Ziqi Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ziqi Zhang has a total of 46 co-authors. Here are the top co-authors by number of shared publications:
- Ding Li
Key Laboratory of High-Confidence Software Technologies (MOE), School of Computer Science, Peking University, Beijing, China
12 shared publications - Yao Guo
Peking University
11 shared publications - Xiangqun Chen
Peking University, Beijing, China
9 shared publications - Yifeng Cai
Peking University, Beijing, China
5 shared publications - Zhineng Zhong
Peking University, Beijing, China
4 shared publications - Junlin Liu
MOE Key Lab of HCST (PKU), School of Computer Science, Peking University, Beijing, China
3 shared publications
Latest Publications
- Open Access
- 30 September 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10 (3) , 1-29
GUIA uditor : Enabling Post-hoc Child Safety Forensics via Action-Guided GUI Provenance on Mobile Devices
The proliferation of smart devices exposes children to online risks like grooming and financial scams that are deeply embedded within legitimate applications. Current approaches rely on automated prevention and detection, a paradigm that is fundamentally limited by its inherent fallibility. Whether rule-based or AI-driven, they inevitably produce false positives and negatives, failing to provide reliable protection. In this paper, we argue for a complementary, human-in-the-loop, post-hoc forensic paradigm. We present GUIA uditor , the first system designed to realize this vision by creating GUI Provenance: a queryable, semantic record of a child's interaction sequence. To generate this, GUIA uditor leverages a Multimodal Large Language Model (MLLM) to translate the temporal sequence of GUI events into a human-understandable narrative. To make this practical on mobile devices, a novel evidence distillation pipeline reduces the data requiring analysis by over 89.2% compared to periodic sampling approaches adopted by industry standards, with negligible impact on accuracy. On a new dataset of 295 interaction clips, GUIA uditor achieves a 95.23% Macro-F1 Score in logging significant events and, crucially, its two-stage forensic query engine successfully retrieves the correct evidence as the top result for over 90.20% of natural language questions. An end-to-end evaluation on three modern smartphones shows that the full pipeline, including on-device MLLM inference, adds 2.1W of power draw and 7.4s of per-event latency, with a peak memory footprint of ~3.1GB. These results show that post-hoc GUI forensics can run on modern mobile devices and provide useful context for guardian-led safety review.
- 30 September 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10 (3) , 1-29
Open AccessGUIA uditor : Enabling Post-hoc Child Safety Forensics via Action-Guided GUI Provenance on Mobile Devices
The proliferation of smart devices exposes children to online risks like grooming and financial scams that are deeply embedded within legitimate applications. Current approaches rely on automated prevention and detection, a paradigm that is fundamentally limited by its inherent fallibility. Whether rule-based or AI-driven, they inevitably produce false positives and negatives, failing to provide reliable protection. In this paper, we argue for a complementary, human-in-the-loop, post-hoc forensic paradigm. We present GUIA uditor , the first system designed to realize this vision by creating GUI Provenance: a queryable, semantic record of a child's interaction sequence. To generate this, GUIA uditor leverages a Multimodal Large Language Model (MLLM) to translate the temporal sequence of GUI events into a human-understandable narrative. To make this practical on mobile devices, a novel evidence distillation pipeline reduces the data requiring analysis by over 89.2% compared to periodic sampling approaches adopted by industry standards, with negligible impact on accuracy. On a new dataset of 295 interaction clips, GUIA uditor achieves a 95.23% Macro-F1 Score in logging significant events and, crucially, its two-stage forensic query engine successfully retrieves the correct evidence as the top result for over 90.20% of natural language questions. An end-to-end evaluation on three modern smartphones shows that the full pipeline, including on-device MLLM inference, adds 2.1W of power draw and 7.4s of per-event latency, with a peak memory footprint of ~3.1GB. These results show that post-hoc GUI forensics can run on modern mobile devices and provide useful context for guardian-led safety review.
