Unlearnable Event Stream (UEVs)
收藏资源简介:
本文提出了一种名为UEVs的数据集,旨在防止未经授权的事件数据集使用。UEVs数据集通过引入一种新的异步事件误差最小化噪声,来保护事件数据集不受未经授权的使用。该噪声能够欺骗未经授权的模型,使其学习嵌入的噪声而不是真实特征。为了与稀疏事件兼容,提出了一种投影策略,将噪声稀疏化以生成UEVs。实验结果表明,UEVs方法有效地保护了事件数据,同时保留了其合法用途的实用性。
This paper proposes a dataset named UEVs, designed to prevent unauthorized usage of event datasets. The UEVs dataset protects event datasets from unauthorized utilization by introducing a novel asynchronous event error-minimizing noise. Such noise can deceive unauthorized models, compelling them to learn the embedded noise instead of the genuine underlying features. To achieve compatibility with sparse events, a projection strategy is proposed to sparsify the noise and generate UEVs. Experimental results show that the UEVs method effectively safeguards event data while preserving the utility for its legitimate applications.




