DP-VOXLET: Provable Speaker Anonymization for Disentangled Speech Representations
本文提出了一种基于差分隐私的说话人匿名化方法DP-VOXLET，用于在保持语义内容不变的同时，为解缠语音表示提供可证明的说话人隐私保护。
0 citationsRead paper
Selected work
Representative Papers
Lazy Arithmetic using Systolic Arrays for Closing the Verification Gap on Embedded Systems
Deploying deep neural networks on resource-constrained embedded systems presents significant challenges in simultaneously achieving energy efficiency, correctness, and security—particularly in safety-critical domains such as healthcare, where models are vulnerable to fault injection attacks and lack dynamic accuracy guarantees. To address this, this work proposes a hardware-software co-design approach. On the software side, it introduces the first adaptive precision quantization method that jointly ensures real-time operation, dynamic adjustability, and formal correctness, leveraging sensitivity analysis to stabilize decision boundaries. On the hardware side, it designs a customized systolic array supporting left-to-right, most-significant-bit (MSB)-first computation to enhance resilience against bit-flip errors. Experimental results demonstrate the feasibility of the proposed framework, with preliminary findings indicating substantial improvements in both reliability and resource efficiency for edge AI deployments.
0 citationsRead paper
Trustworthy Runtime Verification via Bisimulation (Extended Experience Report)
This work proposes CopilotVerifier, an automated verification framework designed to enhance the correctness and trustworthiness of runtime monitoring code in safety-critical systems by complementing the Copilot compiler. CopilotVerifier is the first to decompose the bisimulation relation between source programs and their compiled C code into verifiable conditions. By integrating symbolic execution (via Crucible) with SMT solving (through What4), the framework automatically generates formal proofs that guarantee semantic equivalence—ensuring identical outputs and consistent crash behaviors under equivalent inputs. This approach significantly strengthens compiler assurance with modest computational overhead and lays the groundwork for producing human-auditable formal arguments of correctness.
0 citationsRead paper
CNnotator: LLM-Guided Memory Safety Annotation Synthesis
This work addresses the persistent challenge of memory safety errors in C programs, which stem from implicit memory usage semantics and the high cost of manual annotation. The authors propose a novel approach that leverages large language models—specifically OpenAI o3 and GPT-4o—to automatically generate formal memory safety annotations. These annotations are then integrated into CN, a hybrid testing and verification framework, enabling end-to-end validation. The method demonstrates substantial improvements in both efficiency and accuracy: OpenAI o3 achieves a 90% success rate on first-attempt annotation generation (97% overall), while GPT-4o attains 65%. These results substantiate the feasibility and practical promise of using AI to assist in rigorous memory safety analysis for low-level code.
0 citationsRead paper
FVSpec: Real-World Property-Based Tests as Lean Challenges
This work addresses the lack of effective evaluation benchmarks for AI models in real-world software formal verification. It presents the first large-scale effort to automatically translate real-world Python property-based tests (PBT) into formal specifications in Lean 4, establishing a reproducible benchmark. By integrating a multi-agent large language model pipeline with Python semantic modeling and dependent type programming techniques, the authors successfully generated 9,415 Lean 4 specifications corresponding to 2,772 tests derived from 11,039 original PBTs. The entire codebase and dataset are publicly released. This contribution substantially advances the practical application of AI-assisted formal verification in real software systems.
0 citationsRead paper
Recent publications
Latest Papers
DP-VOXLET: Provable Speaker Anonymization for Disentangled Speech Representations
本文提出了一种基于差分隐私的说话人匿名化方法DP-VOXLET，用于在保持语义内容不变的同时，为解缠语音表示提供可证明的说话人隐私保护。
0 citationsRead paper
Lazy Arithmetic using Systolic Arrays for Closing the Verification Gap on Embedded Systems
Deploying deep neural networks on resource-constrained embedded systems presents significant challenges in simultaneously achieving energy efficiency, correctness, and security—particularly in safety-critical domains such as healthcare, where models are vulnerable to fault injection attacks and lack dynamic accuracy guarantees. To address this, this work proposes a hardware-software co-design approach. On the software side, it introduces the first adaptive precision quantization method that jointly ensures real-time operation, dynamic adjustability, and formal correctness, leveraging sensitivity analysis to stabilize decision boundaries. On the hardware side, it designs a customized systolic array supporting left-to-right, most-significant-bit (MSB)-first computation to enhance resilience against bit-flip errors. Experimental results demonstrate the feasibility of the proposed framework, with preliminary findings indicating substantial improvements in both reliability and resource efficiency for edge AI deployments.
0 citationsRead paper
Trustworthy Runtime Verification via Bisimulation (Extended Experience Report)
This work proposes CopilotVerifier, an automated verification framework designed to enhance the correctness and trustworthiness of runtime monitoring code in safety-critical systems by complementing the Copilot compiler. CopilotVerifier is the first to decompose the bisimulation relation between source programs and their compiled C code into verifiable conditions. By integrating symbolic execution (via Crucible) with SMT solving (through What4), the framework automatically generates formal proofs that guarantee semantic equivalence—ensuring identical outputs and consistent crash behaviors under equivalent inputs. This approach significantly strengthens compiler assurance with modest computational overhead and lays the groundwork for producing human-auditable formal arguments of correctness.
0 citationsRead paper
CNnotator: LLM-Guided Memory Safety Annotation Synthesis
This work addresses the persistent challenge of memory safety errors in C programs, which stem from implicit memory usage semantics and the high cost of manual annotation. The authors propose a novel approach that leverages large language models—specifically OpenAI o3 and GPT-4o—to automatically generate formal memory safety annotations. These annotations are then integrated into CN, a hybrid testing and verification framework, enabling end-to-end validation. The method demonstrates substantial improvements in both efficiency and accuracy: OpenAI o3 achieves a 90% success rate on first-attempt annotation generation (97% overall), while GPT-4o attains 65%. These results substantiate the feasibility and practical promise of using AI to assist in rigorous memory safety analysis for low-level code.
0 citationsRead paper
FVSpec: Real-World Property-Based Tests as Lean Challenges
This work addresses the lack of effective evaluation benchmarks for AI models in real-world software formal verification. It presents the first large-scale effort to automatically translate real-world Python property-based tests (PBT) into formal specifications in Lean 4, establishing a reproducible benchmark. By integrating a multi-agent large language model pipeline with Python semantic modeling and dependent type programming techniques, the authors successfully generated 9,415 Lean 4 specifications corresponding to 2,772 tests derived from 11,039 original PBTs. The entire codebase and dataset are publicly released. This contribution substantially advances the practical application of AI-assisted formal verification in real software systems.
0 citationsRead paper