Atay Ozgovde
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Atay Ozgovde has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Cem Ersoy
Department of Computer Engineering Bogazici University Istanbul Turkey
22 shared publications - Ozgun Pinarer
Galatasaray University
7 shared publications Ahmet Cihat Baktir
Computer Engineering, Bogaziçi University, Istanbul, Turkey
4 shared publications- Fatma Başak Aydemir
Utrecht University, Utrecht, Netherlands
4 shared publications Ahmet Cihat Baktır
3 shared publications- Cagatay Sonmez
Computer Engineering, Bogazici Universitesi, Istanbul, Istanbul, Turkey
3 shared publications
Latest Publications
- 12 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Games
- Vol. PP (24751502) , 1-14
UBCL: A Reinforcement Learning Framework for Controllable and Diverse Player Behaviors
This paper introduces a reinforcement learning framework that enables controllable and diverse player behaviors without relying on human gameplay data. Existing approaches often require large-scale player trajectories, train separate models for different player types, or provide no direct mapping between interpretable behavioral parameters and the learned policy, limiting their scalability and controllability. We define player behavior in an N-dimensional continuous space and uniformly sample target behavior vectors from a region that encompasses the subset representing real human styles. During training, each agent receives both its current and target behavior vectors as input, and the reward is based on the normalized reduction in distance between them. This allows the policy to learn how actions influence behavioral statistics, enabling smooth control over attributes such as aggressiveness, mobility, and cooperativeness. A single PPO-based multi-agent policy can reproduce new or unseen play styles without retraining. Experiments conducted in a custom multi-player Unity game show that the proposed framework produces significantly greater behavioral diversity than a win-only baseline and matches specified behavior vectors across diverse targets, with highest accuracy on score-based dimensions and moderate degradation on cooperation and mobility. The method offers a scalable solution for automated playtesting, game balancing, human-like behavior simulation, and replacing disconnected players in online games.
- 12 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Games
- Vol. PP (24751502) , 1-14
UBCL: A Reinforcement Learning Framework for Controllable and Diverse Player Behaviors
This paper introduces a reinforcement learning framework that enables controllable and diverse player behaviors without relying on human gameplay data. Existing approaches often require large-scale player trajectories, train separate models for different player types, or provide no direct mapping between interpretable behavioral parameters and the learned policy, limiting their scalability and controllability. We define player behavior in an N-dimensional continuous space and uniformly sample target behavior vectors from a region that encompasses the subset representing real human styles. During training, each agent receives both its current and target behavior vectors as input, and the reward is based on the normalized reduction in distance between them. This allows the policy to learn how actions influence behavioral statistics, enabling smooth control over attributes such as aggressiveness, mobility, and cooperativeness. A single PPO-based multi-agent policy can reproduce new or unseen play styles without retraining. Experiments conducted in a custom multi-player Unity game show that the proposed framework produces significantly greater behavioral diversity than a win-only baseline and matches specified behavior vectors across diverse targets, with highest accuracy on score-based dimensions and moderate degradation on cooperation and mobility. The method offers a scalable solution for automated playtesting, game balancing, human-like behavior simulation, and replacing disconnected players in online games.
- Open Access
Instruction-Tuning Open-Weight Language Models for BPMN Model Generation
Domain models are central to software engineering, as they enable a shared understanding, guide implementation, and support automated analyses and model-driven development. Yet, despite these benefits, practitioners often skip modeling because it is time-consuming and demands scarce expertise. We address this barrier by investigating whether open-weight large language models, adapted via instruction tuning, can generate high-quality BPMN process models directly from natural language descriptions in a cost-effective and privacy-preserving way. We introduce InstruBPM, a reproducible approach that prepares paired text-diagram data and instruction tunes an open source large language model with parameter-efficient fine-tuning and quantization for on-prem deployment. We evaluate the tuned model through complementary perspectives: (i) text/code similarity using BLEU, ROUGE-L, and METEOR, (ii) structural fidelity using Relative Graph Edit Distance, (iii) guidelines conformance using external tool checks, and (iv) a small expert review. Using a curated subset of a multi-domain BPMN dataset, we compare the tuned model with untuned open-weight baselines and strong proprietary models under consistent prompting regimes. Our compact tuned model outperforms all baselines across sequence and structural metrics while requiring substantially fewer resources; guideline analysis and expert feedback further indicate that the generated diagrams largely follow BPMN best practices and are useful starting points that reduce modeling effort. Overall, instruction tuning improves structural accuracy and robustness compared to untuned baselines and reduces reliance on heavy prompt scaffolding. We publicly share the trained models and scripts to support reproducibility and further research.
