Yushun Dong

Stop model extraction attacks on your GNNs without sacrificing accuracy: CITED uses a novel decision boundary signature to prove ownership at both the embedding and label levels.

Finally, a defense against model extraction comes with rigorous theoretical guarantees, using mutual information to certify DNN ownership.

Suppressing non-stationary frequencies in time series data yields surprisingly large gains in forecasting accuracy and computational efficiency.

A unified benchmark reveals the fragmented landscape of RAG security, highlighting vulnerabilities to knowledge-extraction attacks and paving the way for robust defense strategies.