PhD Researcher · Rensselaer Polytechnic Institute

Meltem Tatli
Machine Learning Researcher

Ph.D. student working at the intersection of wireless communications and machine learning, focused on risk-aware and preference-centric bandit algorithms for reliable decision-making.

Experience ↓ Publications
Professional Snapshot
Research + Industry ML Experience
From theoretical guarantees in risk-sensitive bandits to simulation and neural-network implementations for wireless and speech applications.
3
Research roles
4
Papers
2
Major awards
Professional Experience
Research and industry impact
Experience spanning theory-driven ML research, wireless communication systems, and industry neural-network deployment.
2023 — Present
Research Assistant
Rensselaer Polytechnic Institute · Troy, NY
Preference-Centric Bandits for Wireless Communications
  • Applied preference-centric bandit algorithms to optimize rate selection under fading, improving epsilon-outage capacity over standard baselines.
  • Integrated theoretical results into simulation-based physical layer evaluations, enabling faster design exploration for wireless systems.
  • Showed mixture-based scheduling can outperform deterministic strategies in 5G-like discrete-rate settings.
Risk-Sensitive Multi-Armed Bandits
  • Designed bandit algorithms that optimize user-specified risk functionals including CVaR, Dual Power, and PHT.
  • Proved finite-time regret guarantees using probabilistic concentration tools.
  • Built and maintained a reproducible Python/NumPy simulation pipeline to validate theory across synthetic bandit instances.
Bandit Theory Wireless Systems Risk Functionals Python/NumPy
2022 — 2023
Undergraduate Researcher
Bilkent University Archlab · Ankara, Turkey
  • Worked on classification tasks using regression models and feature-analysis tools such as PCA.
  • Implemented custom neural-network architectures in Python with Keras.
Keras PCA Classification
Summer 2022
Machine Learning Intern
Fraunhofer IIS · Erlangen, Germany
  • Tuned hyperparameters for a neural network used in MPEG-H speech-background separation.
  • Added FCNN components and implemented multi-domain, multi-resolution loss functions in PyTorch.
PyTorch MPEG-H Speech ML
Publications
Selected works
Conference papers, journal submissions, and workshop presentations in machine learning theory and wireless systems.
2025 AISTATS
Risk-sensitive bandits: Arm-mixture optimality and regret-efficient algorithms
Meltem Tatli, Arpan Mukherjee, Prashanth L.A., Karthikeyan Shanmugam, Ali Tajer
2025 NeurIPS
Preference-centric bandits in wireless communications: Theory and applications
Meltem Tatli, Ali Tajer
★ Oral Presentation — AI4NextG Workshop
2025 Submitted
Preference-centric Bandits: Optimality of Mixtures and Regret-efficient Algorithms
Meltem Tatli, Arpan Mukherjee, Prashanth L.A., Karthikeyan Shanmugam, Ali Tajer
IEEE Transactions on Information Theory
2026 Submitted
Tail-aware bandit learning for communication
Meltem Tatli, Ali Tajer
Under review
Research Areas
What I study
β
Risk-Sensitive Bandits
Bandit algorithms that optimize user-specified risk functionals such as CVaR, Dual Power, and PHT — going beyond expected reward to capture tail behavior.
CVaR Regret bounds Concentration Mixtures
π
Preference-Centric Learning
Decision-making frameworks where the learner's objective is shaped by user preferences, with mixture-based strategies that outperform deterministic approaches.
Preferences Mixture policies Outage capacity
λ
Wireless Communications
Applying bandit theory to rate selection under fading, enabling fast physical-layer design exploration for 5G-like discrete-rate systems.
Fading channels 5G Rate selection Scheduling
Selected Projects
Applied work
Industry internships, course projects, and applied research spanning audio ML, computer vision, robotics, and energy optimization.
Industry · Arçelik 2023
Smart Fridge Energy Optimization
Led ML sub-team building LSTM temperature forecasting coupled with PID control, deployed on microcontroller hardware for a leading appliance company.
Demonstrated measurable energy savings on real hardware
TensorFlow LSTM PID Embedded
Industry · Fraunhofer IIS 2022
MPEG-H Speech Separation
Tuned neural network hyperparameters for audio codec, added FCNN components and implemented multi-domain loss functions in PyTorch.
Contributed to production audio codec pipeline
PyTorch FCNN MPEG-H
Course 2025
Transformer Image Captioning
Built Transformer-based captioning model from scratch in PyTorch, benchmarked on Multi30K.
BLEU 37 on Multi30K benchmark
PyTorch Transformers Vision + NLP
Course 2024
POMDP Robot Planning
Implemented MCTS, forward search with rollouts, and QMDP for robot decision-making under partial observability.
4 planning algorithms compared
NumPy MCTS QMDP
Curriculum Vitae Full academic CV with complete publication list, teaching, and service record.
Request CV
Education & Awards
Background
2023 — Present
Rensselaer Polytechnic Institute
Ph.D. in Electrical Engineering · Troy, NY
GPA: 4.00 / 4.00
2019 — 2023
Bilkent University
B.Sc. in Electrical & Electronics Engineering · Ankara, Turkey
GPA: 3.90 / 4.00 · Ranked 2nd / 152
2024
Founders Award of Excellence
Rensselaer Polytechnic Institute
Recognized for academic achievement, leadership, creativity, and contributions to the RPI community.
2024
Jerry Dziuba ECSE Graduate Student Service Award
Rensselaer Polytechnic Institute
Awarded for exceptional service and dedication to the graduate community.
Contact
Get in touch

I'm seeking a 2026 industry internship in machine learning, applied research, or ML engineering. I'd welcome the chance to connect.

meltemtatli27@gmail.com LinkedIn