arXiv subjects

Dongdong Zhao

Publications and source records attributed to Dongdong Zhao.

16 recordsLinked to original sources

arXiv2026

Keep Evaluation Fair: Detecting Data Leakage in Code Generation Benchmarks via Membership Inference Attacks

Code generation benchmarks are widely used to evaluate Large Language Models (LLMs), but benchmark data leakage into training sets can inflate performance and undermine evaluation validity. DetectLeak, a method specifically designed for code generation benchmark leakage detection, relies on perplexity scores to identify likely leaked samples. However, perplexity mainly reflects general familiarity with code patterns and may perform poorly on complex or rare samples. It also overlooks other useful signals, such as code similarity, functional correctness, and semantic representations. To address these limitations, we propose CGMIA (Code-Generation-specific Membership Inference Attack), a method for detecting leakage in code generation benchmarks. CGMIA fine-tunes a shadow model on a subset of benchmark samples to construct labeled member and non-member data. For each sample, it collects the input prompt, generated code, and reference solution, and extracts expert features, including CodeBLEU, edit distance, test pass rate, and perplexity, together with semantic features from CodeBERT embeddings. An integrated learning module combines these features to capture both surface-level memorization signals and deeper behavioral patterns, enabling a classifier to predict whether a sample was included in the target model's training set. Experiments on eight code generation benchmarks show that CGMIA outperforms eight existing membership inference methods in most cases. It also effectively detects known leaked APPS samples in StarCoder-7B's training data.

cs.SE↗

arXiv2026

Benign on Label, Malicious by Design: Clean-Label Dormant-to-Activated Backdoor via Machine Unlearning with Removable Camouflage

Existing backdoor attacks often become effective immediately after backdoor implantation and may therefore be exposed before exploitation. Machine unlearning activated dormant backdoors mitigate such behavioral exposure by remaining inactive after training and becoming effective only after selected training records are unlearned. However, existing methods struggle to simultaneously achieve a low pre-unlearning attack success rate and strong post-unlearning activation under clean-label constraints and realistic unlearning requests. Achieving this transition requires jointly establishing a persistent latent association and a removable suppressive influence. To address this challenge, we propose a clean-label unlearning-activated backdoor framework based on dual-generator learning and formulate it as a bilevel optimization problem: By simulating latent backdoor establishment and machine unlearning, the framework alternately learns sample-specific triggers that establish a latent trigger-to-target association and label-consistent camouflage samples that provide removable suppression. Once a small subset of camouflage samples is unlearned, the suppression is lifted and the dormant backdoor is activated. Experiments on CIFAR-10 and ImageNet-10 show that our method maintains lower pre-unlearning attack success rates while achieving stronger post-unlearning activation across multiple unlearning algorithms than representative backdoor baselines. These results demonstrate that reliable dormancy-to-activation transitions can be achieved by coordinating a persistent latent association with removable suppression under clean-label and realistic deletion constraints.

cs.CR↗

arXiv2026

From Classification to Consistent Templates: Multiple Permuted-Label Classifier Encoding for Biometric Template Protection

Biometric template protection (BTP) must secure stored templates while tolerating intra-class variations. Existing methods rely on protected-domain similarity matching, error correction, or predefined-template mappings, potentially retaining exploitable similarity structures, introducing helper-data risks, depending on artificial targets, or coupling protection to specific modalities. Storing only cryptographic hash digests eliminates directly comparable representations and conceals pre-hash templates, but hash-based exact-match verification requires genuine samples to generate identical intermediate templates before hashing. Identity classification is naturally suited to this requirement because it maps variable biometric samples to stable and discriminative identity-level outputs. Based on this insight, we propose Multiple Permuted-Label Classifier Encoding (MPLCE). Through classifier-specific label permutations, MPLCE assigns each identity different labels across multiple classifiers. The predicted labels are encoded and concatenated to form an intermediate template, preventing repeated encodings of a single identity label and enlarging the effective candidate space while preserving classification consistency. The template is randomized with an application-specific XOR string and cryptographically hashed, enabling exact-match verification without error correction codes or biometric-dependent helper data. Using modality-specific classifiers, MPLCE retains the same template generation and protection procedure across modalities. On four face and two iris datasets, MPLCE achieves competitive performance, including a GAR of 98.61\% at a FAR of 5.51$\times$10\textsuperscript{-5}\% on YTF and a GAR of 99.10\% at a FAR of 0.00\% on CASIA-Iris-Lamp. Security analyses and attack evaluations support its irreversibility, revocability, and unlinkability under the threat model.

