Asynchronous Event Error-Minimizing Noise for Safeguarding Event Dataset

With more event datasets being released online, safe-guarding the event dataset against unauthorized usage hasbecome a serious concern for data owners. UnlearnableExamples are proposed to prevent the unauthorized ex-ploitation of image datasets. However, it’s unclear how tocreate unlearnable asynchronous event streams to preventevent misuse. In this work, we propose the first unlearn-able event stream generation method to prevent unautho-rized training from event datasets. A new form of asyn-chronous event error-minimizing noise is proposed to per-turb event streams, tricking the unauthorized model intolearning embedded noise instead of realistic features. Tobe compatible with the sparse event, a projection strategyis presented to sparsify the noise to render our unlearnableevent streams (UEvs). Extensive experiments demonstratethat our method effectively protects event data from unau-thorized exploitation, while preserving their utility for legit-imate use. We hope our UEvs contribute to the advance-ment of secure and trustworthy event dataset sharing. Codeis available at: https://github.com/rfww/uevs.