Open AccessInstruction-Tuning Open-Weight Language Models for BPMN Model Generation
Domain models are central to software engineering, as they enable a shared understanding, guide implementation, and support automated analyses and model-driven development. Yet, despite these benefits, practitioners often skip modeling because it is time-consuming and demands scarce expertise. We address this barrier by investigating whether open-weight large language models, adapted via instruction tuning, can generate high-quality BPMN process models directly from natural language descriptions in a cost-effective and privacy-preserving way. We introduce InstruBPM, a reproducible approach that prepares paired text-diagram data and instruction tunes an open source large language model with parameter-efficient fine-tuning and quantization for on-prem deployment. We evaluate the tuned model through complementary perspectives: (i) text/code similarity using BLEU, ROUGE-L, and METEOR, (ii) structural fidelity using Relative Graph Edit Distance, (iii) guidelines conformance using external tool checks, and (iv) a small expert review. Using a curated subset of a multi-domain BPMN dataset, we compare the tuned model with untuned open-weight baselines and strong proprietary models under consistent prompting regimes. Our compact tuned model outperforms all baselines across sequence and structural metrics while requiring substantially fewer resources; guideline analysis and expert feedback further indicate that the generated diagrams largely follow BPMN best practices and are useful starting points that reduce modeling effort. Overall, instruction tuning improves structural accuracy and robustness compared to untuned baselines and reduces reliance on heavy prompt scaffolding. We publicly share the trained models and scripts to support reproducibility and further research.
- Open Access
Learning Controllable and Diverse Player Behaviors in Multi-Agent Environments
This paper introduces a reinforcement learning framework that enables controllable and diverse player behaviors without relying on human gameplay data. Existing approaches often require large-scale player trajectories, train separate models for different player types, or provide no direct mapping between interpretable behavioral parameters and the learned policy, limiting their scalability and controllability. We define player behavior in an N-dimensional continuous space and uniformly sample target behavior vectors from a region that encompasses the subset representing real human styles. During training, each agent receives both its current and target behavior vectors as input, and the reward is based on the normalized reduction in distance between them. This allows the policy to learn how actions influence behavioral statistics, enabling smooth control over attributes such as aggressiveness, mobility, and cooperativeness. A single PPO-based multi-agent policy can reproduce new or unseen play styles without retraining. Experiments conducted in a custom multi-player Unity game show that the proposed framework produces significantly greater behavioral diversity than a win-only baseline and reliably matches specified behavior vectors across diverse targets. The method offers a scalable solution for automated playtesting, game balancing, human-like behavior simulation, and replacing disconnected players in online games.
Open AccessLearning Controllable and Diverse Player Behaviors in Multi-Agent Environments
This paper introduces a reinforcement learning framework that enables controllable and diverse player behaviors without relying on human gameplay data. Existing approaches often require large-scale player trajectories, train separate models for different player types, or provide no direct mapping between interpretable behavioral parameters and the learned policy, limiting their scalability and controllability. We define player behavior in an N-dimensional continuous space and uniformly sample target behavior vectors from a region that encompasses the subset representing real human styles. During training, each agent receives both its current and target behavior vectors as input, and the reward is based on the normalized reduction in distance between them. This allows the policy to learn how actions influence behavioral statistics, enabling smooth control over attributes such as aggressiveness, mobility, and cooperativeness. A single PPO-based multi-agent policy can reproduce new or unseen play styles without retraining. Experiments conducted in a custom multi-player Unity game show that the proposed framework produces significantly greater behavioral diversity than a win-only baseline and reliably matches specified behavior vectors across diverse targets. The method offers a scalable solution for automated playtesting, game balancing, human-like behavior simulation, and replacing disconnected players in online games.
