observed and hidden variables involved can be much higher than other applications. These variables may also have uneven spacing in space and time, posing a challenge for standard deep learning architectures. Furthermore, climate observations are usually burdened with uncertainties, as they are indirect observations (e.g. remote-sensed reflectances).

What data will be worked on?

This project will be primarily based on datasets from satellite remote sensing, weather station data, climate model outputs and reanalysis data. If the participant is particularly interested in some other earth science dataset it might be possible to use this instead.

What tasks will this project involve?

Designing and implementing neural density estimators in pytorch, and applying these to large datasets. Quantitatively benchmarking these results against previous methods.

What makes this project interesting to work on?

Quantifying uncertainty in earth system models,and assessing the predictability of the main patterns of climate variability, is both one of the most critical challenges facing humanity, and one of the most fascinating and challenging problems computational problems a data scientist can work on. The central challenge is to combine scientific insights into the inner workings of the natural world with the effectiveness and scalability of machine learning.

What skills are necessary for this project?

Computational models, Data mining / Machine learning

What infrastructure, programs will be used for this project?

python, pytorch, git, GPU-computing, cluster computing

What will be the project's outcome?

Contribution to research paper, Contribution to software development

Which part of the project could be worked on remotely?

All tasks and aspects of the project can be worked on remotely. Potential datasets are readily available for download, and can also be readily accessed by connections to HZG computing facilities. Collaborative coding will be carried out using git, github and jupyter notebooks.

How would you work together remotely?

Video and audio calls will be used to discuss mathematical, conceptual and programming ideas, along with online virtual whiteboards. These which will be laid out in collaborative latex-based documents on overleaf. Code will be reviewed, run and tested collaboratively.

Which part of the project would be worked on at the research center?

More detailed discussion, data analysis and presentation of the results, along with final improvements to the code and ideas developed so far. Essentially, in person we would carry out the same tasks, but better.

Additional information
Is the data open source?	Yes
What level of experience is necessary for this project?	Master level
Would the participant be working as part of a team or individually?	Team

Project 5 - Analysis of X-ray scattering data with neural networks

Host name

Nico Hoffmann

Link to host lab

https://www.hzdr.de/db/Cms?pNid=1483

Group description:

The long term and sustainable success of X-ray community essentially depends on its ability to meet growing challenges in handling and analyzing data of increasing volume and complexity. ML techniques provide a smart solution enabling dramatically increasing the output of x-ray scattering facilities regarding acceleration of data analysis. That analysis allows us to learn about nanometer scale structures of future materials, biological macromolecules or very fast processes of Matter under extreme conditions or warm dense matter (Plasma Physics, Astrophysics).

What is the data science project's research question?

Analysis of X-ray scattering data is a very time-consuming process as it requires solving an ill-posed inverse problem to infer properties describing the imaged object. We approach this task by neural networks that basically estimate the most important parameters of the objects in a single step given the experimentally acquired scattering image. Some parameters inhibit low contribution to the acquired data which also hampers reliable predictions and discrimination of these parameters. We will tackle that issue by methods for estimation of the posterior distribution of these parameters to resolve ambiguous situations and provide information about the reliability of the estimate.

What data will be worked on?

Synthetic + experimental Small Angle X-Ray scattering (SAXS) data

What tasks will this project involve?

Development of training data generation methods and recent neural network architectures for one step reconstruction of SAXS data (Variational Autoencoder, Invertible Neural Networks). That tasks base on preliminary works of my ML students so you don't have to start from scratch.

What makes this project interesting to work on?

The reconstruction of X-ray scattering data allows us to understand matter under extreme conditions! We'll be able to understand the interaction of an ultra high intensity laser with solid targets which helps us to actually understand processes inside super hot plasmas. At the same time, you are going to learn and actually do big data image analysis (>100 GB) on a pretty decent GPU cluster (>180 NVIDIA V100 gpus). You'll be working in a team with two-three more ML students of my group that are experienced in Deep Learning based image analysis. Additionally, you are getting supervised by me (ML group leader at HZDR) and an X-ray scattering expert (Dr. Thomas Kluge). We really want you to become a part of our nice team of machine learning experts and physicists and learn quite a lot about analysis of big data.

What skills are necessary for this project?

Data analytics / statistics, Data mining / Machine learning, Deep learning, High-performance computing

What infrastructure, programs will be used for this project?

Python; PyTorch; Horovod; large IBM Power9 GPU cluster

What will be the project's outcome?

Contribution to research paper

Which part of the project could be worked on remotely?

Most of the work is done on a HPC machine anyway for what reason no presence is required though I guess it would be great to jointly+locally work on the problem.

How would you work together remotely?

There will be a weekly video conference for discussions related to the current progress and potential issues. Furthermore, I'm going to set up a slack channel for all people contributing to the project (at BGU + of my lab) for discussions on minor issues and maybe distribution of work. One master's student of mine also agreed on supporting our collaborators form BGU and a PhD student joining my lab in June is going to support the work, too. Our bi-weekly machine learning paper reading group is also open for our friends at BGU.

Which part of the project would be worked on at the research center?

See above.

Additional information
Is the data open source?	Yes
What level of experience is necessary for this project?	Master level
Would the participant be working as part of a team or individually?	Team