- Open Access
- 27 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 693-703
Governing Global ASID/PCID Management for QoS Isolation in Shared-Kernel Mixed-Priority Systems
- 27 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 693-703
Open AccessGoverning Global ASID/PCID Management for QoS Isolation in Shared-Kernel Mixed-Priority Systems
- Open Access
- 12 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
A General Framework for Defending against Query-based Black-box Attacks on Neural Network Models
As Machine Learning as a Service (MLaaS) becomes increasingly prevalent, query-based black-box attacks pose a significant threat to machine learning systems. These attacks manipulate input queries to generate adversarial examples, resulting in misclassification or undesired outputs. To address these threats, researchers have developed various query moderation techniques. However, recent studies have revealed that these techniques are vulnerable to sophisticated adaptive attacks. In this paper, we propose a general framework, Query Provenance Analysis (QPA), to defend against query-based black-box attacks both robustly (against non-adaptive and adaptive attacks) and efficiently (in real-time). Our key insight is that, instead of focusing on individual queries, utilizing features from the query sequence (termed query provenance ) can distinguish malicious queries from benign queries more effectively. We construct a query provenance graph to capture the relationship between a new query and prior historical queries, and then design efficient algorithms to detect malicious queries based on the query provenance graphs. We evaluate QPA on image classification and Natural Language Processing (NLP) tasks. QPA significantly reduces the Attack Success Rate (ASR) compared to baselines, achieving up to 20 \(\times\) lower ASR for image classification and 10 \(\times\) lower ASR for NLP, while also improving throughput and latency across all tasks.- 12 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
Open AccessA General Framework for Defending against Query-based Black-box Attacks on Neural Network Models
As Machine Learning as a Service (MLaaS) becomes increasingly prevalent, query-based black-box attacks pose a significant threat to machine learning systems. These attacks manipulate input queries to generate adversarial examples, resulting in misclassification or undesired outputs. To address these threats, researchers have developed various query moderation techniques. However, recent studies have revealed that these techniques are vulnerable to sophisticated adaptive attacks. In this paper, we propose a general framework, Query Provenance Analysis (QPA), to defend against query-based black-box attacks both robustly (against non-adaptive and adaptive attacks) and efficiently (in real-time). Our key insight is that, instead of focusing on individual queries, utilizing features from the query sequence (termed query provenance ) can distinguish malicious queries from benign queries more effectively. We construct a query provenance graph to capture the relationship between a new query and prior historical queries, and then design efficient algorithms to detect malicious queries based on the query provenance graphs. We evaluate QPA on image classification and Natural Language Processing (NLP) tasks. QPA significantly reduces the Attack Success Rate (ASR) compared to baselines, achieving up to 20 \(\times\) lower ASR for image classification and 10 \(\times\) lower ASR for NLP, while also improving throughput and latency across all tasks. - Open Access
- 26 May 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Cyber-Physical Systems
- Vol. 10 (4) , 1-31
Optical Lens Attack on Monocular Depth Estimation for Autonomous Driving
Monocular Depth Estimation (MDE) is a pivotal component of vision-based Autonomous Driving (AD) systems, enabling vehicles to estimate the depth of surrounding objects using a single camera image. This estimation guides essential driving decisions, such as braking before an obstacle or changing lanes to avoid collisions. In this paper, we explore vulnerabilities of MDE algorithms in AD systems, presenting LensAttack , a novel physical attack that strategically places optical lenses on the camera of an autonomous vehicle to manipulate the perceived object depths. LensAttack encompasses two attack formats: concave lens attack and convex lens attack, each utilizing different optical lenses to induce false depth perception. We first develop a mathematical model that outlines the parameters of the attack, followed by simulations and real-world evaluations to assess its efficacy on state-of-the-art MDE models. Additionally, we adopt an attack optimization method to further enhance the attack success rate by optimizing the attack focal length. To better evaluate the implications of LensAttack on AD, we conduct comprehensive end-to-end system simulations using the CARLA platform. The results reveal that LensAttack can significantly disrupt the depth estimation processes in AD systems, posing a serious threat to their reliability and safety. Finally, we discuss some potential defense methods to mitigate the effects of the proposed attack.
