1. passive defense 被动防御

不改变model,增加一个Anomaly Detection

1.1. 增加一个filter,例如smoothing

1.2. Feature Squeeze

1.3. Randomization at Inference Phase

1.4. 缺点

如果防御机制泄漏,攻击仍会生效

2. Proactive Defence 主动防御

Training a model that is robust to adversarial attack

2.1. 找出漏洞,补起来

根据图像攻击算法,找到攻击图像,把攻击图像当作训练样本来训练

2.2. 缺点

训练时只能列举有限的攻击算法,换用算法攻击,仍然能攻破

results matching ""

    No results matching ""