Projects

A resource-adaptive federated fine-tuning method that fixes the expert-imbalance and gradient-sparsity discordances of sparse MoE, achieving up to 45% computation reduction and 8.7x better low-resource performance over heterogeneous LoRA-rank methods.

A semi-supervised approach that discovers latent architectural patterns within fake images to achieve cross-generator generalization in deepfake detection.

A comprehensive survey of proactive deepfake defense strategies including disruption and watermarking approaches across visual and audio modalities.

A learning-to-unlearn framework that incorporates transformations during federated learning to reduce memorization and simplify subsequent unlearning.

An online test-time adaptation method that improves deepfake detector adaptability during inference without requiring training data or labels.

A novel cancellable biometrics hashing scheme for voice-based speaker verification that records ranking of elements instead of maximum values.

A study of the resilience of the Deepfake CAPTCHA system under transferable imperceptible adversarial attacks, with a more robust defense using adversarial training.

A novel framework combining differential privacy with meta-learning to simultaneously address privacy leakage and non-IID data challenges in cross-silo federated learning.