Gradient-guided Random Search:单次前反向传播即可匹敌 sparse-AutoAttack 的对抗攻击方法
Propose with the Gradient, Accept with the Loss: One Hedged Attack for Pixel-Sparse Robustness
做对抗鲁棒性评测的可以留意,这篇把 sPGD 和 Sparse-RS 统一成局部搜索框架,用五条梯度规则合成 GRS。18 个设置里 17 个追平或超过整个 sAA 集成,还快了 3.7 到 5 倍,跑鲁棒性基准的成本能降不少。
arXiv 论文提出 Gradient-guided Random Search(GRS),针对像素稀疏对抗攻击场景(攻击者可修改少量像素为任意值)。现有评测标准是集成式攻击 sparse-AutoAttack(sAA),其成员包括稀疏投影梯度下降(sPGD)和 Sparse-RS,两者各有盲区:sPGD 依赖的一阶模型在整像素改动下失效,Sparse-RS 则完全忽略梯度信息。
论文将两类攻击统一为“提议分布加接受规则”的局部搜索框架,用泰勒界推导出五条规则:用梯度符号设定新像素值、宽松参考梯度对候选像素的排序、更紧密参考其对待删像素的排序(后者更可靠)、同一像素在移动被接受前不重复提议、接受决策只看损失值。由此得到每步只需一次前向和一次反向传播的 GRS。
在 13 个 CIFAR-10 配置和 5 个 ImageNet 模型上,GRS 达到或超过 sAA 最佳单成员,17/18 个设置中与 sAA 相差不到一个点或更强,其中 8 个设置显著更强。计算开销上只需 3 个前向传播单元(sAA 最多需 7 个),在 CIFAR-10 上比 sAA 成员完整运行快 3.7 到 5.0 倍。
Propose with the Gradient, Accept with the Loss: One Hedged Attack for Pixel-Sparse Robustness
A pixel-sparse adversary may change a few pixels of an image to any value. Robustness to such attacks is currently measured with an ensemble, sparse-AutoAttack (sAA), because its members fail in different places: sparse projected gradient descent (sPGD) trusts a first-order model that a full pixel change breaks, and sparse random search (Sparse-RS) ignores the gradient even where it is informative. We show that both attacks are local searches, each a proposal distribution plus an acceptance rule, and ask what such a search can learn from the gradient. A Taylor bound says what to measure, and measurements at the states the search visits give five rules. The sign of the gradient is its most reliable use at a single pixel and sets a new pixel's value. Its ranking of candidate pixels should be followed loosely, and its ranking of the pixels to remove, which is often more reliable, can be followed more closely. A pixel should not be proposed twice before a move is accepted, and acceptance should be decided by the loss alone. These rules define Gradient-guided Random Search (GRS), which runs once and costs one forward and one backward pass per step. On 13 CIFAR-10 configurations and 5 ImageNet models, GRS matches or beats the best single member of sAA and is within one point of sAA itself or stronger on 17 of 18 settings, and it is significantly stronger than sAA on 8. It uses 3 instead of up to 7 forward-pass units, and on CIFAR-10 it runs 3.7 to 5.0 times faster than the members of sAA run in full.
本条情报归属 AITOP「论文」核心追踪栏目。涉及核心实体与技术议题包括:对抗攻击、robustness、sparse-AutoAttack、CIFAR-10,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明:本文原始线索采编自「arXiv cs.LG」(@Hon Ming Yan, Chun Pong Lau),经 AITOP 多维评分管线结构化提炼。