
Research Assistant | School of Engineering and
Applied Sciences, Harvard University
Time: Jun 2021 - Present. Advisor:
Prof. Hanspeter Pfister
As an AI enthusiast with a solid theoretical foundation and considerable practical experience, my mission is to develop innovative AI solutions that can make a positive impact on the world. I earned both my B.Sc. in Engineering Science and my M.Sc. in Robotics, Cognition, Intelligence from the Technical University of Munich (TUM). During my Master's studies, I had the honor of being invited to Harvard's Visual Computing Group (VCG) as a research fellow. There, I was given the opportunity to lead the development of CySGAN, a 3D instance segmentation method for neurobiology. Currently, I'm continuing my work at Harvard as a Research Assistant, advancing the development of SynAnno, a proofreading tool for synaptic polarity annotations. Between my undergraduate and graduate studies, I spent a year in Berlin working as a Data Scientist at deevio and later as a Technical Project Lead and Integrated Systems Engineer at Predicpro (now Stryza), a WATTx venture. While completing my M.Sc., I gained valuable experience as a working student in the perception and software development department at Agile Robots AG, Germany's first robotics unicorn, and at the national research institute fortiss. Most recently, I held the position of Research Software Engineer at Merantix Momentum, an AI Consultancy and Research Hub. In this role, I developed MLOps tools for scalable and traceable ML training, designed CI/CD systems for safety-critical ML components, and provided MLOps training to my colleagues.
Research Assistant | School of Engineering and
Applied Sciences, Harvard University
Time: Jun 2021 - Present. Advisor:
Prof. Hanspeter Pfister
Research Software Engineer | Research Team, Merantix
Momentum
Time: Feb 2023 - Jul 2023. Mentor:
Dr. Johannes Otterbach
Mr. Lauenburg is distinguished by his excellent team-oriented work ethic, outstanding communication skills, and very high flexibility. - Dr. Johannes Otterbach
Research Fellow | School of Engineering and Applied
Sciences, Harvard University
Time: Oct 2021 - Jul 2022. Advisor:
Prof. Hanspeter Pfister
We would like to strongly recommend him [Leander] for any academic programs or industry positions without reservation. - Prof. Hanspeter Pfister
Working Student | Department for Perception and
Software Development, AgileRobots
Time: Dec 2020 - Jul 2021. Mentor:
Patrick Schulte
Leander has been an exceptional dedicated and driven team member who consistently demonstrated his highly efficient and independent work style. - Patrick Schulte
Entrepreneur in Residence (Tech Lead) | Stryza
(formerly Predicpro), WATTx
Time: May 2019 - Oct 2019. Mentor:
Dr. Martin Mittermeier
Leander's work laid the new foundation for Predicpro [Stryza] for which we are entirely grateful. His vision and hard-work ultimately resulted, among other things, in a pilot project with a DAX listed company." - Dr. Martin Mittermeier
Data Scientist Intern | Engineering Team, deevio
Time: Nov 2018 - May 2019. Mentor:
Dr. Tassilo Glander
We would like to highlight Leander’s versatility and independent working style, as he proactively identified and resolved issues related to his current tasks. - Dr. Tassilo Glander
Working Student | Smart Energy Team, fortiss
Time: Jun 2017 - Sep 2018. Mentor:
Dr. Markus Duchon
Mr. Lauenburg shows an extraordinary level of commitment, an exceptional willingness to perform, and a great eagerness to learn. - Dr. Markus Duchon
SynAnno: Proofreading Tool for Synaptic Polarity
Annotations
Leander Lauenburg, Zudi Lin, Jakob Troidl,
Johanna Beyer, Hanspeter Pfister, Donglai Wei
SynAnno is a tool for proofreading and correcting
synaptic polarity annotations in electron microscopy
volumes. The tool is developed by Harvard's Visual
Computing Group and Lichtman Lab, it integrates with
Seung Lab's CAVE (Connectome Annotation Versioning
Engine). This project is currently being prepared for
first-author submission to IEEE VIS 24.
Code
Lynx
Leander Lauenburg
Lynx is a lightweight tool that automates and schedules
sequential ML subtasks, tailored for experimental ML MVPs
that are not ready for a full-fledged Flyte setup. Ideal
for research, testing, and debugging, Lynx can run tasks
once or on a cronjob-like schedule, executing pipelines in
the foreground or background.
Code
3D Domain Adaptive Instance Segmentation via Cyclic
Segmentation GANs
Leander Lauenburg, Zudi Lin, Ruihan Zhang, Márcia dos
Santos, Siyu Huang, Ignacio Arganda-Carreras, Edward S.
Boyden, Hanspeter Pfister, Donglai Wei
(IEEE JBHI, 2023)
The paper '3D Domain Adaptive Instance Segmentation via
Cyclic Segmentation GANs', is available via open access in
the IEEE Journal of Biomedical and Health Informatics. The
paper is the result of my Master's thesis written during
my research stay with the Visual Computing Group at
Harvard.
Research Code
Production Code
IEEE JBHI
Master's Thesis
Reinforcement Learning for Solving Robotic Reaching Tasks
in the Neurorobotics Platform
Márton Szep, Leander Lauenburg, Kevin Farkas, Xiyan
Su, Chuanlong Zang
(HBP Student Conference, 2022)
The paper is my team's contribution to the master course
"Cloud-Based Machine Learning in Robotics" (grade 1.0). I
presented the work at the 6th Human Brain Project Student
Conference.
Code
arXiv Paper
Conference Abstract
PackLog Solutions: Estimation Tool for Streamlining
Logistic Operations
Franz Schubart, Leander Lauenburg, Lukas Pries
Marcel, Perez San Blas
(First Place in TechChallenge BEFIVE, Innovation
Challenge by
UnternehmerTUM, 2021)
In close collaboration with our industry partners, we
developed an MVP for streamlining planning and communication
processes between the various players in the logistics of
large manufacturers.
Code
Video
ELA - Object SLAM
Leander Lauenburg, Andy Chen, Ezgi Cakir
The work is my team's contribution to the master course
"Advanced Topics in 3D Computer Vision" (grade 1.0). It is
an extension of
CubeSLAM: Monocular 3D Object SLAM, IEEE Transactions on
Robotics 2019, S. Yang, S. Scherer. In addition to cleaning up, streamlining, and dockerizing
CubeSLAM, we improved the work by adding dynamic object
filtering, object class-dependent scaling, and embedding
stream enrichments.
Code