- 26 May 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Cyber-Physical Systems
- Vol. 10 (4) , 1-31
Open AccessOptical Lens Attack on Monocular Depth Estimation for Autonomous Driving
Monocular Depth Estimation (MDE) is a pivotal component of vision-based Autonomous Driving (AD) systems, enabling vehicles to estimate the depth of surrounding objects using a single camera image. This estimation guides essential driving decisions, such as braking before an obstacle or changing lanes to avoid collisions. In this paper, we explore vulnerabilities of MDE algorithms in AD systems, presenting LensAttack , a novel physical attack that strategically places optical lenses on the camera of an autonomous vehicle to manipulate the perceived object depths. LensAttack encompasses two attack formats: concave lens attack and convex lens attack, each utilizing different optical lenses to induce false depth perception. We first develop a mathematical model that outlines the parameters of the attack, followed by simulations and real-world evaluations to assess its efficacy on state-of-the-art MDE models. Additionally, we adopt an attack optimization method to further enhance the attack success rate by optimizing the attack focal length. To better evaluate the implications of LensAttack on AD, we conduct comprehensive end-to-end system simulations using the CARLA platform. The results reveal that LensAttack can significantly disrupt the depth estimation processes in AD systems, posing a serious threat to their reliability and safety. Finally, we discuss some potential defense methods to mitigate the effects of the proposed attack.
- Open Access
- 25 February 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
- Vol. 35 (10) , 1-37
LLMGuard : Safeguarding Real-Time Inference for Large Language Models on Edge Devices
TEE-shielded secure inference offers an efficient solution to protect valuable edge-deployed models from potential thefts. Nevertheless, existing methods are lack of theoretical security analysis, failing to achieve the optimal security. Furthermore, while feasible for small models, existing methods are excessively heavyweight for Large Language Models (LLM). For LLaMA-7B, they introduce GB-level secure memory requirement and hundredfold inference latency, severely compromising real-time utility. To solve these problems, we first present a Bayesian theory framework of Model Stealing (MS) attacks, which decomposes MS into prior and posterior knowledge leakage. Based on this framework, LLMGuard is proposed, which presents two components: First, Intrinsic Parameters Shielding is designed to shield all private parameters, preventing prior knowledge leakage. This approach significantly decreases the secure memory usage and achieves inference speedup. Second, since OTP is not applicable to LLMs, Random Slices Composition is developed to obfuscate intermediate distributions with no computational overhead, minimizing posterior knowledge leakage efficiently. Experimental results demonstrate that LLMGuard downgrades model to black-box inference with negligible accuracy loss, while delivering \(43\times\) inference speedup on LLaMA compared to fully-shielded methods. The proposed LLMGuard effectively addresses concerns related to intellectual property theft on edge, boosting the secure deployment of LLMs on untrusted devices.- 25 February 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
- Vol. 35 (10) , 1-37
Open AccessLLMGuard : Safeguarding Real-Time Inference for Large Language Models on Edge Devices
TEE-shielded secure inference offers an efficient solution to protect valuable edge-deployed models from potential thefts. Nevertheless, existing methods are lack of theoretical security analysis, failing to achieve the optimal security. Furthermore, while feasible for small models, existing methods are excessively heavyweight for Large Language Models (LLM). For LLaMA-7B, they introduce GB-level secure memory requirement and hundredfold inference latency, severely compromising real-time utility. To solve these problems, we first present a Bayesian theory framework of Model Stealing (MS) attacks, which decomposes MS into prior and posterior knowledge leakage. Based on this framework, LLMGuard is proposed, which presents two components: First, Intrinsic Parameters Shielding is designed to shield all private parameters, preventing prior knowledge leakage. This approach significantly decreases the secure memory usage and achieves inference speedup. Second, since OTP is not applicable to LLMs, Random Slices Composition is developed to obfuscate intermediate distributions with no computational overhead, minimizing posterior knowledge leakage efficiently. Experimental results demonstrate that LLMGuard downgrades model to black-box inference with negligible accuracy loss, while delivering \(43\times\) inference speedup on LLaMA compared to fully-shielded methods. The proposed LLMGuard effectively addresses concerns related to intellectual property theft on edge, boosting the secure deployment of LLMs on untrusted devices. - 18 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1, 1-7
SecureInfer: Heterogeneous TEE-GPU Architecture for Privacy-Critical Tensors for Large Language Model Deployment
With the increasing deployment of Large Language Models (LLMs) on mobile and edge platforms, securing them against model extraction attacks has become a pressing concern. However, protecting model privacy without sacrificing the performance benefits of untrusted AI accelerators, such as GPUs, presents a challenging trade-off. In this paper, we initiate the study of high-performance execution on LLMs and present SecureInfer, a hybrid framework that leverages a heterogeneous Trusted Execution Environments (TEEs)-GPU architecture to isolate privacy-critical components while offloading compute-intensive operations to untrusted accelerators. Building upon an outsourcing scheme, SecureInfer adopts an information-theoretic and threat-informed partitioning strategy: security-sensitive components, including non-linear layers, projection of attention head, FNN transformations, and LoRA adapters are executed inside an SGX enclave, while other linear operations (matrix multiplication) are performed on the GPU after encryption and are securely restored within the enclave. We implement a prototype of SecureInfer using the LLaMA-2 model and evaluate it across performance and security metrics. Our results show that SecureInfer offers strong security guarantees with reasonable performance, offering a practical solution for secure on-device model inference.
