Sanberk Serbest

About Me

I recently graduated (ranked 1st in the class) with an MSc in Electrical and Electronics Engineering from EPFL, specializing in Data Science and Systems. My research interests include robotics, generative models, reinforcement learning, optimization, and speech enhancement. I enjoy developing efficient algorithms that advance the capabilities of intelligent systems. I am currently a research intern at ETH Zürich (until April 30, 2026), focusing on robotics and reinforcement learning. Previously, I studied at Boğaziçi University with a strong focus on signal processing and computer engineering.

Education

• EPFL (Switzerland) — 5.84 / 6.0 - MSc Electrical and Electronics Engineering, 2023–2025
• Bogazici University (Turkey) — 3.98 / 4.0 — BSc Electrical and Electronics Engineering, 2017–2022
• Bogazici University (Turkey) — 3.72 / 4.0 — Minor in Computer Engineering, 2020–2022

Research Interests

Optimization algorithms, generative models, reinforcement learning, deep learning theory, imitation learning, speech enhancement, computer vision, theoretical aspects of generalization and convergence of learning algorithms.

Publications

• Serbest, S., et al. “DeepFilterGAN: A Full-band Real-time Speech Enhancement System with GAN-based Stochastic Regeneration.” accepted to Interspeech 2025. [arxiv]
• Serbest, S., et al. “Improving Experience Replay Architecture with K-Means Clustering.” SIU 2023. [DOI]
• Serbest, S., et al. "Autonomous Air Combat with Reinforcement Learning under Different Noise Conditions." SIU 2023. [DOI]
• Serbest, S., et al. “A Biodegradable Implant Antenna Detecting Post-Surgical Infection.” EuCAP 2020. [DOI]

Projects

• Master Thesis: Beyond AdaGrad: Adaptive Methods for Non-Euclidean Optimization (LIONS, EPFL, 2025)

  Investigating norm constrained linear minimization oracles (LMOs) and developing an optimizer adaptive to the non-euclidean geometry of the loss landscape, with convergence guarantees for nonconvex optimization, specifically targeting deep learning trainings.

• MSc Semester Project: Exploration in deep imitation learning (LIONS, EPFL, 2024)

  Extending an exploration bonus computation utilized in the linear function approximation to the nonlinear function approximation setting, more specifically neural networks. As the baseline method, a Soft Actor-Critic based imitation learning algorithm is used with the addition of exploration bonus.

• BSc Senior Project: A Phase Unwrapping Algorithm for Magnetic Resonance Imaging (MRI) Applications (Bogazici University, 2022)

  Design of a segmentation based path-following algorithm for phase unwrapping of MR images as the first step of Quantitative Susceptibility Mapping (QSM), funded by The Scientific and Technological Research Council of Turkey, TUBITAK.

• Course Project: Dataset Based Imitation Learning (Reinforcement Learning Course, EPFL, 2024)

  Theoretical comparison of three dataset-based imitation learning algorithms: Online Apprenticeship Learning (OAL), Model-based Transition-aware Adversarial Imitation Learning (MB-TAIL), and Imitation Learning via Adversarial Reinforcement Learning (ILARL).
  References: "Lior Shani, Tom Zahavy, and Shie Mannor. Online apprenticeship learning, 2021", "Tian Xu, Ziniu Li, Yang Yu, and Zhi-Quan Luo. Provably efficient adversarial imitation learning with unknown transitions, 2023", "Luca Viano, Stratis Skoulakis, and Volkan Cevher. Better imitation learning in discounted linear MDP, 2024.

• Course Project: Reproducibility Challenge: Multi-Task Learning using Uncertainty (Fundamentals of Inference and Learning, EPFL, 2024)

  Reproduction and analysis of "Multi-task learning using uncertainty to weigh losses for scene geometry and semantics" (CVPR 2018). Theoretical and implementational aspects examined in light of experimental results.
  Reference: "Kendall, Alex, Yarin Gal, and Roberto Cipolla. "Multi-task learning using uncertainty to weigh losses for scene geometry and semantics." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018."

• Course Project: Pose Memory Game for Android Devices (Android Development Course, EPFL, 2024)

  Development of an Android game using pose estimation and smartwatch integration. Users mimic yoga poses within a time constraint; scoring is based on pose accuracy and response time. Included real-time database management.

• Course Project: Object Tracking on RGB-D Video Frames (Machine Vision Course, Bogazici University, 2022)

  Detected horizontally flat surfaces were filtered out, and the remaining objects tracked across frames using PCL and OpenCV in C++. Focused on RGB-D data analysis and 3D object segmentation.

Teaching Experience

• EE‑568 Reinforcement Learning — EPFL, Spring 2025

Awards & Grants

• Anna Barbara Reinhard Prize, The highest ranking graduate in Electrical and Electronic Engineering, EPFL (2025)
• Grant for Senior Project, The Scientific and Tech. Research Council of Turkey, TUBITAK (2022)
• High Honors, Electrical and Electronics Engineering, Bogazici University, Istanbul (2022)
• 2^nd Place Graduate, Electrical and Electronics Engineering, Bogazici University, Istanbul (2022)
• 3^rd Place Graduate, Faculty of Engineering, Bogazici University, Istanbul (2022)
• Merit‑based Scholarship, Funded by Turkish Government for taking the 61st place in the matriculation exam (2018-2022)
• 1^st Place Graduate, Cagaloglu Anadolu Lisesi (High School), Istanbul (2017)

Contact

Email: sanberkserbest@gmail.com

LinkedIn: sanberk-serbest