Portfolio
A curated showcase of my research — papers where I made a primary contribution and key collaborations at top venues.
Featured Work
Distributional Active Inference
A unified framework bridging distributional reinforcement learning and active inference, enabling sample-efficient control without modeling transition dynamics. Published at ICML 2026.
MOMBO: Deterministic Uncertainty Propagation for Offline RL
Moment matching replaces Monte Carlo sampling in pessimistic offline RL, cutting suboptimality and accelerating convergence across D4RL benchmarks. Published at NeurIPS 2024.
EPPO: Evidential Proximal Policy Optimization
Evidential uncertainty quantification in the critic network preserves plasticity and enables directed exploration, outperforming state-of-the-art on-policy methods in non-stationary environments. Published in TMLR 2025.
CDDP: Continual Learning of Multi-modal Dynamics
A neural episodic memory with a Dirichlet Process prior enables a dynamics model to continually learn new behavioral modes without forgetting old ones. Published at L4DC 2024.
Evidential Turing Processes
A unified Bayesian framework combining global and local uncertainty through an external memory mechanism, achieving simultaneous model calibration, class overlap quantification, and out-of-distribution detection on five real-world classification benchmarks. Published at ICLR 2022.
iS-QL: Bridging Target-free and Target-based Reinforcement Learning
Parameter sharing between online and target networks (keeping only the final linear layer separate) closes the stability gap of target-free RL while halving memory, with gains across Atari, DMC, and language. Published at ICLR 2026.
PAC4SAC: PAC-Bayesian Soft Actor-Critic Learning
The first actor-critic algorithm to use a PAC-Bayesian generalization bound as its critic training objective. A single randomized critic paired with critic-guided multiple shooting delivers consistent sample efficiency and regret improvements over SAC. Published at AABI 2024.
ObjectRL: An Object-Oriented Reinforcement Learning Codebase
An open-source deep RL research codebase built on object-oriented principles: encapsulation, inheritance, and polymorphism mirror the natural structure of RL algorithms, enabling rapid prototyping of new ideas with minimal code changes.
BFL: Aggregating Variational Bayesian Networks in Federated Learning
An empirical survey of five statistical aggregation rules for Variational Bayesian Neural Networks in federated learning, revealing that the variance (spread) of the aggregated distribution is the dominant factor in federated performance. Published at NeurIPS 2022.
Thesis
Probabilistic Methods for Sample-Efficient Reinforcement Learning
Doctoral thesis presenting six peer-reviewed algorithms at NeurIPS, ICML, ICLR, TMLR, and UAI, unified by one claim: probabilistic uncertainty representations make reinforcement learning agents faster, more adaptive, and more data-efficient.
Memory-based Approaches to Problems in Probabilistic Modeling
Master's thesis at Istanbul Technical University demonstrating that external memory solves two open problems in probabilistic ML: total calibration of neural networks (ETP, ICLR 2022) and continual learning of multi-modal dynamical systems (CDDP, L4DC 2024).