Ábel Ilyés-Kun

I am Ábel, a computer science graduate from RWTH Aachen University. My academic interests include AI for music generation, natural language processing, and computational neuroscience. Beyond my studies, I am passionate about playing the piano and enjoy transcribing recordings from my favorite musicians, Brad Mehldau and Chick Corea.

Email | GitHub | LinkedIn | CV

News

  • [October 2025] I will begin my master thesis on real-time human-AI cooperative jamming at the Chair for Artificial Intelligence Methodology (AIM) at RWTH Aachen.
  • [September 2025] My co-authored paper "Generating Piano Music with Transformers: A Comparative Study of Scale, Data and Metrics" was accepted at the NeurIPS 2025 Workshop on AI for Music!
  • [September 2024] I started my exchange semester at the Korean Advanced Institute of Science and Technology (KAIST) and received the DUO-Korea scholarship.

Research and Projects

🎹 Real-Time Human-AI Cooperative Jamming

Fall 2025

I am currently pursuing my master’s thesis at the Chair for Artificial Intelligence Methodology of RWTH Aachen. The goal is to develop a real-time system for both educational and creative applications where a human musician cooperates with the agent on a MIDI instrument, e.g., the Yamaha Disklavier. Inspired by recent real-time music generation projects like jam_bot and ReaLchords, this work aims to enable planned improvisation: given a Jazz lead sheet, the system will interactively accompany the performer by generating harmonies for a played melody, or conversely, by providing melodic lines over harmonies played by the user.

🎹 Generating MIDI Piano Performance with Transformers

Summer 2025

Confusion Matrix

As part of a university lab project, I worked on generating MIDI piano performances with Transformers. We systematically compared different datasets, model architectures, model sizes, and training strategies to evaluate their impact on generative quality. To support model development and evaluation, we examined a range of quantitative metrics and analyzed how well they correlate with human judgment collected through listening studies. Our best-performing model, a 950M-parameter transformer trained on 80K MIDI files from diverse genres, produces outputs that are often rated as human-composed in a Turing-style listening survey.

VR Game for Learning Git

Summer 2025

VR_Git

Co-developed a 3D VR game in Unity to teach git in an interactive, hands-on environment. Branches are represented as color-coded shelves, and files as items that can be put on the shelves (e.g., cubes for .py, books for .docx), allowing students to visualize and experiment with core git commands like add, commit, merge, and push. The user can trigger git commands from a UI panel with the VR controller and observe the effect in the immersive environment. The game provides a risk-free space to build mental models, reinforce correct workflows, and reduce fear of mistakes, preparing learners for real-world git projects.

🐍 Mamba State-space Model

Summer 2024

Over the past summer at my home university, I participated in a research seminar at the Machine Learning and Reasoning chair involving Mamba, a recent state-space model. The experience sparked my interest in continuing to explore state-space models on music data! Since Mamba-variants can process extremely long sequences more efficiently than Transformers, it can be interesting to see how they handle long temporal dependencies in music data.

🧠 Thesis at Institute for Computational and Systems Neuroscience

October 2022 – May 2023

Bachelor’s thesis on hyper-parameter optimization: The goal here was to optimize the hyper-parameters of a spiking neural network simulator with Optuna. The work involved parallel computation on the JURECA cluster and experimentation with sampling algorithms (TPE, random).

🚙 Practical at Cyber-Physical Mobility Lab

October 2021 – February 2022

Implemented trajectory planning and collision avoidance for model vehicles. Scrum-based workflow in a team of six using Git.

🎵 Computer-generated Music

Summer 2020

During my bachelor degree at RWTH Aachen University, I did a seminar on computer-generated music, where I covered recent neural-network-based approaches like Google Magenta or the Bachbot, also discussing the LZ compression algorithm within the OpenMusic software.



Work Experience

Tutor at Research Group for Programming Languages and Verification

October 2021 – March 2022, October 2022 – March 2023

Taught weekly classes (~15 students), graded coding assignments and exams. Tutored students in Java, Haskell, Prologand Verification.

Internship at BWI GmbH

March 2020

Intro to Scrum-based project management. Observed live JIRA workflows.



Education

Kaist Logo

KAIST MSc Computer Science Exchange, AI Track
September 2024 – December 2024
Received DUO-Korea Scholarship

RWTH Logo

RWTH Aachen University MSc Computer Science, AI Track
October 2023 – Present

RWTH Logo

RWTH Aachen University BSc Computer Science
October 2019 – June 2023
- Thesis at Forschungszentrum Jülich on hyperparameter optimization
- Projects on neuro-inspired computing, microcontroller programming, and model vehicle control

Goethe Gymnasium Bad Ems A levels (Abitur)
- Majors: English, Mathematics, Physics
- DPG Abitur Prize in Physics



Programming Stack

Languages & Tools
Python, C++, Java, C#, SQL, Bash, Git, Linux, Slurm

Machine Learning & Deep Learning
PyTorch, NumPy, Pandas, Matplotlib, Optuna, Jupyter



Societies

  • Member of 창작동화 Jazz Band (2024–)
  • Rowing Club Bad Ems (2016–2019)


Interests

  • Music Information Retrieval
  • Natural Language Processing
  • Jazz Piano
  • Video and Music Editing
  • Running, Table Tennis, Football
  • Psychology