cs.CR↗

arXiv2026

Discard the Dross and Select the Essential: Pre-query Sample Selection for Black-box Membership Inference Attacks

Black-box membership inference attacks (MIAs) rely on target-model queries to infer whether candidate samples were used for training. However, membership signals are highly non-uniform across samples: some candidate samples support strong member/non-member separability, whereas many others provide little useful signal. Consequently, indiscriminate querying can incur substantial query cost and increase query-induced exposure, with limited marginal benefit for inference. This raises a key question: which candidate samples are worth querying for black-box MIAs? To address this question, we propose PSS-MIA, a pre-query sample selection framework which can be embedded with any existing MIA methods. PSS-MIA proceeds in two stages: it first ranks candidate samples and selects a subset expected to support stronger membership inference, then queries the selected samples and uses the returned outputs for an existing black-box MIA, thereby reducing query cost and query-induced exposure. In the first stage, we propose Loss-Gap Ranking (LGR), which ranks candidate samples by estimating the strength of their membership signal using loss gaps computed from reference models. Experiments on CIFAR-10, CIFAR-100, and CINIC-10 with five representative black-box MIA methods demonstrate that PSS-MIA with LGR consistently outperforms all other compared methods. Moreover, under a 0.1% FPR constraint, PSS-MIA can save at least 83.1%, 60.6%, and 80.4% of the query budget for the three datasets, respectively.

cs.CR↗

arXiv2026

Bionic Human-Motion Style Transfer for Physically Executable Whole-Body Control of Humanoid Robots

Expressive whole-body motion is important for humanoid robots operating in human environments, where robots are expected to move stably while presenting readable and adjustable body behaviors. However, most expressive motions are still obtained from fixed demonstrations or manually designed scripts, making it difficult to reuse a demonstrated style across different motion contents. Inspired by the way human motion styles convey affective and intentional cues through gait rhythm, posture, arm swing and body sway, this paper proposes a bionic generation-to-control framework for exemplar-driven style transfer on humanoid robots. Given a short human style exemplar and a target content motion, the proposed framework generates a stylized whole-body reference that preserves the intended motion content while transferring the demonstrated style. A physics-aware multi-condition latent diffusion model is developed to fuse style, content and trajectory conditions, and classifier-free guidance is used to adjust the style intensity without retraining. To improve hardware executability, contact-consistency and temporal-smoothness regularization are imposed on decoded motions during training. The generated references are then converted into G1-compatible robot references and executed by a preview-based whole-body tracking policy trained with a cluster-and-distill strategy. Simulation and Unitree G1 experiments show that the proposed method can transfer short human style exemplars to diverse robot motion contents, reduce contact and jitter artifacts compared with animation-oriented style-transfer baselines, and achieve a 96.0% success rate over 125 reported real-robot trials. The results demonstrate the feasibility of using short human motion exemplars as reusable bionic sources for physically executable expressive humanoid motion.

cs.RO↗

arXiv2026

BackWeak: Backdooring Knowledge Distillation Simply with Weak Triggers and Fine-tuning

Knowledge Distillation (KD) is essential for compressing large models, yet relying on pre-trained "teacher" models downloaded from third-party repositories introduces serious security risks--most notably backdoor attacks. Existing KD backdoor methods are typically complex and computationally intensive: they employ surrogate student models and simulated distillation to guarantee transferability, and construct triggers similar to universal adversarial perturbations (UAPs), which being not stealthy in magnitude, inherently exhibit strong adversarial behavior. This work questions whether such complexity is necessary and constructs stealthy "weak" triggers--imperceptible perturbations that have negligible adversarial effect. We propose BackWeak, a simple, surrogate-free attack paradigm. BackWeak shows that a powerful backdoor can be implanted by simply fine-tuning a benign teacher with a weak trigger using a very small learning rate. We demonstrate that this delicate fine-tuning is sufficient to embed a backdoor that reliably transfers to diverse student architectures during a victim's standard distillation process, yielding high attack success rates. Extensive empirical evaluations on multiple datasets, model architectures, and KD methods show that BackWeak is efficient, simpler, and often more stealthy than previous elaborate approaches. This work calls on researchers studying KD backdoor attacks to pay particular attention to the trigger's potential adversarial characteristics.

