ZoAQ: Adaptive Zeroth-Order Querying via Query-Reuse Coupling
ZoAQ: Adaptive Zeroth-Order Querying via Query-Reuse Coupling
ZoAQ:基于查询重用耦合的自适应零阶查询方法
Abstract: Zeroth-order optimization (ZOO) estimates updates from function evaluations, making perturbation queries a primary cost. Fixed budgets spend the same number of queries at every step, while adaptive controllers may offset their savings by using additional oracle calls to test estimator reliability.
摘要: 零阶优化(ZOO)通过函数评估来估计更新,这使得扰动查询成为主要的成本来源。固定预算在每一步都消耗相同数量的查询,而自适应控制器虽然能节省查询,却往往需要通过额外的预言机(oracle)调用来测试估计器的可靠性,从而抵消了节省的成本。
We introduce ZoAQ, an adaptive ZOO method built around query reuse. Rather than discarding past evaluations after each step, ZoAQ makes them useful for both the next update and the decision to query further. This enables adaptive query allocation without extra validation queries.
我们引入了 ZoAQ,这是一种围绕“查询重用”构建的自适应 ZOO 方法。ZoAQ 不会在每一步之后丢弃过去的评估结果,而是将其同时用于下一次更新和决定是否需要进一步查询。这使得在无需额外验证查询的情况下,即可实现自适应的查询分配。
Our analysis characterizes when this agreement identifies an update that supports descent and guides the controller to a sufficient query budget. On synthetic objectives, ZoAQ reduces queries by 43-48% relative to fixed baselines using 1.2M queries.
我们的分析刻画了这种一致性在何时能识别出支持下降的更新,并引导控制器确定充足的查询预算。在合成目标函数上,与使用 120 万次查询的固定基准相比,ZoAQ 将查询量减少了 43-48%。
In black-box attacks, it reaches 100% success with 320 and 625 average queries on MNIST and CIFAR-10, respectively. Across four OPT fine-tuning settings, ZoAQ saves 43-46% forward evaluations relative to fixed K=4, with accuracy changes within tasks ranging from -0.018 to +0.010.
在黑盒攻击中,它在 MNIST 和 CIFAR-10 数据集上分别以平均 320 次和 625 次查询达到了 100% 的成功率。在四种 OPT 微调设置中,与固定 K=4 的基准相比,ZoAQ 节省了 43-46% 的前向评估,且任务内的准确率变化范围仅在 -0.018 到 +0.010 之间。