CURRICULUM VITAE¶
Dr. MARIO MICHAEL KRELL¶
| krell at uni-bremen.de | languages | English (fluent), German (fluent), French (basic) | |
| address | Redwood City | hobbies | choir singing, ballroom dancing, hiking/running |
SUMMARY
- Senior Machine Learning Researcher with 13-year of cumulative experience in many applications
- Strong problem solving, mathematical analysis, interdisciplinary teamwork, and leadership skills (10 years)
- Interested in leading applied research in the Bay Area that helps humanity
EDUCATION
| Time | Degree | Institute | Advisor | GPA |
| 2017 | Postdoc | UC Berkeley, ICSI, USA | G Friedland | |
| 03/15 | PhD in CS (machine learning) | University of Bremen, Robotics Group | F Kirchner | 4.0 |
| 03/09 | degree in mathematics (“Diplom”) | Humboldt University of Berlin | B Kummer | 3.9 |
SKILLS
- problem solving, machine learning, deep learning, mathematical analysis, optimization
- leadership experience with up to 8 direct reports, stakeholder interaction, and as scrum master/PO
- collaboration in multicultural/interdisciplinary teams (engineers, computer/neuro-scientists, manager, PO)
- independent research and data analysis (>40 publications, >500 citations), teaching, scientific presentation
- software development: Python, NumPy, Git, Sphinx, HPC, TensorFlow, PyTorch, documentation, PySpark
- GNN, SVM, SVR, CNN (ResNet-50), NLP (BERT), probabilistic models (ABC, AdGMoM), RL (MiniGo), evaluation, capacity, differential privacy, signal processing, clustering, source localisation
- basic knowledge in brain-machine interfaces, robotics, multimedia, cars, hardware acceleration (IPUs)
SHORT WORK SUMMARY
| Time | Title | Employer | Reference |
| 08/22 - 10/22 | AI Engineering Manager | Graphcore | J Irwin |
| 09/19 - 08/22 | AI Applications Specialist | Graphcore | M Iyer |
| 01/18 - 07/19 | Principal ML/Data Scientist | Mercedes-Benz R&D, USA | H Endt |
| 02/17 - 12/17 | Postdoc | UC Berkeley, ICSI, USA | G Friedland |
| 03/15 - 01/17 | Postdoc and Senior ML Researcher | University of Bremen, GER | F&E Kirchner |
| 07/10 - 03/15 | Machine Learning Researcher | University of Bremen, GER | S Straube |
| 05/09 - 06/10 | Jr. ML Researcher | DFKI GmbH, Bremen, GER | A Seeland |
AWARDS AND GRANTS
| 2021: | Top performer at Graphcore |
|---|---|
| 2017: | DAAD research scholarship for a project at ICSI, Berkeley |
| 2017: | Lead DFKI activity for H2020 Grant (InFuse), 3.5 Mio. Euro |
| 2017: | Industry project funding by local government (xMove), 200.000 Euro |
| 2017: | Second prize for best student poster at OCEANS 2017 MTS/IEEE Aberdeen |
| Scholarships: | Hans-Böckler Stiftung (2005-2009), University of Bremen (2015), Yerun (2016) |
WORK EXPERIENCE
| 09/19-10/22: | Principal Machine Learning Lead at Graphcore
|
|---|---|
| 01/18-07/19: | Principal Data Scientist at Mercedes-Benz R&D North America in the Statistics, Optimization, Machine Learning, and Analytics (SOMA) team
|
| 02/17-12/17: | Postdoctoral Research Scholar in Machine Learning at ICSI (International Computer Science Institute), University of California Berkeley
|
|---|---|
| 05/15-01/17: | Sr. Machine Learning Researcher at the Robotics Group, University of Bremen, and |
| 07/10-04/15: | Machine Learning Researcher at the Robotics Group, University of Bremen
|
| 05/09-06/10: | Jr. ML Researcher at the DFKI GmbH, Robotics Innovation Center, Bremen |
|---|
RECOMMENDATIONS
James Irwin (Director at Graphcore): “Mario faced the fire-hose fast-track of everything that would normally be exaggerated training examples. He must have grown a full foot in the course of his tenure as an AI engineering manager. A baptism of fire would have been a cake-walk compared to what was thrown at him. He did not flinch, and he did not cower. He did not dismiss as irrelevant & divert to HR truly important personal staff issues that immediately tested his values as a manager. Mario is empathetic and resourceful. He acted, showing he knew his role was to enable the team to be successful. A dazzlingly promising engineering manager. That’s the part that matters here since Mario’s undoubted credibility as an engineer will put a team at ease quickly.”
