CURRICULUM VITAE

Dr. MARIO MICHAEL KRELL

PERSONAL INFORMATION

contact:krell@uni-bremen.de
born:Germany, Frankfurt (Oder), 25th February 1984, German nationality


SUMMARY

  • Postdoc (senior scientist) with 8-year background in machine learning and data analysis
  • Strong problem solving, mathematical analysis, and interdisciplinary teamwork skills
  • Interested in leading data analysis and machine learning algorithm development at university as well as in industry for interesting applications that help humanity


EDUCATION

Time Degree Institute Advisor GPA
03/15 PhD in machine learning University of Bremen, Robotics Group Kirchner, F. 4.0
03/09 degree in mathematics (“Diplom”) Humboldt University of Berlin Kummer, B. 3.9
07/06 pre-degree in computer science Humboldt University of Berlin   3.8
06/03 university-entrance diploma C.-F.-Gauß-Gymnasium Frankfurt (O.)   3.8
Final Thesis:


SKILLS

  • problem solving, machine learning expertise, mathematical analysis, optimization
  • independent research and data analysis, research writing, teaching, scientific presentation
  • interdisciplinary teamwork (system engineers, computer scientists, neuroscientists, management, etc.)
  • knowledge of robotics and related problems, man-machine interfaces, and electroencephalographic data
  • project acquisition and fast writing of parts from project proposals
  • Python programming, reStructuredText, Sphinx, YAML, OmniGraffle, Git (see pySPACE - machine learning environment written in Python to easily configure and run complex evaluations in parallel)
  • LaTeX markup, Mac OS X, MacPorts, basics of Microsoft Office
  • languages: German (mother tongue), English (fluent), French (basic)
  • hobbies: singing (chamber choir), dancing, music, running, cycling
  • international experience


AWARDS AND GRANTS

2017:DAAD research scholarship for a project at ICSI, Berkeley
2016:YERUN scholarship for Big Data and Analytics Summer School at the University of Essex
2015:Scholarship of University of Bremen for 29th Machine Learning Summer School, Kyoto
2005-2009:Scholarship of Hans-Böckler Stiftung (Hans Böckler Foundation)


WORK EXPERIENCE

since 02/17:

Postdoctoral Research Scholar at ICSI (International Computer Science Institute), University of California Berkeley, Supervisor: Gerald Friedland

  • implementing tools for performing studies in robotics, environmental science, social sciences, biology, etc. using the YFCC100m multimedia dataset
  • leading a team of 5 undergraduate students for URAP
05/15-01/17:

Sr. Scientist at the Robotics Group, University of Bremen, Head: F Kirchner

  • organizer of the machine learning and optimization workgroup
  • organizer of the signal processing workgroup
  • work in different projects like RECUPERA-Reha, BesMan, Entern, Robocademy
  • support of projects with or for the industry (health devices, cars, airplanes, wearables, telecommunication, art, employer’s liability insurance association)
  • project acquisition (contribution to more than 10 project proposals, successful H2020 proposal for the ESA data fusion project: InFuse)
  • counseling of students and employees in the context of machine learning or pySPACE
  • improved (online) algorithms to better operate when few data or few resources are available
  • software development (lead developer of pySPACE)
  • master thesis supervision (data selection strategies for SVMs; automatic processing chain optimization)
07/10-04/15:

Scientist at the Robotics Group, University of Bremen, Head: F Kirchner

  • successfully finished the project IMMI (intelligent man-machine interface)
  • general concepts for connecting SVM variants to improve understanding (especially for teaching) and usability
  • backtransformation (new visualization concept for signal processing chains)
  • new parameter optimization algorithms and sensor selection algorithms
  • lead developer of pySPACE
  • open source release of pySPACE (improving, refactoring, and restructuring)
  • improved documentation, documentation concept, automatic API documentation generator
  • support for other software users and large simplification of user interface
  • generic unit testing framework in pySPACE
  • numerous further improvements of pySPACE (e.g., multi-class classification, regression, bug fixes, style improvements, data handling)
  • contribution to project proposals
  • supervision of student assistants and a master thesis
05/09-06/10:

Jr. Scientist at the DFKI GmbH (German Research Center for Artificial Intelligence), Robotics Innovation Center, Bremen, Head: F Kirchner

  • successfully finished the project VI-Bot (exoskeleton operator surveillance)
  • contribution to the project proposal of the follow up project IMMI (e.g., literature research on brain-computer interfaces)
  • classification, performance evaluation, and further algorithms added to pySPACE
  • faster processing by algorithm tuning in pySPACE
  • mathematical model for space simulation in the project Inveritas


