Journals
Locally Adaptive One‑Class Classifier Fusion with Dynamic ℓp‑Norm Constraints for Robust Anomaly Detection
2025
Sepehr Nourmohammadi, Arda Sarp Yenicesu, Shervin R. Arashloo, Ozgur S. Oguz
Pattern Recognition
Introduces an interior‑point optimisation scheme for locally adaptive ensemble fusion, delivering significant speed‑ups and accuracy on benchmark anomaly‑detection sets.
H‑MaP: An Iterative and Hybrid Sequential Manipulation Planner
2025
Berk Cicek*, Arda Sarp Yenicesu*, Cankut Bora Tuncer*, Kutay Demiray, Ozgur S. Oguz*Equal contribution
IEEE Robotics and Automation Letters
Combines sampling and optimisation‑based planning to tackle sequential manipulation tasks involving tool use and bimanual coordination.
Interpretable Responsibility Sharing as a Heuristic for Task and Motion Planning
2024
Arda Sarp Yenicesu*, Sepehr Nourmohammadi*, Berk Cicek, Ozgur S. Oguz*Equal contribution
Robotics & Autonomous Systems — under review
Proposes the IRS heuristic that leverages human biases and auxiliary objects to make task‑and‑motion planning both more efficient and interpretable.
Workshops
Cross‑Lingual Transfer Learning for Misinformation Detection
2023
Oguzhan Ozcelik*, Arda Sarp Yenicesu*, Onur Yildirim*, Dilruba S. Haliloglu*, Erdem E. Eroglu*, Fazli Can*Equal contribution
International Workshop on Disinformation and Toxic Content Analysis
Studies transformer‑based cross‑lingual misinformation detection across English, Arabic, Chinese, Turkish and Polish, highlighting language‑pair asymmetries.
Preprints
CUER: Corrected Uniform Experience Replay for Off‑Policy Continuous Deep RL Algorithms
2024
Arda Sarp Yenicesu*, Furkan B. Mutlu*, Suleyman S. Kozat, Ozgur S. Oguz*Equal contribution
arXiv preprint
Mitigates sampling bias in replay buffers, yielding faster and stabler learning across MuJoCo benchmarks.
FViT‑Grasp: Grasping Objects with Fast Vision Transformers
2023
Arda Sarp Yenicesu*, Berk Cicek*, Ozgur S. Oguz*Equal contribution
arXiv preprint
Achieves 150 FPS grasp‑pose estimation using a lightweight transformer backbone while maintaining 88 % accuracy on the Cornell Grasp dataset.