2021/05/20 by Jack W. Stokes, Stokes, Jack W., P. England +3
Computer Science · Decision Sciences · #Advanced Malware Detection Techniques #Cryptography and Security (cs.CR) #Digital and Cyber Forensics #FOS: Computer and information sciences #Scientific Computing and Data Management
paper · pdf · doi:10.48550/arxiv.2105.10051
openalex publication_date 2021/05/20 · openalex created_date 2022/10/01 · openalex updated_date 2026/07/28
Recent research has successfully demonstrated new types of data poisoning\nattacks. To address this problem, some researchers have proposed both offline\nand online data poisoning detection defenses which employ machine learning\nalgorithms to identify such attacks. In this work, we take a different approach\nto preventing data poisoning attacks which relies on cryptographically-based\nauthentication and provenance to ensure the integrity of the data used to train\na machine learning model. The same approach is also used to prevent software\npoisoning and model poisoning attacks. A software poisoning attack maliciously\nalters one or more software components used to train a model. Once the model\nhas been trained it can also be protected against model poisoning attacks which\nseek to alter a model's predictions by modifying its underlying parameters or\nstructure. Finally, an evaluation set or test set can also be protected to\nprovide evidence if they have been modified by a second data poisoning attack.\nTo achieve these goals, we propose VAMP which extends the previously proposed\nAMP system, that was designed to protect media objects such as images, video\nfiles or audio clips, to the machine learning setting. We first provide\nrequirements for authentication and provenance for a secure machine learning\nsystem. Next, we demonstrate how VAMP's manifest meets these requirements to\nprotect a machine learning system's datasets, software components, and models.\n