NTR organizes and hosts scientific webinars on neural networks and invites speakers from all over the world to present their recent work.
On June 28 Pavel Korshunov, Idiap Research Institute, Martigny, Switzerland, presented a technical Zoom webinar on Improving Generalization of Deepfake Detection.
About the webinar:
Typical deepfake detection systems lack generalization capabilities on new datasets or unseen deepfakes. In this webinar, Pavel demonstrated this issue and proposed some generic approaches that aim to improve the generalization of the deepfake detection models.
Pavel Korshunov considered such approaches as increasing variability of the training data with aggressive “data farming” strategies, a few shot model tuning on new data, and training models for attribution instead of for binary classification.
Additional links:
- Pavel Korshunov and Sébastien Marcel, “Improving Generalization of Deepfake Detection with Data Farming and Few-Shot Learning”, IEEE Transactions on Biometrics, Behavior, and Identity Science, 2021. https://publications.idiap.ch/attachments/papers/2021/Korshunov_TBIOM_2021.pdf
- Anubhav Jain, Pavel Korshunov, and Sébastien Marcel, “Improving Generalization of Deepfake Detection by Training for Attribution”, MMSP 2021. http://publications.idiap.ch/downloads/papers/2021/Jain_MMSP_2021.pdf
- Pavel Korshunov, Anubhav Jain, and Sébastien Marcel, “Custom attribution loss for improving generalization and interpretability of deepfake detection”, ICASSP 2022. https://publications.idiap.ch/attachments/papers/2022/Korshunov_ICASSP_2022.pdf
Slide deck link https://disk.yandex.ru/i/UcFeC6tZBnHMMg
Moderator and contact: NTR CEO Nick Mikhailovsky: nickm@ntrlab.com.