UNIVERSITY TEACHING

Semester Type Title Organizer
SP2017 seminar Undergrad. Research Apprentice Program G Friedland
WS2016 seminar decision models in natural sciences HG Döbereiner
WS2016 complete lecture machine learning for autonomous robots F Kirchner
SS2016 lecture+tutorial reinforcement learning F Kirchner
WS2015 complete lecture machine learning for autonomous robots F Kirchner
SS2015 lecture reinforcement learning F Kirchner
SS2015 corrected exams fundamentals in computer science 2 F Kirchner
WS2014 coordination behaviour based robotics F Kirchner
WS2014 lecture+coord. machine learning for autonomous robots F Kirchner
WS2013 lecture+tutorial machine learning for autonomous robots F Kirchner
SS2012 tutorial analysis 2 (mathematics) B Stratmann
WS2011 tutorial analysis 1 (mathematics) B Stratmann
SS2010 tutorial mathematics 2 (computer science) R Stöver
WS2009 tutorial mathematics 1 (computer science) R Stöver
before exercise sheets corrections for mathematics lectures Various


MAJOR PUBLICATIONS


TALKS AND WORKSHOPS

  • pySPACE workshop (2015), deep learning workshop (2016), machine learning workshop (2016), DFKI RIC, Bremen, Germany
  • Representation of the DFKI RIC at the CeBIT international computer expo (2015), Hannover, Germany
  • Introduction to the Signal Processing and Classification Environment pySPACE (2014), PyData Berlin 2014, Berlin, Germany
  • Our Tools for Large Scale or Embedded Processing of Physiological Data (2014), Passive BCI Community Meeting, Delmenhorst, Germany
  • Introduction to pySPACE workflows (2013), NIPS workshop Machine Learning Open Source Software: Towards Open Workflows, Lake Tahoe, Nevada, USA


RESEARCH TOPICS

Framework - pySPACE:
 

pySPACE is the abbreviation for signal processing and classification environment written in Python which is supporting parallelization and intuitive configuration (based on YAML).

I am not the original developer of pySPACE but I am the lead developer and I contributed the major parts to it like documentation, usability, numerous algorithms, tuned performance, classification, regression, evaluation metrics, parameter optimization, etc. Furthermore, I was responsible for making this software open source.

Currently, I am working on optimizing processing chains. One of my approaches uses deep learning on EEG data and the other is to integrate domain/expert knowledge and the optimization library Hyperopt into pySPACE.

Intelligent Man-Machine Interaction (IMMI):
 

I contributed to the proposal of this project and I have worked in IMMI from 05/10 to 04/15. My tasks involved the optimization, descriptive analysis, and online adaptation of data processing chains for electroencephalographic (EEG) data. One task to detect the perception of rare infrequent important events in contrast to the perception of more frequent but irrelevant in the very noisy data. The other task was to predict upcoming movements by detecting the movement preparation in the EEG. Most parts of my thesis are results from my work in this project and they are related to support vector machines, decoding of EEG data processing chains, and the processing optimization with the framework pySPACE.

Support Vector Machines (SVMs):
 

Due to their generalization capability on few data with high dimensions, the SVM is a common classifier in EEG data processing. I encountered numerous variants of this algorithm and derived general and intuitive concepts how these variants are connected. The approaches could be used to improve the understanding of these algorithms and to easily teach a class of algorithms to students. My concept included the (smooth) connections to linear discriminant analysis, support vector regression, relative margin machine, one-class SVM, and the online passive-aggressive algorithm. Recently, I compared different possibilities of online learning SVMs which consider resource limitations.

Decoding:

To avoid that a complex data processing chain is relying on artifacts it is crucial to determine how it interacts with the data in total. I implemented a generic approach to decode (and visualize) these processing chains. As a side product, I developed and compared algorithms which are capable of reducing the segment length and the number of sensors used in the processing of segmented time series data from multiple sensors (e.g., EEG and robotics data). Currently, I am working on improving the interpretability with the help of source localization methods.

RECUPERA-Reha:

I have been mainly supporting this project since 05/2015. Its objective is to create a full-body exoskeleton for the support in stroke rehabilitation. EEG data is used to improve the exoskeleton control and to get insights into the rehabilitation process of the user. Additionally to the processing chain decoding, I am contributing with approaches that tackle the major problem of too few training data.

Robotics:

Even though my main responsibility has always been on EEG data processing, I regularly support colleagues with mathematical or machine learning problems in robotics. This includes processing chain construction/optimization, soil detection, sensor fault detection, outlier detection, underwater vehicle movement modeling, reinforcement learning, automotive control, etc.


REVIEWING

  • Robotics Group and Robotics Innovation Center for internal quality control before first submission,
  • Pattern Recognition, Expert Systems with Applications, Information Sciences,
  • Chemometrics and Intelligent Laboratory Systems, Biomedical Signal Processing and Control, International Journal of Machine Learning and Cybernetics, Neural Computing and Applications, and Recent Patents on Electrical & Electronic Engineering


MINOR PUBLICATIONS