- 18 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1, 1-7
SecureInfer: Heterogeneous TEE-GPU Architecture for Privacy-Critical Tensors for Large Language Model Deployment
With the increasing deployment of Large Language Models (LLMs) on mobile and edge platforms, securing them against model extraction attacks has become a pressing concern. However, protecting model privacy without sacrificing the performance benefits of untrusted AI accelerators, such as GPUs, presents a challenging trade-off. In this paper, we initiate the study of high-performance execution on LLMs and present SecureInfer, a hybrid framework that leverages a heterogeneous Trusted Execution Environments (TEEs)-GPU architecture to isolate privacy-critical components while offloading compute-intensive operations to untrusted accelerators. Building upon an outsourcing scheme, SecureInfer adopts an information-theoretic and threat-informed partitioning strategy: security-sensitive components, including non-linear layers, projection of attention head, FNN transformations, and LoRA adapters are executed inside an SGX enclave, while other linear operations (matrix multiplication) are performed on the GPU after encryption and are securely restored within the enclave. We implement a prototype of SecureInfer using the LLaMA-2 model and evaluate it across performance and security metrics. Our results show that SecureInfer offers strong security guarantees with reasonable performance, offering a practical solution for secure on-device model inference.
- Open Access
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 269-282
Cost-Efficient Cloud Infrastructure with Hugepage-aware Memory Deduplication
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 269-282
Open AccessCost-Efficient Cloud Infrastructure with Hugepage-aware Memory Deduplication
- Open Access
- 4 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 9-11
WiP: Beyond App Blocking: Fine-Grained Child Online Safety via MLLM-based GUI Provenance on the Edge
- 4 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 9-11
Open AccessWiP: Beyond App Blocking: Fine-Grained Child Online Safety via MLLM-based GUI Provenance on the Edge
- Open Access
- 4 October 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
- Vol. 34 (8) , 1-28
Not All Exceptions Are Created Equal: Triaging Error Logs in Real-World Enterprises
Error logs like Java exceptions play a crucial role in diagnosing and resolving errors within the industry. Nonetheless, the extensive logging of Java exceptions may result in exception fatigue in large-scale Java systems at an industrial level, where the frequency of Java exceptions being generated surpasses developers’ ability to manage them effectively. Regrettably, there is a lack of research on the seriousness, prevalence, and solutions to this problem. To close this gap, we first make a comprehensive investigation into the exception fatigue problem within a prominent Internet corporation in China, namely Alibaba, confirming its importance in the industry. Consequently, we introduce a novel solution called ABEL , designed to automatically pinpoint the most relevant exceptions associated with software failures. The key challenge lies in the randomness of exceptions, which prevents existing sequence-based techniques from being effective. To address this challenge, ABEL establishes correlations between Java exceptions and the Key Performance Indicator (KPI) of applications, enabling the identification of exceptions leading to irregularities in KPI. Our evaluation of ABEL across four Java applications and five business KPIs within Alibaba illustrates its capability to pinpoint the primary cause of exception logs with an AC@5 (top-5 accuracy) exceeding 90%, effectively mitigating the exception fatigue problem within Alibaba. Furthermore, it can identify the root-cause exceptions in a real software failure within just 4 minutes, outperforming the manual investigation process by over an hour.
