Detecting threats in generative AI: security, interpretability and ethical issues in deepfake content
Martini, Piergiorgio (A.A. 2024/2025) Detecting threats in generative AI: security, interpretability and ethical issues in deepfake content. Tesi di Laurea in Big data and smart data analytics, Luiss Guido Carli, relatore Irene Finocchi, pp. 76. [Master's Degree Thesis]
|
PDF (Full text)
Download (2MB) | Preview |
Abstract/Index
Overview of generative models. Definition of generative models. Historical evolution and development of generative models. Applications of generative models. Security and ethics in deepfake video generation. Deepfakes and disinformation. Manipulation of textual and visual content. Attacks on generative models. Reliability in generative models. Interpretability methods. Security and governance in generative models. Ethical issues and bias. Regulation and policy. Experimental context, dataset and evaluation criteria. Experimental objectives and context. Dataset structure and composition. Class distribution and imbalance. Evaluation metrics for deepfake detection models. The integrated use of metrics in model validation. Model engineering and architectural choices. Introduction to the EfficientNet family. Why EfficientNet-B7 was selected. EfficientNet-B7 in detail. Architectural comparison: alternative models to EfficientNet-B7. Computing environment: use of a Virtual machine (VM). Experimental Pipeline and model implementation. Preprocessing and dataset preparation. Construction and organization of the supervised dataset. Implementation of a custom DataLoader. Training strategy. Testing and inference. Experimental variations. Comparative experiments with alternative architectures. Empirical evaluation and results. Overview of the evaluation protocol. Key experimental variables. Performance on the 10 GB test set. Performance on the 50 GB test set. Comparison with alternative architectures. Efficiency and trade-off analysis. Robustness and repeatability. Comparative visualization of performance.
References
Bibliografia: pp. 74-75.
| Thesis Type: | Master's Degree Thesis |
|---|---|
| Institution: | Luiss Guido Carli |
| Degree Program: | Master's Degree Programs > Master's Degree Program in Data Science e Management (LM-91) |
| Chair: | Big data and smart data analytics |
| Thesis Supervisor: | Finocchi, Irene |
| Thesis Co-Supervisor: | Coppa, Emilio |
| Academic Year: | 2024/2025 |
| Session: | Summer |
| Deposited by: | Alessandro Perfetti |
| Date Deposited: | 16 Sep 2025 08:07 |
| Last Modified: | 16 Sep 2025 08:07 |
| URI: | https://tesi.luiss.it/id/eprint/43172 |
Downloads
Downloads per month over past year
Repository Staff Only
 |
View Item |
