Adversarial Robustness

Motivation : a limitation of the (supervised) ML framework \scriptsize{\text{: a limitation of the (supervised) ML framework}} : a limitation of the (supervised) ML framework

The current way of training and inference is to train the ML model on some randomly drawn samples and test on some other independently drawn samples. A crucial assumption behind this is that the distribution that we use to train the ML model is exactly the distribution that the model will encounter. In reality, this is not the case as there are various forms of covariate shifts. Due to lack of this assumption, ML (particularly DL) predictions are brittle despite (mostly) accurate.

ML and adversarially robust ML

learning objective

In fact, lack of adversarial robustness is not at odds with what we currently want our ML models to achieve.

  • standard generalization (average case performance): E ( x , y ) ∼ D [ l o s s ( θ , x , y ) ] \mathbb{E}_{(x,y)\sim D} [loss(\theta,x,y)] E(x,y)D[loss(θ,x,y)]
  • adversarial robust generalization (worst-case notion, measure zero event): E ( x , y ) ∼ D [ max ⁡ δ ∈ Δ l o s s ( θ , x + δ , y ) ] \mathbb{E}_{(x,y)\sim D} [\max_{\delta \in \Delta} loss(\theta,x+\delta,y)] E(x,y)D[maxδΔloss(θ,x+δ,y)]
generalization of standard and robust deep networks

Not only does the optimization problem become more challenging and the model become larger(training), adversarial robustness has a significantly larger sample complexity, thus its generalization requires more data (test).

No ‘free lunch’: can exist a trade-off between accuracy and robustness. Intuitively,

  • in standard training, all correlation among features is a good correlation (so-called non-robust features in [3]);
  • if we want robustness, must avoid weakly correlated features.
interpretability

Adversarial Examples and Verification (inner minimization)

min ⁡ θ ∑ ( x , y ) ∼ S max ⁡ δ ∈ Δ L ( θ , x + δ , y ) ⏟ Part I. create AEs or verify it does not exist \min_\theta \sum_{(x,y)\sim S} \underbrace{\max_{\delta \in \Delta} L(\theta,x+\delta,y)}_\text{Part I. create AEs or verify it does not exist} θmin(x,y)SPart I. create AEs or verify it does not exist δΔmaxL(θ,x+δ,y)

In the linear case with a norm ball perturbation region, the maximization has exact solution based on dual norm; a simple instance of robust optimization. That is,
max ⁡ δ L ( θ T ( x + δ ) ⋅ y ) = L ( min ⁡ δ θ T ( x + δ ) ⋅ y ) = L ( θ T x ⋅ y − ∥ θ ∥ ∗ ) \begin{aligned} \max_\delta L(\theta^T (x + \delta) \cdot y) &=L(\min_\delta \theta^T (x + \delta) \cdot y)\\ &=L(\theta^T x \cdot y - \| \theta \|_\ast) \end{aligned} δmaxL(θT(x+δ)y)</

  • 0
    点赞
  • 2
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值