Rong Huang

Deep neural networks (DNNs) have become important intellectual assets, and ownership verification for misappropriated DNNs is increasingly important in Machine Learning as a Service (MLaaS) settings. Among existing DNN watermarking methods, backdoor watermarking is a typical approach for deployed ownership verification. However, existing methods still face two limitations. When verification relies on a finite trigger set, forged ownership evidence becomes difficult to rule out once the trigger samples are leaked or closely imitated. In addition, when watermark embedding modifies the service backbone, the predictor used for routine service is directly altered rather than kept unchanged. To address these limitations, we propose a backdoor DNN watermarking framework that combines secret-key-driven trigger group construction with a plug-and-play LoRA component. The proposed method regenerates the trigger groups used for verification from benign image pairs under a valid key whenever ownership needs to be checked, so ownership verification no longer depends on a finite stored trigger set. Meanwhile, watermark embedding is carried by an external LoRA component rather than by modifying the service backbone. In addition, we further optimize the LoRA configuration through a genetic search. Experiments on five benchmark datasets show that under the intended deployment protocol, the proposed method keeps the service predictor unchanged, enables effective ownership verification, and makes it difficult for attackers without the valid key to reproduce the verification behavior of the legitimate watermark under a large number of repeated attack trials.

目的镜面反射会干扰视觉感知系统，降低目标检测和三维重建等任务的性能。尽管现有高光去除方法已有进展，但大多依赖合成数据训练，难以有效处理真实场景中的反射。针对这一问题，本文构建了多光照条件下的汽车高光数据集，并提出掩码引导的高光去除网络（Mask Guided Transformer Unet，MG-TransUnet），显著提升高光去除效果。方法首先，本文构建了汽车高光数据集，覆盖多种光照条件及不同强度、分布的高光图像对，并提供高光掩码。并且提出协同优化的两阶段网络架构，第一阶段采用轻量级U形网络（U-shaped network，U-Net）实现高光检测，生成高光区域掩码；第二阶段将掩码作为空间先验，指导基于改进型窗口移位注意力（Shifted Window Attention Transformer，SWAT）的U-Net高光去除网络。通过端到端联合训练，实现检测与去除网络的参数协同优化。结果本文在自建数据集和三个公开数据集（PSD、SHIQ、SSHR）上进行了系统评估，与8种最先进方法进行对比。在PSD数据集上，本方法以26.37dB的峰值信噪比（pe-ak signal-to-noise ratio，PSNR）位列第二；在SHIQ数据集上以35.24dB的PSNR值优于第二名（34.13dB）1.11dB；在SSHR数据集上以36.54dB的PSNR值领先第二名（36.16dB）0.38dB；在自建数据集上以25.14dB的PSNR值优于第二名（25.03dB）0.11dB。此外，消融实验表明掩码引导网络和SWAT模块均能提升高光去除性能。结论本文构建了面向真实场景的汽车高光数据集，提出了基于窗口移位注意力机制的掩码引导网络（MG-TransUNet）。实验证明，本文方法在自建数据集上性能达到最优，并在多个公开数据集上取得了具有竞争力的表现。代码地址：https：//github.com/chenWULUQI/MG-Trans_Unet

Restricted preemption plays a crucial role in reducing total completion time while controlling preemption overhead. A typical version of restricted preemptive models is k-restricted preemptive scheduling, where preemption is only allowed after a task has been continuously processed for at least k units of time. Though solving this problem of minimizing the makespan on parallel machines is NP-hard in general, it is of vital importance to obtain the optimal solution for small-sized problems, as well as for evaluation of heuristics. This paper proposes optimal strategies to the aforementioned problem. Motivated by the dramatic speed-up of Boolean Satisfiability (SAT) solvers, we make the first step towards a study of applying a SAT solver to the k-restricted scheduling problem. We set out to encode the scheduling problem into propositional Boolean logic and determine the optimal makespan by repeatedly calling an off-the-shelf SAT solver. Moreover, we move one step further by encoding the problem into Partial Maximum Satisfiability (PMS), which is an optimized version of SAT, so that the explicit successive calls of the solver can be eliminated. The optimal makespan of the problem and the performance of the proposed methods are studied experimentally. Furthermore, an existing heuristic algorithm is evaluated by the optimization methods.