Murat Onur Yildirim

Hi, I am a PhD candidate in the AMOR/e lab under the supervision of Joaquin Vanschoren at the Mathematics and Computer Science department in TU Eindhoven.

I develop efficient post-training adaptation strategies for continually learning AI systems by leveraging neuroscientific principles such as sparsity and modularity.

Email  -  Scholar  -  Github  -  In  -  CV

Murat Onur Yildirim

News

  • 02/2026, "Automated Machine Learning for Unsupervised Tabular Tasks" has been accepted to MLJ!
  • 02/2026, "Unlocking [CLS] Features for Continual Post-Training" has been accepted to TMLR!
  • 01/2026, "Pruned Adaptation Modules: A Simple yet Strong Baseline for Continual Foundation Models" has been accepted to CPAL!
  • 05/2025, "Self-Regulated Neurogenesis for Online Data-Incremental Learning" has been accepted for an oral presentation at CoLLAs!
  • 04/2025, I have joined the General Assembly of the EU-project SYNERGIES in Paris, France.
  • 03/2025, I have participated in an inspiring spring school on "Learning Over Time" at Certosa di Pontignano, Italy.
  • 06/2024, I have attended an amazing summer school on Generative Modelling "GeMMS" at TU Eindhoven.
  • 11/2023, "Continual Learning with Dynamic Sparse Training: Exploring Algorithms for Effective Model Updates" has been accepted for an oral presentation at CPAL!
  • 09/2023, "AdaCL: Adaptive Continual Learning" has been accepted for an oral presentation at the CLAI!
  • 04/2023, "Condition Monitoring and Predictive Maintenance in Railways" is granted to international patent.
  • 10/2022, "Automated Machine Learning Approach in Material Discovery of Hole Selective Layers for Perovskite Solar Cells" has been published in the Energy Technology.
  • 11/2021, "Predicting Perovskite Bandgap and Solar Cell Performance with Machine Learning" has been published in the Solar RRL.
  • 05/2021, "A Machine Learning Approach for Metal Oxide Based Polymer Composites as Charge Selective Layers in Perovskite Solar Cells" published in the ChemPlusChem.
  • 08/2020, "Comparison of Machine Learning Models on Performance of Single- and Dual-Type Electrochromic Devices" has been published in the ACS Omega.
See more

Research

Publication Image 8

Unlocking [CLS] Features for Continual Post-Training.
Murat Onur Yildirim, Elif Ceren Gok Yildirim, Joaquin Vanschoren.
TMLR, 2026 - NeurIPS, 2025 CCFM Workshop.
Unlocks the frozen [CLS] token in vision transformers to drive continual post-training — matching fine-tuned accuracy with far fewer learnable parameters.
paper  -  github  -  blog

Publication Image 8

Pruned Adaptation Modules: A Simple yet Strong Baseline for Continual Foundation Models.
Elif Ceren Gok Yildirim, Murat Onur Yildirim, Joaquin Vanschoren.
CPAL, 2026 - NeurIPS, 2025 CCFM Workshop.
Prunes adapter modules to a sparse subset after each task — yielding a memory-efficient baseline that is surprisingly hard to beat in continual foundation model settings.
paper  -  github  -  blog

Publication Image 7

Self-Regulated Neurogenesis for Online Data-Incremental Learning.
Murat Onur Yildirim, Elif Ceren Gok Yildirim, Decebal Constantin Mocanu, Joaquin Vanschoren.
CoLLAs, 2025 (oral).
Dynamically grows new neurons when needed and prunes redundant ones, letting a network self-regulate its own capacity as new data streams arrive.
paper  -  github  -  blog

Publication Image 6

AdaCL: Adaptive Continual Learning.
Elif Ceren Gok Yildirim, Murat Onur Yildirim, Mert Kilickaya, Joaquin Vanschoren.
ContinualAI, 2024 (oral).
Automatically adjusts the stability–plasticity balance across tasks, removing the need to manually tune continual learning hyperparameters for each new dataset.
paper  -  github  -  blog

Publication Image 5

Continual Learning with Dynamic Sparse Training: Exploring Algorithms for Effective Model Updates.
Murat Onur Yildirim, Elif Ceren Gok Yildirim, Ghada Sokar, Decebal Constantin Mocanu, Joaquin Vanschoren.
CPAL, 2024 (oral).
Benchmarks dynamic sparse training algorithms for continual learning, identifying which sparsity-based update strategies best retain old knowledge while absorbing new tasks.
paper  -  github

Publication Image 4

Automated Machine Learning in Material Discovery of Hole Selective Layers for Perovskite Solar Cells.
Murat Onur Yildirim, Elif Ceren Gok Yildirim, Esin Eren, Peng Huang, Muhammed Haris, Samrana Kazim, Joaquin Vanschoren, Aysegul Uygun Oksuz, Shahzada Ahmad.
Energy Technology, 2023.
Automates the search for high-performing hole-selective layer materials in perovskite solar cells, letting ML rank candidate formulations and cut experimental trial-and-error.
paper

Publication Image 3

Predicting Perovskite Bandgap and Solar Cell Performance with Machine Learning.
Elif Ceren Gok, Murat Onur Yildirim, Muhammed Haris, Esin Eren, Meenakshi Pegu, Naveen Harindu Hemasiri, Peng Huang, Samrana Kazim, Aysegul Uygun Oksuz, Shahzada Ahmad.
Solar RRL, 2022.
Predicts perovskite bandgap and power conversion efficiency from composition data alone, accelerating candidate screening before lab synthesis.
paper

Publication Image 2

A Machine Learning Approach for Metal Oxide Based Polymer Composites as Charge Selective Layers in Perovskite Solar Cells.
Murat Onur Yildirim, Elif Ceren Gok, Naveen Harindu Hemasiri, Esin Eren, Samrana Kazim, Aysegul Uygun Oksuz, Shahzada Ahmad.
ChemPlusChem, 2021.
Uses ML to identify high-performing metal oxide-polymer blends for charge-selective layers, guiding synthesis toward more efficient perovskite cells.
paper

Publication Image 1

Comparison of Machine Learning Models on Performance of Single-and Dual-Type Electrochromic Devices.
Elif Ceren Gok, Murat Onur Yildirim, Esin Eren, Aysegul Uygun Oksuz.
ACS Omega, 2020.
Benchmarks ML models on predicting coloration efficiency and switching speed of electrochromic devices, establishing best-performing algorithms for device design.
paper

See more

Supervision

If you are excited about exploring machine learning through the lens of efficient adaptation and post-training strategies, I would love to brainstorm ideas and support your journey in a collaborative, low-pressure environment. I have been lucky to work with such talented and curious people:

Student Image 4

Danushkumar Venkadesh
Clover: Revisit-Aware Benchmark Suite for Continual Post-Training — 2026.
paper

Student Image 3

Shreya Sajid
Muting Tokens: Exploring Token Sparsification for Efficient Audio Transformers — co-supervised by NXP, 2025.
paper

Student Image 2

Athira Kulampurath
Reliability and Uncertainty in Foundation Model-Based Class-Incremental Learning — 2025.
paper

Student Image 1

Coen Schouten
On-the-Fly Curriculum Design for Online Class-Incremental Learning — cum laude, 2024.
paper