Institution profile

Bogazici University

Academic institutioneurope · tr
Official website
Research library71linked papers
Opportunities0open roles
Selected work

Representative Papers

Online Monitoring of Metric Temporal Logic using Sequential Networks

Jan 01, 2019arXiv.org

This paper addresses efficiency and scalability bottlenecks in online monitoring of Metric Temporal Logic (MTL) for cyber-physical systems. Methodologically, it introduces a unified monitoring framework supporting both discrete- and dense-time behaviors. Its core contribution is the first “point-free” dense-time modeling approach for MTL, circumventing theoretical limitations inherent in traditional point-based semantics and thereby enhancing monitor composability and engineering practicality. The framework integrates sequence-network-based modeling, syntactic parsing of the past-time fragment of MTL, a unified semantic construction for discrete and dense time, and synergistic optimization via term rewriting and automata-theoretic techniques. Experimental evaluation demonstrates that the prototype monitor outperforms state-of-the-art tools in both runtime speed and memory footprint, achieving real-time deployment capability.

16 citationsRead paper

Affirmative Action in India: Restricted Strategy Space, Complex Constraints, and Direct Mechanism Design

Oct 04, 2023Social Science Research Network

This paper addresses resource allocation imbalances under India’s multi-layered reservation system—combining vertical (social group-based) and horizontal (cross-cutting criteria, e.g., gender, disability) quotas—amid practical constraints including quota rollback conflicts, restricted preference expression, and deeply nested priority structures. We introduce the Generalized Lexicographic (GL) family of selection rules, the first formal framework unifying legally mandated hierarchical priorities across reservation layers. Integrating a deferred-acceptance algorithm with a law-mechanism co-design architecture, we propose a direct matching mechanism that is constitutionally compliant, strategy-proof, and fair. It guarantees full utilization of reserved positions and significantly improves substantive representation of disadvantaged subgroups—including women and persons with disabilities—in education and public employment. Our mechanism offers a scalable, legally grounded paradigm for multidimensional affirmative action policy design.

4 citationsRead paper

Generative language models on nucleotide sequences of human genes

Jul 20, 2023Scientific Reports

Generative modeling of human genomic DNA sequences remains underexplored. Method: This work pioneers systematic autoregressive language modeling for biological sequences, treating DNA as a one-dimensional quaternary string (A/C/G/T) and employing RNNs and N-gram models—augmented with genomics-specific preprocessing and evaluation on real biological tasks (e.g., promoter identification, splice site prediction)—to circumvent the computational intractability of whole-genome modeling. Contribution/Results: On small-scale genomic datasets, RNNs significantly outperform Transformers, challenging the universality of the “large-model + large-data” paradigm in minimal-alphabet settings. DNA sequences admit effective formal-language modeling, yet their quaternary nature does not reduce data requirements. We establish RNNs as the current optimal architecture for few-shot generative genomics tasks and introduce a new paradigm for generative bioinformatics.

3 citationsRead paper

Unsupervised Meta-Testing With Conditional Neural Processes for Hybrid Meta-Reinforcement Learning

Oct 01, 2024IEEE Robotics and Automation Letters

To address low sample efficiency in meta-testing due to the absence of reward signals, this paper proposes UMCNP, an unsupervised meta-reinforcement learning framework. UMCNP integrates policy gradient optimization with task inference, enabling implicit environment dynamics modeling and adaptive policy optimization from a single trajectory of an unseen task. Its key contributions are threefold: (1) decoupling policy learning from task inference to support offline reuse of meta-training data; (2) employing Conditional Neural Processes (CNPs) for unsupervised task representation learning; and (3) combining parameterized policy gradients with a model-predictive control–inspired self-generated rollout mechanism. Evaluated on benchmarks—including 2D point navigation, biased-sensor CartPole, and dynamics-randomized Walker—UMCNP reduces meta-test sample requirements by over 50% while significantly improving few-shot adaptation performance.

1 citationsRead paper
Recent publications

Latest Papers

StARS: Socially Appropriate Robot Actions via a Recommender System-Driven Approach

Jul 23, 2026

This work addresses the challenge that socially appropriate robot behaviors in human–robot interaction vary across individuals by proposing the StARS framework, which introduces recommender system principles to this domain for the first time. Treating annotators as users, interaction contexts as items, and appropriateness ratings of robot actions as preference signals, StARS combines collaborative filtering with learnable context representations to model personalized social norms. The framework is model-agnostic, compatible with diverse backbone architectures and context encoding strategies, and robust to sparse feedback. Evaluated on the MannersDB+ and SocNav1 datasets, StARS significantly improves both action recommendation performance and alignment with individual user ratings, effectively enabling the generation of robot behaviors tailored to personal social expectations.

0 citationsRead paper

TreeThink: A Modular Tree Search Library for Mathematical Reasoning with LLMs

Jul 13, 2026

This work addresses the lack of native support for formal verifiers in existing large language model (LLM) tree search libraries and the limited generality of theorem-proving systems, which often rely on task-specific search strategies that hinder unified, efficient cross-lingual mathematical reasoning. To bridge this gap, the authors propose a modular, fully asynchronous open-source tree search framework that, for the first time, natively integrates interactive theorem proving in Lean4, Coq (Rocq), and Isabelle/HOL. The framework unifies natural and formal language reasoning by coupling vLLM inference pipelines, lightweight heuristics, and neural evaluators through REPL interfaces. Experiments on the miniF2F and MATH500 benchmarks demonstrate that the approach effectively enables cross-lingual formal proof synthesis and achieves up to a 6.3× end-to-end speedup over prior methods.

0 citationsRead paper