I am a PhD researcher at the Smart Embedded Systems Lab of OTH Regensburg, in a cooperative doctorate with the Technical University of Munich, supervised by Prof. Johannes Reschke and Prof. Björn W. Schuller. I build detectors for AI-generated text and images that hold up outside the lab: across datasets, generators and domains, under adversarial manipulation of their input, and with decisions that can be explained to the people who rely on them.
|
DeBERTa-ConPara Attack-aware, deployment-realistic detection of AI-generated text. Official RAID leaderboard: AUROC 99.61 %, TPR 96.57 % at 1 % FPR. Normalising Unicode at inference lifts homoglyph and zero-width-space attacks from 11.05 % and 1.12 % to 96.98 % TPR. |
Datasets Academic-Text-arxiv-gpt-gemini: 669,008 academic paragraphs, human (arXiv, before 2022) and AI-generated (GPT-3.5-Turbo, Gemini 2.0 Flash). HC3-Gemini-Flash-Responses: 23,463 Gemini 2.0 Flash answers to the HC3 questions, for measuring generator shift. Now: what do AI-text detectors actually look at? Faithfulness and ground-truth evaluation of their explanations. |
|
AI-Generated-Text-Detector Extracts 119+ linguistic and topological features (lexical diversity, syntax, readability, discourse, persistent-homology dimension) from text, to study and classify human versus AI-generated writing. GUI and command line, files up to 1 GB. |
PhD exposé The research plan behind this work: robust and interpretable detection of AI-generated text and images, with explainable AI for the people who rely on the decisions. |
- DeBERTa-ConPara: Attack-Aware and Deployment-Realistic Detection of AI-Generated Text. M. Mady, Y. Li, J. Reschke, B. W. Schuller. AACL-IJCNLP 2026. arXiv · code
- Beyond Accuracy: ARIA-Rubrics for Evaluating Audio Reasoning in Large Audio Language Models. Y. Li, Q. Sun, M. Mady, C. Wang, Z. Gong, B. Sisman, B. W. Schuller. Findings of AACL-IJCNLP 2026. arXiv
- AI-Generated Content Detection: A Cross-Modal Survey of Methods, Challenges, and Future Directions. M. Mady, Y. Li, B. W. Schuller, B. Sisman, J. Reschke. Preprint, 2026. Research Square
- Feature-Augmented Transformers for Robust AI-Text Detection Across Domains and Generators. M. Mady, J. Reschke, B. W. Schuller. arXiv, 2026. arXiv
- ficonTEC Service GmbH: production automation for photonics and semiconductor assembly (FAU-to-PIC alignment, laser soldering, wafer probers, SECS/GEM).
- BMW Group: prediction of road-induced vibration (95 %+ accuracy), about 30 % fewer physical prototype tests.
- Boehringer Ingelheim: automated real-world-evidence pipelines and R Shiny dashboards, 40 % faster clinical data processing.
- Ph.D. candidate, AI-generated content detection, TUM and OTH Regensburg
- M.Eng. AI for Smart Sensors and Actuators, Technische Hochschule Deggendorf, 2024
- B.Eng. Electronics and Communications, Mansoura University, 2022
Python PyTorch Hugging Face Transformers scikit-learn Integrated Gradients SHAP LIME Grad-CAM OpenCV Docker FastAPI R
mohamed.mady@tum.de · mohamed.mady@st.oth-regensburg.de · Mohamed.Mady@gmx.de · LinkedIn · Google Scholar · ORCID