cs.CR↗

arXiv2025

Model Inversion Attack Against Deep Hashing

Deep hashing improves retrieval efficiency through compact binary codes, yet it introduces severe and often overlooked privacy risks. The ability to reconstruct original training data from hash codes could lead to serious threats such as biometric forgery and privacy breaches. However, model inversion attacks specifically targeting deep hashing models remain unexplored, leaving their security implications unexamined. This research gap stems from the inaccessibility of genuine training hash codes and the highly discrete Hamming space, which prevents existing methods from adapting to deep hashing. To address these challenges, we propose DHMI, the first diffusion-based model inversion framework designed for deep hashing. DHMI first clusters an auxiliary dataset to derive semantic hash centers as surrogate anchors. It then introduces a surrogate-guided denoising optimization method that leverages a novel attack metric (fusing classification consistency and hash proximity) to dynamically select candidate samples. A cluster of surrogate models guides the refinement of these candidates, ensuring the generation of high-fidelity and semantically consistent images. Experiments on multiple datasets demonstrate that DHMI successfully reconstructs high-resolution, high-quality images even under the most challenging black-box setting, where no training hash codes are available. Our method outperforms the existing state-of-the-art model inversion attacks in black-box scenarios, confirming both its practical efficacy and the critical privacy risks inherent in deep hashing systems.

cs.CV↗

arXiv2025

Known Meets Unknown: Mitigating Overconfidence in Open Set Recognition

Open Set Recognition (OSR) requires models not only to accurately classify known classes but also to effectively reject unknown samples. However, when unknown samples are semantically similar to known classes, inter-class overlap in the feature space often causes models to assign unjustifiably high confidence to them, leading to misclassification as known classes -- a phenomenon known as overconfidence. This overconfidence undermines OSR by blurring the decision boundary between known and unknown classes. To address this issue, we propose a framework that explicitly mitigates overconfidence caused by inter-class overlap. The framework consists of two components: a perturbation-based uncertainty estimation module, which applies controllable parameter perturbations to generate diverse predictions and quantify predictive uncertainty, and an unknown detection module with distinct learning-based classifiers, implemented as a two-stage procedure, which leverages the estimated uncertainty to improve discrimination between known and unknown classes, thereby enhancing OSR performance. Experimental results on three public datasets show that the proposed framework achieves superior performance over existing OSR methods.

cs.CV↗

arXiv2025

Injection, Attack and Erasure: Revocable Backdoor Attacks via Machine Unlearning

Backdoor attacks pose a persistent security risk to deep neural networks (DNNs) due to their stealth and durability. While recent research has explored leveraging model unlearning mechanisms to enhance backdoor concealment, existing attack strategies still leave persistent traces that may be detected through static analysis. In this work, we introduce the first paradigm of revocable backdoor attacks, where the backdoor can be proactively and thoroughly removed after the attack objective is achieved. We formulate the trigger optimization in revocable backdoor attacks as a bilevel optimization problem: by simulating both backdoor injection and unlearning processes, the trigger generator is optimized to achieve a high attack success rate (ASR) while ensuring that the backdoor can be easily erased through unlearning. To mitigate the optimization conflict between injection and removal objectives, we employ a deterministic partition of poisoning and unlearning samples to reduce sampling-induced variance, and further apply the Projected Conflicting Gradient (PCGrad) technique to resolve the remaining gradient conflicts. Experiments on CIFAR-10 and ImageNet demonstrate that our method maintains ASR comparable to state-of-the-art backdoor attacks, while enabling effective removal of backdoor behavior after unlearning. This work opens a new direction for backdoor attack research and presents new challenges for the security of machine learning systems.