Phil Brown (VP at Graphcore): “Mario is fantastic to work with. In addition to a strong AI Engineering skill set he is extremely positive, proactive and dedicated, helping lift and enhance his team and colleagues. He made key contributions across a range of areas, including implementing and optimizing a reinforcement learning application with a novel execution scheme, driving our MLPerf submissions - particularly finding a series of significant optimizations on a key CNN benchmark, and working with a number of customers on new Graph Neural Networks with fantastic results. More recently Mario has transitioned successfully into team leadership, helping steer his team through organisational change whilst always keeping the group upbeat and engaged.”
Alexander Tsyplikhin (Sr. Manager): “I had the pleasure and honor to work with Mario at Graphcore. He has incredible technical depth, an ability to learn very quickly, and outstanding communication skills. He led multiple activities at the same time, ranging from MLperf submissions to research papers and customer projects. During his time as IC, he worked on a vast spectrum of ML models and domains: reinforcement learning, computer vision scaling optimization, audio processing, PDE solvers, fraud detection, differential privacy, approximate Bayesian computation, et. al. He was productive and efficient at every project, and he always had time to coach co-workers on technical and non-technical questions. Taking over the team as a manager was a natural change for him. Mario was amazing as a manager, keeping his engineers happy and engaged, including during periods of uncertainty. I will be happy to work with Mario in the future and can recommend him to anyone extremely highly.”
TEACHING
| Semester | Type | Title | Organizer |
|---|---|---|---|
| 2019/20/21 | seminar | work topics related reading group | MM Krell |
| 2019 | seminar | discuss different ML and CS algorithm | MM Krell |
| FA2017 | seminar | Undergrad. Research Apprentice Program | (G Friedland) |
| SP2017 | seminar | Undergrad. Research Apprentice Program | (G Friedland) |
| 2014-17 | seminar | machine learning workgroup | MM Krell |
| 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 |
MAJOR PUBLICATIONS
- Tuple Packing: Efficient Batching of Small Graphs in Graph Neural Networks, MM Krell, M Lopez, S Anand, H Helal, AW Fitzgibbon (2022), arXiv
- Efficient Sequence Packing without Cross-contamination: Accelerating Large Language Models without Impacting Performance, MM Krell, M Kosec, SP Perez, AW Fitzgibbon (2022), arXiv
- Hardware-accelerated Simulation-based Inference of Stochastic Epidemiology Models for COVID-19, S Kulkarni, MM Krell, S Nabarro, CA Moritz (2022), In ACM Journal on Emerging Technologies in Computing Systems 18(2): 1-24, doi: 10.1145/3471188
- Accelerating ResNet-50 Training on the IPU: Behind our MLPerf Benchmark, MM Krell, Zhenying Liu, Emmanuel Menage, Bartosz Bogdanski (2022), Towards Data Science
- A Capacity Scaling Law for Artificial Neural Networks, G Friedland, MM Krell (2018), arXiv
- Data Augmentation for Brain-Computer Interfaces: Analysis on Event-Related Potentials Data, MM Krell, A Seeland, SK Kim (2018), arXiv
- Empirical comparison of distributed source localization methods for single-trial detection of movement preparation, A Seeland, MM Krell, S Straube, EA Kirchner (2018), In Front. Hum. Neurosci., doi: 10.3389/fnhum.2018.00340
- Field Studies with Multimedia Big Data: Opportunities and Challenges (Extended Version), MM Krell, J Bernd, D Ma, J Choi, D Borth, G Friedland (2017), arXiv
- Generalizing, Decoding, and Optimizing Support Vector Machine Classification, MM Krell (2015), PhD Thesis, University of Bremen, Bremen, 1-236
- How to evaluate an agent’s behaviour to infrequent events? — Reliable performance estimation insensitive to class distribution, S Straube, MM Krell (2014), In Front. Comput. Neurosci. 8(43): 1-6, doi:10.3389/fncom.2014.00043
