MSc student in Computer Science & Engineering at Sabancı University and researcher at TÜBİTAK BİLGEM. I work on trustworthy and efficient machine learning: Byzantine-robust and private federated learning, backdoors in graph neural networks, adversarial robustness and distillation, and low-bit quantization of multilingual language models.
| Project | Focus |
|---|---|
| structure-aware-fedala | Personalised federated graph learning with topology-weighted aggregation and degree-calibrated differential privacy |
| tr-quantbench | How calibration language shapes 4-bit GPTQ/AWQ quantization of multilingual LLMs, with a Turkish focus |
| gnn-backdoor-attacks | GNNs for graph, node and link tasks with Graph Trojaning backdoors on graph classification |
| cnn-robustness-cifar10 | Corruption and adversarial robustness of transfer-learned and distilled CNNs |
| transformer-feedback-channel-coding | Transformer channel codes for the AWGN channel with feedback, learned end to end |
| aspect-based-recommender | Recommendation from reviews with instruction-tuned aspect-based sentiment analysis |
More: adversarial-ml-mnist · columnist-persona-llm · mlp-ablation-suite