- 4 October 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
- Vol. 34 (8) , 1-28
Open AccessNot All Exceptions Are Created Equal: Triaging Error Logs in Real-World Enterprises
Error logs like Java exceptions play a crucial role in diagnosing and resolving errors within the industry. Nonetheless, the extensive logging of Java exceptions may result in exception fatigue in large-scale Java systems at an industrial level, where the frequency of Java exceptions being generated surpasses developers’ ability to manage them effectively. Regrettably, there is a lack of research on the seriousness, prevalence, and solutions to this problem. To close this gap, we first make a comprehensive investigation into the exception fatigue problem within a prominent Internet corporation in China, namely Alibaba, confirming its importance in the industry. Consequently, we introduce a novel solution called ABEL , designed to automatically pinpoint the most relevant exceptions associated with software failures. The key challenge lies in the randomness of exceptions, which prevents existing sequence-based techniques from being effective. To address this challenge, ABEL establishes correlations between Java exceptions and the Key Performance Indicator (KPI) of applications, enabling the identification of exceptions leading to irregularities in KPI. Our evaluation of ABEL across four Java applications and five business KPIs within Alibaba illustrates its capability to pinpoint the primary cause of exception logs with an AC@5 (top-5 accuracy) exceeding 90%, effectively mitigating the exception fatigue problem within Alibaba. Furthermore, it can identify the root-cause exceptions in a real software failure within just 4 minutes, outperforming the manual investigation process by over an hour.
- Open Access
- 31 July 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
- Vol. 34 (6) , 1-49
TEESlice : Protecting Sensitive Neural Network Models in Trusted Execution Environments when Attackers Have Pre-Trained Models
Trusted Execution Environments (TEEs) are used to safeguard on-device models. However, directly employing TEEs to secure the entire DNN model is challenging due to the limited computational speed. Utilizing GPU can accelerate DNN’s computation speed but widely available commercial GPUs usually lack security protection. To this end, scholars introduce TEE-Shielded DNN Partition (TSDP), a method that protects privacy-sensitive weights within TEEs and offloads insensitive weights to GPUs. Nevertheless, current methods do not consider the presence of a knowledgeable adversary who can access abundant publicly available pre-trained models and datasets. This article investigates the security of the existing methods against such a knowledgeable adversary and reveals their inability to fulfill their security promises. Consequently, we introduce a novel partition before training strategy, which effectively separates privacy-sensitive weights from other components of the model. Our evaluation demonstrates that our approach can offer full model protection with a computational cost reduced by a factor of 10. In addition to traditional CNN models, we also demonstrate the scalability to large language models. Our approach can compress the private functionalities of the large language model to lightweight slices and achieve the same level of protection as the shielding-whole-model baseline.
- 31 July 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
- Vol. 34 (6) , 1-49
Open AccessTEESlice : Protecting Sensitive Neural Network Models in Trusted Execution Environments when Attackers Have Pre-Trained Models
Trusted Execution Environments (TEEs) are used to safeguard on-device models. However, directly employing TEEs to secure the entire DNN model is challenging due to the limited computational speed. Utilizing GPU can accelerate DNN’s computation speed but widely available commercial GPUs usually lack security protection. To this end, scholars introduce TEE-Shielded DNN Partition (TSDP), a method that protects privacy-sensitive weights within TEEs and offloads insensitive weights to GPUs. Nevertheless, current methods do not consider the presence of a knowledgeable adversary who can access abundant publicly available pre-trained models and datasets. This article investigates the security of the existing methods against such a knowledgeable adversary and reveals their inability to fulfill their security promises. Consequently, we introduce a novel partition before training strategy, which effectively separates privacy-sensitive weights from other components of the model. Our evaluation demonstrates that our approach can offer full model protection with a computational cost reduced by a factor of 10. In addition to traditional CNN models, we also demonstrate the scalability to large language models. Our approach can compress the private functionalities of the large language model to lightweight slices and achieve the same level of protection as the shielding-whole-model baseline.