- pySPACE — a signal processing and classification environment in Python, MM Krell, S Straube, A Seeland, H Wöhrle, Johannes Teiwes, JH Metzen, EA Kirchner, F Kirchner (2013), In Front. Neuroinform. 7(40): 1-11, doi:10.3389/fninf.2013.00040
- Comparison of Sensor Selection Mechanisms for an ERP-Based Brain-Computer Interface, David Feess, MM Krell*, JH Metzen (2013), In PLoS ONE 8(7): e67543 1-9, doi:10.1371/journal.pone.0067543
- Generalized Derivatives in Nonsmooth Analysis: Connections and Computability, MM Krell (2009), diploma thesis, Humboldt University of Berlin, Berlin, 1-91
PRESENTATIONS AND WORKSHOPS
- Graphcore at Fürberg Workshop: Hybrid AI - combining symbolic, deep learning and neuromorphic (2022)
- Software/hardware co-optimization on the IPU: An MLPerf™ case study (2021), Hot Chips 33 Symposium
- pySPACE workshop (2015), DL workshop (2016), ML workshop (2016), DFKI RIC, Bremen, Germany
- Representation of the DFKI RIC at the CeBIT international computer expo (2015), Hannover, Germany
- Introduction to 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
REVIEWING
ICML, NeurIPS, Pattern Recognition, Expert Systems with Applications, Information Sciences, Sensors, IEEE MultiMedia, ACM Multimedia, Chemometrics and Intelligent Laboratory Systems, Biomedical Signal Processing and Control, International Journal of Machine Learning and Cybernetics, Neural Computing and Applications, Recent Patents on Electrical & Electronic Engineering, Progress in Artificial Intelligence, Neuroadaptive Technology Conference, and internal reviews
MINOR PUBLICATIONS
- Packing: Towards 2x NLP BERT Acceleration, M Kosec, Sheng Fu, MM Krell (2021), arXiv
- NanoBatch DPSGD: Exploring Differentially Private learning on ImageNet with low batch sizes on the IPU, EH Lee, MM Krell, A Tsyplikhin, V Rege, E Colak, KW Yeom (2021), arXiv
- A First Step Towards Distribution Invariant Regression Metrics, MM Krell, B Wehbe, (2020), arXiv
- Classifier Transfer with Data Selection Strategies for Online Support Vector Machine Classification with Class Imbalance, MM Krell, N Wilshusen, A Seeland, SK Kim (2017), Journal of Neural Engineering 14(2), IOP Publishing, doi: 10.1088/1741-2552/aa5166
- Backtransformation: A new representation of data processing chains with a scalar decision function, MM Krell, S Straube (2017), Advances in Data Analysis and Classification 11 (2): 415-439, doi:10.1007/s11634-015-0229-3
- Online Model Identification for Underwater Vehicles through Incremental Support Vector Regression, B Wehbe, A Fabisch, MM Krell (2017), IROS 2017
- Learning Coupled Dynamic Models of Underwater Vehicles using Support Vector Regression, B Wehbe, MM Krell (2017), OCEANS 2017
- hyperSPACE: Automated Optimization of Complex Processing Pipelines for pySPACE, T Hansing, MM Krell, F Kirchner (2016), NIPS workshop: BayesOPT2016
- raxDAWN: Circumventing Overfitting of the Adaptive xDAWN, MM Krell, A Seeland, H Wöhrle (2015), In International Congress on Neurotechnology, Electronics and Informatics: 68-75, ScitePress, doi:10.5220/0005657500680075
- Accounting for Task-Difficulty in Active Multi-Task Robot Control Learning, A Fabisch, JH Metzen, MM Krell, F Kirchner (2015), In KI - Künstliche Intelligenz, 1-9, doi:10.1007/s13218-015-0363-2
- An Adaptive Spatial Filter for User-Independent Single Trial Detection of Event-Related Potentials, H Wöhrle, MM Krell, S Straube, SK Kim, EA Kirchner, F Kirchner (2015), In IEEE Transactions on Biomedical Engineering 62(7): 1696-1705, doi:10.1109/TBME.2015.2402252
- New one-class classifiers based on the origin separation approach, MM Krell, H Wöhrle (2015), In Pattern Recognition Letters 53: 93-99, doi:10.1016/j.patrec.2014.11.008
