Keywords
Quantum, Deepfake, Machine Learning

Technologies
Amazon Braket

This project investigates whether quantum machine learning can contribute to deepfake detection. Deepfakes are synthetic media generated using deep learning techniques and are becoming increasingly realistic, raising concerns around misinformation, fraud, and digital trust. The objective of this project is to evaluate whether a quantum model could be integrated into a modern deep learning pipeline to increase performance.

A hybrid quantum–classical approach was developed in which a pretrained ResNet18 model extracts features from face images. These features are compressed to a low-dimensional space and encoded into a 4-qubit variational quantum circuit, which performs the final classification between real and fake images. The quantum model leverages an exponentially large state space and quantum feature interactions to represent complex relationships between image features.

Results show that the model achieves strong performance on a held-out dataset, reaching 95% accuracy. The findings show that near-term quantum models can be successfully applied to realistic machine learning tasks and highlight the potential of hybrid architectures as quantum hardware continues to develop.