Biography
Hi, I’m Lukas, and I am a Machine Learning and Cheminformatics Intern at Pangea Bio, where I apply my strong background in machine learning and data science to solve challenging problems in drug discovery and natural product research. I’m passionate about using machine learning to make a positive impact on human life!
Nine years of experience in deep learning, from training reinforcement learning agents on real robots to finetuning a large language model on ethnobotanical data. Currently looking for a job!
- Machine Learning
- Data Science
- Cheminformatics
- Large Language Models
- Robotics
-
M.Sc. in Computer Science, 2019
University of Freiburg
-
B.Sc. in Computer Science, 2015
University of Freiburg
Featured Work
News
- 2 Jun 2023 We received the 2022 IEEE Robotics and Automation Letters Best Paper Award for CALVIN: A Benchmark for Language-conditioned Policy Learning for Long-horizon Robot Manipulation Tasks
Skills
Experience
- Benchmarking machine learning models and molecular fingerprints for chemical activity prediction using RDKit.
- Fine‑tuning Llama2 and prompt engineering of LLMs for medical entity and relationship extraction from ethnobotanical literature.
- Integration of heterogeneous data sources such as ethnobotany, natural product datasets, and bioactivity databases like PubChem and ChEMBL and implementation of analysis workflow to prioritize potential compound‑target pairs.
- Building docker images and cloud deployment on AWS, Lambda Labs and Vast.ai.
- Machine learning research for robot manipulation.
- Created CALVIN, a benchmark, dataset and state-of-the-art Transformer‑based architecture for learning language‑conditioned robot control policies from unstructured data.
- Self-supervised learning of robotic skills from demonstrations (HULC).
- Optimized dataloading and distributed training on high‑performance SLURM cluster.
- Developed a Python framework for the fast design of platform-independent robot experiments.
- Implemented robot control on three different robots (KUKA iiwa, Franka Emika Panda, UR3).
- Designed and implemented curriculum learning strategies for deep reinforcement learning based on Proximal Policy Optimization.
- Successfully applied the algorithm to solve real‑world robot manipulation tasks with KUKA LBR iiwa (ICRA 2020 paper).
Education
Minor: Cognitive Science
Thesis: Adaptive Curriculum Generation from Demonstrations
Supervised by Prof. Dr. Wolfram Burgard
Specializations:
- Robotics
- Machine Learning
- Computer Vision
Minor: Cognitive Science
Thesis: Hand Orientation Estimation using Deep Neural Networks
Supervised by Prof. Dr. Wolfram Burgard
Students Supervised
- Ilia Dobrusin, Self-Supervised Consistency Loss for Sim-to-Real Domain Adaptation, Master Thesis, 2021.
- Mikel Martinez, Self-supervised Control with Vision and Language, Master Project, 2021.
- Jessica Borja, Affordance Learning from Play for Sample-Efficient Policy Learning, Master Project, 2021.
- Deep Learning Lab Course, 2021