cs.CR↗

arXiv2025

A Privacy-Preserving Data Collection Method for Diversified Statistical Analysis

Data perturbation-based privacy-preserving methods have been widely adopted in various scenarios due to their efficiency and the elimination of the need for a trusted third party. However, these methods primarily focus on individual statistical indicators, neglecting the overall quality of the collected data from a distributional perspective. Consequently, they often fall short of meeting the diverse statistical analysis requirements encountered in practical data analysis. As a promising sensitive data perturbation method, negative survey methods is able to complete the task of collecting sensitive information distribution while protecting personal privacy. Yet, existing negative survey methods are primarily designed for discrete sensitive information and are inadequate for real-valued data distributions. To bridge this gap, this paper proposes a novel real-value negative survey model, termed RVNS, for the first time in the field of real-value sensitive information collection. The RVNS model exempts users from the necessity of discretizing their data and only requires them to sample a set of data from a range that deviates from their actual sensitive details, thereby preserving the privacy of their genuine information. Moreover, to accurately capture the distribution of sensitive information, an optimization problem is formulated, and a novel approach is employed to solve it. Rigorous theoretical analysis demonstrates that the RVNS model conforms to the differential privacy model, ensuring robust privacy preservation. Comprehensive experiments conducted on both synthetic and real-world datasets further validate the efficacy of the proposed method.

cs.CR↗

arXiv2025

High-Precision Transformer-Based Visual Servoing for Humanoid Robots in Aligning Tiny Objects

High-precision tiny object alignment remains a common and critical challenge for humanoid robots in real-world. To address this problem, this paper proposes a vision-based framework for precisely estimating and controlling the relative position between a handheld tool and a target object for humanoid robots, e.g., a screwdriver tip and a screw head slot. By fusing images from the head and torso cameras on a robot with its head joint angles, the proposed Transformer-based visual servoing method can correct the handheld tool's positional errors effectively, especially at a close distance. Experiments on M4-M8 screws demonstrate an average convergence error of 0.8-1.3 mm and a success rate of 93\%-100\%. Through comparative analysis, the results validate that this capability of high-precision tiny object alignment is enabled by the Distance Estimation Transformer architecture and the Multi-Perception-Head mechanism proposed in this paper.

cs.CV↗

arXiv2024

BioDeepHash: Mapping Biometrics into a Stable Code

With the wide application of biometrics, more and more attention has been paid to the security of biometric templates. However most of existing biometric template protection (BTP) methods have some security problems, e.g. the problem that protected templates leak part of the original biometric data (exists in Cancelable Biometrics (CB)), the use of error-correcting codes (ECC) leads to decodable attack, statistical attack (exists in Biometric Cryptosystems (BCS)), the inability to achieve revocability (exists in methods using Neural Network (NN) to learn pre-defined templates), the inability to use cryptographic hash to guarantee strong security (exists in CB and methods using NN to learn latent templates). In this paper, we propose a framework called BioDeepHash based on deep hashing and cryptographic hashing to address the above four problems, where different biometric data of the same user are mapped to a stable code using deep hashing instead of predefined binary codes thus avoiding the use of ECC. An application-specific binary string is employed to achieve revocability. Then cryptographic hashing is used to get the final protected template to ensure strong security. Ultimately our framework achieves not storing any data that would leak part of the original biometric data. We also conduct extensive experiments on facial and iris datasets. Our method achieves an improvement of 10.12$\%$ on the average Genuine Acceptance Rate (GAR) for iris data and 3.12$\%$ for facial data compared to existing methods. In addition, BioDeepHash achieves extremely low False Acceptance Rate (FAR), i.e. 0$\%$ FAR on the iris dataset and the highest FAR on the facial dataset is only 0.0002$\%$.

cs.CR↗

arXiv2024

Inoculating solid-state homogeneous precipitation by impurity atoms through a spinodal decomposition like pathway

Solid-state homogeneous precipitation of nano-sized precipitates is one of the most effective processes to strengthen metal alloys, where the final density and size distribution of precipitates are largely controlled by the precipitation kinetics. Here, we report a strategy to inoculate the homogeneous precipitation of coherent precipitates to enhance the precipitation strengthening. Using the technologically important dilute Al-Zr alloys as an example, we demonstrate that an addition of a trace level of economical and readily available, non-L1$_{2}$ phase forming impurity atoms, X (X= Sn, Sb, Bi or Cd) and Si, can significantly enhance the diffusivity of Zr atoms and overturn the precipitation of L1$_{2}$-structured Al$_{3}$Zr nanoparticles from the classical homogeneous nucleation and growth pathway into a nonclassical nucleation pathway: Al$_{3}$Zr forms through the spontaneous formation of nano-scale local concentration fluctuations of Zr atoms on Zr-X(-Si)-vacancy clusters followed by a continuous increase of the concentration and chemical short-range ordering (CSRO). Such an impurity atoms induced heterogeneous nucleation based on a "spinodal decomposition like" mechanism dramatically accelerates the precipitation kinetics, leading to an order of magnitude higher number density of precipitates and a record high hardening efficiency of solute Zr atoms. By formulating the generalized selection principles for inoculating impurity elements, this inoculation strategy should be extendable to a broader range of materials to further explore the precipitation strengthening potentials.

