用约束求解生成测试用例，高效发现深度学习框架的内存错误
GPU-Fuzz: Finding Memory Errors in Deep Learning Frameworks
- 将算子参数建模为约束，用求解器生成边界测试用例
- 在三个主流框架中发现13个未知内存错误
- 适合框架开发者和安全研究人员排查底层缺陷
GPU内存错误是深度学习框架的重大威胁，可能导致崩溃甚至安全漏洞。我们提出GPU-Fuzz，通过将算子参数建模为形式化约束，利用约束求解器生成测试用例，系统性地探测GPU内核中的易错边界条件。该方法应用于PyTorch、TensorFlow和PaddlePaddle，共发现13个未知漏洞，验证了GPU-Fuzz在定位内存错误方面的有效性。
原文摘要 · Abstract (English)
GPU memory errors are a critical threat to deep learning (DL) frameworks, leading to crashes or even security issues. We introduce GPU-Fuzz, a fuzzer locating these issues efficiently by modeling operator parameters as formal constraints. GPU-Fuzz utilizes a constraint solver to generate test cases that systematically probe error-prone boundary conditions in GPU kernels. Applied to PyTorch, TensorFlow, and PaddlePaddle, we uncovered 13 unknown bugs, demonstrating the effectiveness of GPU-Fuzz in finding memory errors.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。