Oğuzhan Ersoy
Researcher
Studies verification and robustness in decentralised training, from pipeline-parallel attacks to collaborative GRPO.
Oğuzhan Ersoy’s research & writing
- ResearchIR3DE-AXL: Collective Inference Network
- ResearchF-TIS: Heterogeneous GRPO Without Homogeneous Assumptions
- ResearchBackdoor in the Middle: Attacking Pipeline Parallelism
- ResearchVerde Verification System In Production
- ResearchHail to the Thief: Exploring Attacks and Defenses in Decentralized GRPO
All Gensyn research
- ResearchThe Role of Feedback Alignment in Self-Distillation
- ResearchLook Beyond One-Size-Fits-All LLMs with IR3DE
- ResearchStop Multitask Training. Just DUME.
- ResearchSAPO, Efficient LM Post-Training with Collective RL
- ResearchCheckFree: fault tolerant training without checkpoints