Mark Leznik joined the Institute of Resource Information Management in January 2018 as a research associate. He holds a M.Sc. degree in Computer Science from Ulm University, which he finished in 2017. The title of his Master's thesis was "Luminance Estimation of Colorized Near-Infrared Images", the work was performed at Daimler AG Research & Developement. Mark obtained his B.Sc. in Computer Engineering from Ulm University of Applied Science in 2013, the subject of his Bachelor's thesis was "Low-light Image Quality Optimization of HDR Cameras", the work was performed at Mercedes-Benz Cars Developement.
As part of my thesis, I am currently working on a range of topics related to time series data (cf. Research Topics Figure).
The time series analysis part (achieved by using both statistics and applied machine learning) includes forecasting, clustering, classification, anomaly detection and quality metrics of time series data. Hereby, I mainly focus on gathering more understanding into anomalies in time series data, as well as providing a set of tools to allow an obejctive analyis of the underlying time series data in terms of its quality (in regards to predictability and general characteristics).
The synthesis part refers to the artificial generation of time series data, mainly for the purpose of its anonymization. Such a mechanism allows for a privacy preserving data publication or exchange for publication or collaboration purposes respectively.
I am also working on intertwining data science methodologies with CI/CD DevOps principles. More precisely, this means adding a higher grade of automation to the developement and testing of data science components (i.e. model training) and better reproducibility across multiple environments (i.e. local workstations, HPC clusters).
- STEALTH: Anonymisation through Privacy-preserving Data Generation
- January 2020 - current
- Vector Stiftung
- AI Investments – an advanced investment tool based on machine learning and big data.
- Subproject: STOQS, Simple Timeseries Objective Quality Measurement Stack
June 2019 - current
- RECAP - Reliable Capacity Provisioning and Enhanced Remediation for Distributed Cloud Applications
- January 2018 - December 2019
Theses & Student Projects
If you are interested in writing a thesis, seminar or doing a project on the topics described above, here are some ideas. If anything sounds interesting, or if you have your own idea, feel free to write me a mail or drop by my office.
Ongoing & Completed Topics
- Evaluating GAN Architectures for Time Series Synthesis (Master Thesis)
- General Game Playing Approaches for Financial Time Series Analysis (Master Thesis)
- Synthesizing Cloud Server Workloads Using Generative Adversarial Networks (Master Thesis)
- The Current State of AutoML, A Comparative Overview (Bachelor Thesis)
- Susceptible Artificial Data Generation Using Generative Adversarial Networks (Bachelor Thesis)
- Current Advances in Time Series Anomaly Detection (Bachelor Thesis)
- An LSTM Framework based on Tensorflow & Keras for Time Series Prediction Using Production and Artificial Data (Student Project)
- The AutoML Jungle – An Overview (Seminar)
- Anomaly Detection in Time Series Data: An Overview (Seminar)
- Is the Cloud ready for Autonomous Driving? (Seminar)
- Deep Learning and Containers? What Gives? (Seminar)
- Synthetic Data Generation Using Deep Learning, An Overview (Seminar)
- General Game Playing and Machine Learning, The Road to MuZero (Seminar)
- Interpretability of Deep Learning (Seminar)
- Explainable and Interpretable Machine Learning, An Overview (Seminar)