- Balanced Relative Margin Machine - The Missing Piece Between FDA and SVM Classification, MM Krell, D Feess, S Straube (2014), In Pattern Recognition Letters 41: 43-52, doi:10.1016/j.patrec.2013.09.018
- On the Applicability of Brain Reading for Self-Controlled, Predictive Human-Machine Interfaces in Robotics, EA Kirchner, SK Kim, S Straube, A Seeland, H Wöhrle, MM Krell, M Tabie, M Fahle (2013), In PLoS ONE 8(12): e817321-19, doi:10.1371/journal.pone.0081732
OTHER PUBLICATIONS
- Accelerating Simulation-based Inference with Emerging AI Hardware, S Kulkarni, A Tsyplikhin, MM Krell, CA Moritz (2020), In Proceedings of IEEE ICRC
- Benchmarking the Performance of Accelerators on National Cyberinfrastructure Resources for Artificial Intelligence / Machine Learning Workloads, A Nasari, H Le, R Lawrence, Z He, X Yang, MM Krell, A Tsyplikhin, M Tatineni, T Cockerill, L Perez, D Chakravorty, H Liu (2022), In Practice and Experience in Advanced Research Computing (PEARC ‘22). ACM 19: 1–9, doi:10.1145/3491418.3530772
- A Practical Approach to Sizing Neural Networks, G Friedland, A Metere, MM Krell (2018), arXiv
- Learning of Multi-Context Models for Autonomous Underwater Vehicles, B Wehbe, O Arriaga, MM Krell, F Kirchner (2018), IEEE OES Autonomous Underwater Vehicle Symposium
- Rotational Data Augmentation for Electroencephalographic Data, MM Krell, SK Kim (2017), 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC’17)
- OrigamiSet1.0: Two New Datasets for Origami Classification and Difficulty Estimation, D Ma, G Friedland, MM Krell (2018), In Proceedings of Origami Science Maths Education, 7OSME, Oxford UK
- Learning Magnetic Field Distortion Compensation for Robotic Systems, L Christensen, MM Krell, F Kirchner (2017), In Proceedings of IROS 2017
- Recupera-Reha: Exoskeleton technology with integrated biosignal analysis for sensorimotor rehabilitation, EA Kirchner et al. (2016), At 2nd trans-disciplinary conference “Technical support systems that people really want”: 535-548, Elsevier
- Comparison of Data Selection Strategies for Online Support Vector Machine Classification, MM Krell, N Wilshusen, AC Ignat, SK Kim (2015), In International Congress on Neurotechnology, Electronics and Informatics: 59-67, ScitePress, doi:10.5220/0005650700590067
- Concept of a Data Thread Based Parking Space Occupancy Prediction in a Berlin Pilot Region, T Tiedemann, T Vögele, MM Krell, JH Metzen, F Kirchner (2015), In Papers from the 2015 AAAI Workshop. Workshop on AI for Transportation (WAIT-2015), Austin, USA, AAAI Press, 58-63
- Generalizing, Optimizing, and Decoding Support Vector Machine Classification, MM Krell, S Straube, H Wöhrle, F Kirchner (2014), In Proceedings of the ECML/PKDD-2014, Nancy
- Reconfigurable Dataflow Hardware Accelerators for Machine Learning and Robotics, H Wöhrle, J Teiwes, MM Krell, A Seeland, EA Kirchner, F Kirchner (2014), In Proceedings of the ECML/PKDD-2014, Nancy
- Introduction to pySPACE, MM Krell, PyData Berlin 2014, Berlin, Germany (2014)
- Memory and Processing Efficient Formula for Moving Variance Calculation in EEG and EMG Signal Processing, MM Krell, M Tabie, H Wöhrle, EA Kirchner (2013), In International Congress on Neurotechnology, Electronics and Informatics: 41-45, ScitePress, doi:10.5220/0004633800410045
- A Dataflow-Based Mobile Brain Reading System on Chip with Supervised Online Calibration, H Wöhrle, J Teiwes, MM Krell, EA Kirchner, F Kirchner (2013), In International Congress on Neurotechnology, Electronics and Informatics: 46-53, ScitePress, doi:10.5220/0004637800460053
- Introduction to pySPACE workflows, MM Krell, NIPS workshop Machine Learning Open Source Software: Towards Open Workflows, Lake Tahoe, Nevada, USA (2013)
- Choosing an Appropriate Performance Measure: Classification of EEG-Data with Varying Class Distribution, S Straube, JH Metzen, A Seeland, MM Krell, EA Kirchner (2011), Proceedings of the 41st Meeting of the Society for Neuroscience 2011, Washington, DC, USA