- 20 November 2025
- Published by Springer Nature
- p. 269-284
Generating Business Process Models with Open Source Large Language Models Using Instruction Tuning
- 20 November 2025
- Published by Springer Nature
- p. 269-284
Generating Business Process Models with Open Source Large Language Models Using Instruction Tuning
- 1 April 2025
- Published by Springer Nature
- p. 157-165
Generating Domain Models with LLMs Using Instruction Tuning: A Research Preview
- 1 April 2025
- Published by Springer Nature
- p. 157-165
Generating Domain Models with LLMs Using Instruction Tuning: A Research Preview
- 1 April 2025
- Published by Springer Nature
- p. 317-329
Requirements Elicitation Workshops Using the Six Thinking Hats Creativity Technique
- 1 April 2025
- Published by Springer Nature
- p. 317-329
Requirements Elicitation Workshops Using the Six Thinking Hats Creativity Technique
- Open Access
AirCompSim: A Discrete Event Simulator for Air Computing
Air components, including UAVs, planes, balloons, and satellites have been widely utilized since the fixed capacity of ground infrastructure cannot meet the dynamic load of the users. However, since those air components should be coordinated in order to achieve the desired quality of service, several next-generation paradigms have been defined including air computing. Nevertheless, even though many studies and open research issues exist for air computing, there are limited test environments that cannot satisfy the performance evaluation requirements of the dynamic environment. Therefore, in this study, we introduce our discrete event simulator, AirCompSim, which fulfills an air computing environment considering dynamically changing requirements, loads, and capacities through its modular structure. To show its capabilities, a dynamic capacity enhancement scenario is used for investigating the effect of the number of users, UAVs, and requirements of different application types on the average task success rate, service time, and server utilization. The results demonstrate that AirCompSim can be used for experiments in air computing.
Open AccessAirCompSim: A Discrete Event Simulator for Air Computing
Air components, including UAVs, planes, balloons, and satellites have been widely utilized since the fixed capacity of ground infrastructure cannot meet the dynamic load of the users. However, since those air components should be coordinated in order to achieve the desired quality of service, several next-generation paradigms have been defined including air computing. Nevertheless, even though many studies and open research issues exist for air computing, there are limited test environments that cannot satisfy the performance evaluation requirements of the dynamic environment. Therefore, in this study, we introduce our discrete event simulator, AirCompSim, which fulfills an air computing environment considering dynamically changing requirements, loads, and capacities through its modular structure. To show its capabilities, a dynamic capacity enhancement scenario is used for investigating the effect of the number of users, UAVs, and requirements of different application types on the average task success rate, service time, and server utilization. The results demonstrate that AirCompSim can be used for experiments in air computing.
- 8 August 2024
- Published by Elsevier in Future Generation Computer Systems
- Vol. 162, 107475
Network-aware federated neural architecture search
- 8 August 2024
- Published by Elsevier in Future Generation Computer Systems
- Vol. 162, 107475
Network-aware federated neural architecture search
- 17 July 2024
- Published by Elsevier in Computer Networks
- Vol. 251
Air computing: A survey on a new generation computation paradigm
- 17 July 2024
- Published by Elsevier in Computer Networks
- Vol. 251
Air computing: A survey on a new generation computation paradigm
- 2 March 2024
- Published by Elsevier in Future Generation Computer Systems
- Vol. 157, 516-528
End-to-end network slicing for edge computing optimization
- 2 March 2024
- Published by Elsevier in Future Generation Computer Systems
- Vol. 157, 516-528
End-to-end network slicing for edge computing optimization