cond-mat.mtrl-sci↗

arXiv2023

Motion-aware Memory Network for Fast Video Salient Object Detection

Previous methods based on 3DCNN, convLSTM, or optical flow have achieved great success in video salient object detection (VSOD). However, they still suffer from high computational costs or poor quality of the generated saliency maps. To solve these problems, we design a space-time memory (STM)-based network, which extracts useful temporal information of the current frame from adjacent frames as the temporal branch of VSOD. Furthermore, previous methods only considered single-frame prediction without temporal association. As a result, the model may not focus on the temporal information sufficiently. Thus, we initially introduce object motion prediction between inter-frame into VSOD. Our model follows standard encoder--decoder architecture. In the encoding stage, we generate high-level temporal features by using high-level features from the current and its adjacent frames. This approach is more efficient than the optical flow-based methods. In the decoding stage, we propose an effective fusion strategy for spatial and temporal branches. The semantic information of the high-level features is used to fuse the object details in the low-level features, and then the spatiotemporal features are obtained step by step to reconstruct the saliency maps. Moreover, inspired by the boundary supervision commonly used in image salient object detection (ISOD), we design a motion-aware loss for predicting object boundary motion and simultaneously perform multitask learning for VSOD and object motion prediction, which can further facilitate the model to extract spatiotemporal features accurately and maintain the object integrity. Extensive experiments on several datasets demonstrated the effectiveness of our method and can achieve state-of-the-art metrics on some datasets. The proposed model does not require optical flow or other preprocessing, and can reach a speed of nearly 100 FPS during inference.

cs.CV↗

arXiv2023

Data-driven prognostics based on time-frequency analysis and symbolic recurrent neural network for fuel cells under dynamic load

Data-centric prognostics is beneficial to improve the reliability and safety of proton exchange membrane fuel cell (PEMFC). For the prognostics of PEMFC operating under dynamic load, the challenges come from extracting degradation features, improving prediction accuracy, expanding the prognostics horizon, and reducing computational cost. To address these issues, this work proposes a data-driven PEMFC prognostics approach, in which Hilbert-Huang transform is used to extract health indicator in dynamic operating conditions and symbolic-based gated recurrent unit model is used to enhance the accuracy of life prediction. Comparing with other state-of-the-art methods, the proposed data-driven prognostics approach provides a competitive prognostics horizon with lower computational cost. The prognostics performance shows consistency and generalizability under different failure threshold settings.

cs.LG↗

arXiv2022

NegDL: Privacy-Preserving Deep Learning Based on Negative Database

In the era of big data, deep learning has become an increasingly popular topic. It has outstanding achievements in the fields of image recognition, object detection, and natural language processing et al. The first priority of deep learning is exploiting valuable information from a large amount of data, which will inevitably induce privacy issues that are worthy of attention. Presently, several privacy-preserving deep learning methods have been proposed, but most of them suffer from a non-negligible degradation of either efficiency or accuracy. Negative database (\textit{NDB}) is a new type of data representation which can protect data privacy by storing and utilizing the complementary form of original data. In this paper, we propose a privacy-preserving deep learning method named NegDL based on \textit{NDB}. Specifically, private data are first converted to \textit{NDB} as the input of deep learning models by a generation algorithm called \textit{QK}-hidden algorithm, and then the sketches of \textit{NDB} are extracted for training and inference. We demonstrate that the computational complexity of NegDL is the same as the original deep learning model without privacy protection. Experimental results on Breast Cancer, MNIST, and CIFAR-10 benchmark datasets demonstrate that the accuracy of NegDL could be comparable to the original deep learning model in most cases, and it performs better than the method based on differential privacy.

cs.CR↗