Posted on May 26, 2025
| 1 minutes
• Other languages: 🇩🇪 Deutsch
In this (german-language) episode of Code for Thought, Carina Haupt (DLR) and I talk with host Peter Schmid about large language models in Research Software Engineering: where tools like Copilot genuinely help (routine tasks, tests, small scripts), and where real-world projects expose their limits (builds, requirements, missing project context). I share my current work on matching papers to repositories via embeddings at the project level rather than just functions - so models carry meaningful context across an entire codebase.
[Read More]
VisoLab is a startup that creates automated self-checkout registers for canteens, where the meals on the tray are recognized on on edge via an iPad, allowing the customer to pay for their meals within seconds without requiring an employee.
The number of available educational resources keeps rising just as the liberty for students
to create a custom curriculum, making recommendation in this sector ever more critical.
There is the need for a methodology to structure these resources unsupervisedly to allow
for explainable recommendations, where users can explicitly demand features such as the
degree how mathematical a course should be. Conceptual Spaces allow to embed resources
into a vector space whose quality dimensions correspond to salient semantic features, but
generating them reliably is an open question. This thesis replicates a method to create these
spaces data-driven from their descriptions and applies it to a given dataset of that domain to
explore its applicability. Our preliminary results show that human categories such as the
faculty a course belongs to are among the detected salient directions. Further qualitative
analysis also indicates that the embeddings capture essential features that can be used in
interpretable classification and explainable recommendation. Additionally, a scalable and
modular architecture for the algorithm is created. It is shown that it can be applied to diverse
variations of dataset and algorithm components. A workflow is implemented and described
to run the algorithm efficiently on compute clusters. The full implementation is open-sourced
to make future research on related problems more accessible.
The number of available educational resources keeps rising just as the liberty for students
to create a custom curriculum, making recommendation in this sector ever more critical.
There is the need for a methodology to structure these resources unsupervisedly to allow
for explainable recommendations, where users can explicitly demand features such as the
degree how mathematical a course should be. Conceptual Spaces allow to embed resources
into a vector space whose quality dimensions correspond to salient semantic features, but
generating them reliably is an open question. This thesis replicates a method to create these
spaces data-driven from their descriptions and applies it to a given dataset of that domain to
explore its applicability. Our preliminary results show that human categories such as the
faculty a course belongs to are among the detected salient directions. Further qualitative
analysis also indicates that the embeddings capture essential features that can be used in
interpretable classification and explainable recommendation. Additionally, a scalable and
modular architecture for the algorithm is created. It is shown that it can be applied to diverse
variations of dataset and algorithm components. A workflow is implemented and described
to run the algorithm efficiently on compute clusters. The full implementation is open-sourced
to make future research on related problems more accessible.
For this thesis, I created a Conceptual Space from Course Descriptions for explainable Recommendation, in a highly performant pipeline on the university-grid.
This text shall serve as a high-level introduction to what I am doing in my master’s thesis. For a longer and more accurate description, it is referred to chapter 2.3 of the thesis.[Read More]
This text shall serve as a high-level introduction to what I am doing in my master’s thesis. For a longer and more accurate description, it is referred to chapter 2.3 of the thesis.[Read More]
The pipeline of this project is computationally expensive, and relies on many different hyperparameters or drop-in algorithms. This makes it the perfect candidate for grid-based parallelization, where different hyperparameter-combinations are executed on different cluster nodes.
[Read More]
The pipeline of this project is computationally expensive, and relies on many different hyperparameters or drop-in algorithms. This makes it the perfect candidate for grid-based parallelization, where different hyperparameter-combinations are executed on different cluster nodes.
[Read More]
In this student research job for the Red Hen Lab, I worked on a pipeline that automatically detects hand gestures in large datasets of videos, involving OpenPose for Pose Recognition, Person detection, Person tracking, Scene detection and more, and deployed it on a HPC using Snakemake and Singularity to create a multimodal communication corpus from the Ellen DeGeneres show.
[Read More]
Delving into the realm of self-driving cars for my bachelor’s thesis, I endeavored to optimize their tactical decisions, particularly speed, using cutting-edge deep neural networks and reinforcement learning in tensorflow within a transformed racing simulation. The resulting research platform, built on Unity, showcased the potential of real-time feedback and adaptability, offering valuable insights into the future integration of reinforcement learning in self-driving cars.
[Read More]
Controlling Self-Driving Race Cars with Deep Neural Networks
Disclaimer: This article was written by chatgpt, as can easily be seen by its style
Navigating the Fast Lane: Reinventing Self-Driving Cars through Reinforcement Learning
Embarking on the journey of my bachelor’s thesis, I delved into the captivating world of self-driving cars, aiming to redefine their capabilities by focusing on the often-neglected tactical decisions, particularly optimizing speed.
Motivation: Unraveling the Dynamics of Self-Driving Cars
Self-driving cars have become a technological marvel, with my motivation rooted in unraveling the intricacies of their operational capabilities. While conventional approaches focus on safety and environmental understanding, my curiosity led me to explore the tactical dimension, specifically the optimization of speed in racing simulations.
Controlling Self-Driving Race Cars with Deep Neural Networks
Disclaimer: This article was written by chatgpt, as can easily be seen by its style
Navigating the Fast Lane: Reinventing Self-Driving Cars through Reinforcement Learning
Embarking on the journey of my bachelor’s thesis, I delved into the captivating world of self-driving cars, aiming to redefine their capabilities by focusing on the often-neglected tactical decisions, particularly optimizing speed.
Motivation: Unraveling the Dynamics of Self-Driving Cars
Self-driving cars have become a technological marvel, with my motivation rooted in unraveling the intricacies of their operational capabilities. While conventional approaches focus on safety and environmental understanding, my curiosity led me to explore the tactical dimension, specifically the optimization of speed in racing simulations.