Department News and Announcements

TÜBİTAK 1001 funding awarded for real-time mobile protection against voice phishing

The project proposal “Development of a Dialog Generator, Dataset, Method and Mobile Platform for the Real-Time Detection of Voice Phishing Attacks” , led by Asst. Prof. Dr. Ahmet Selman Bozkır of our department, has been awarded funding under the TÜBİTAK 1001 – Scientific and Technological Research Projects Support Program.

Voice phishing has become an increasingly significant threat with the spread of AI-assisted voice cloning and sophisticated social engineering techniques, often targeting elderly individuals and people with limited technological awareness. Over the course of 18 months, the project will create an original and balanced synthetic conversation dataset reflecting Turkish-language scenarios and diverse speaker profiles.

The project will also develop a lightweight deep learning method capable of assessing risk at each turn while a call is still in progress, rather than analysing the conversation only after it has ended. The method will take emotional and thematic cues into account, and a new evaluation metric will be proposed to measure how early a threat can be detected reliably.

One of the project outcomes will be a free Android application that performs all analyses directly on the mobile device. By avoiding the transmission of conversation data to a remote server, the application is designed to protect user privacy.

Assoc. Prof. Dr. Murat Aydos will serve as a researcher on the project, which will also support one MSc scholarship student.

Our students win the Turkcell Code Night Türkiye final

Artificial Intelligence Engineering students Muhammed Ekinci and Murat Dinç, who will continue their studies within the Department of Artificial Intelligence and Data Engineering, and Computer Engineering student İbrahim Yılmaz won first place in the Turkcell Code Night Türkiye final.

The team competed against participants from across Türkiye, including teams with graduates and master’s and doctoral students, and received the award from Turkcell CEO Dr. Ali Taha Koç. We congratulate our students on this achievement.

Our students' capstone project leads to a paper accepted at the GAIA Workshop at ECCV 2026

Building on capstone projects completed by our students, the paper “When Does the Second View Help? Auditing Cross-View Fusion in Urban Vision-Language Models” has been accepted to the GAIA Workshop, held in conjunction with ECCV 2026 in Malmö this September.

The study introduces CityViews, a benchmark spanning 12 task families across 40 cities in 32 countries. Each example pairs satellite imagery with four road-aligned street-level views, allowing the same question to be evaluated with satellite-only, street-view-only, and combined inputs. The analysis shows that the benefit of adding a second view is task-dependent: it can be limited when one view already contains sufficient evidence, but substantial when the answer depends on correspondence between the two viewpoints.

The paper was co-authored by Computer Engineering student Ezel Bayraktar; Artificial Intelligence Engineering students Alperen Demirci, Mukhamediyar Amanzhol and Adam Sattout; Aykut Erdem (Koç University) and Erkut Erdem.

GAIA Workshop at ECCV 2026 →

The Turkish radiology report generation study conducted as part of our students’ capstone project was presented at SİU 2026

The paper “Multimodal Large Language Models for Turkish Radiology Report Generation from 3D CT Images” was accepted and presented at SİU 2026, where the team discussed the study with members of the research community and received valuable feedback.

The study compares general-purpose and medically specialized multimodal large language models for generating Turkish radiology reports directly from three-dimensional CT volumes. It evaluates slice-based and volumetric representations under zero-shot and parameter-efficient fine-tuning settings. The results show that fine-tuning is particularly important for a low-resource clinical language such as Turkish and that, under the evaluated settings, slice-based representations outperform the volumetric approach.

The work was carried out by Artificial Intelligence Engineering students Emir Çağıl Kökdener, Ender Orman, Gülse Atun, and Abdullah Edik as part of their capstone project, under the supervision of Erkut Erdem and Aykut Erdem (Koç University).

Students present medical AI research at the CVPR 2026 Med-Reasoner Workshop

Alperen Demirci and Bora Dere, students in Hacettepe University’s Artificial Intelligence Engineering undergraduate program, together with Computer Engineering student Ezel Bayraktar, presented their joint research at the Medical Reasoning with Vision-Language Foundation Models (Med-Reasoner) Workshop, held in conjunction with CVPR 2026 in Denver, Colorado, USA.

Their paper, “When the Right Answer Is Missing: Probing Hallucination in Medical Reasoning Models via NOTA Evaluation”, investigates whether medical vision-language models can recognize when a correct answer is unavailable and how reliably they avoid hallucinating under such conditions.

The study was carried out under the supervision of Erkut Erdem and Aykut Erdem and appears in the CVPR 2026 Workshop Proceedings.