Publications & research
Academic work
I work on making deep learning models accurate, robust and efficient on real-world data. My computer vision papers are in medical imaging; my current focus is efficient inference for large language models.
- Machine Learning
- Efficient Machine Learning
- Efficient LLM
- Efficient Inference
Highlights
-
International collaboration
90 citations
My most-cited paper is the results paper of the international MIDOG 2022 challenge, published in Medical Image Analysis.
-
Contribution to the field
2 new datasets
The MITNET and Tubule-U-Net papers introduced new datasets alongside the models.
-
From research to production
LLM efficiency
My MSc thesis is on efficient LLM inference on GPUs; today I optimize GPU and token throughput at Vodafone.
Research areas
Each area lists the skills that work called for. Select an area to filter the related publications.
-
2025 Master's thesis
Studies how to run large language models with less memory and lower latency on GPUs by quantizing them after training.
Post-Training Quantization for Efficient Inference of Large Language Models On Modern GPU Architectures
MSc thesis, Bahçeşehir University (YÖK National Thesis Center)
No citations yet Open paper → -
2024 Journal
The results paper of an international challenge that measured how well mitosis detection models generalize across tumour types, labs and species.
Domain generalization across tumor types, laboratories, and species — insights from the 2022 edition of the Mitosis Domain Generalization Challenge
Medical Image Analysis, vol. 94
90 citations Open paper → -
2022 Journal
A two-stage deep learning approach that finds dividing cells in breast cancer tissue, together with a new dataset built for it.
MITNET: a novel dataset and a two-stage deep learning approach for mitosis recognition in whole slide images of breast cancer tissue
Neural Computing and Applications, vol. 34
37 citations Open paper → -
2023 Journal
A U-Net-based model that segments tubule structures in breast cancer slides, together with a new dataset built for it.
Tubule-U-Net: a novel dataset and deep learning-based tubule segmentation framework in whole slide images of breast cancer
Scientific Reports, vol. 13
39 citations Open paper → -
2023 Book chapter
CNN approaches that grade breast cancer across whole slides by working on small image patches.
Patch-based approaches to whole slide histologic grading of breast cancer using convolutional neural networks
In: Diagnostic Biomedical Signal and Image Processing Applications with Deep Learning
5 citations -
2021 Conference
Detecting cell nuclei in histopathology images with RetinaNet.
Nuclei Detection on Breast Cancer Histopathology Images Using RetinaNet
SIU 2021 — 29th IEEE Signal Processing and Communications Applications Conference
4 citations Open paper → -
2022 Abstract
Whole-slide breast cancer grading with CNNs.
Whole slide histologic grading of breast cancer using convolutional neural networks
Journal of Clinical Oncology, vol. 40, no. 16_suppl
2 citations -
2021 Conference
Scoring how much cell nuclei vary in shape (pleomorphism) at patch level with CNNs.
Patch-Level Nuclear Pleomorphism Scoring Using Convolutional Neural Networks
CAIP 2021 — 19th International Conference on Computer Analysis of Images and Patterns
1 citation Open paper → -
2024 Abstract
AI-based automated analysis of mitotic figures in soft-tissue sarcoma slides.
AI-Powered Automated Analysis of Mitotic Figures in Soft Tissue Sarcomas Whole Slide Images
Laboratory Investigation, vol. 104, no. 3
No citations yet