Cover image for Is Building Secure ML Possible?

Is Building Secure ML Possible?

Author Catherine Nelson
Release Date: 2021/10/01
ISBN: 9781098107338
Topic:
0%
8
Chapters
0-1
Hours read
0k
Total Words
    Start Reading Now    
       Add to Wishlist       
View table of contents

Security isn't considered a high priority when it comes to machine learning systems. But given the speed of innovation in this area, the rapid advances in ML present a whole new set of security risks that are quite different from those of traditional software. This report reviews known security risks for ML systems and examines why security in this area is particularly important today.

Catherine Nelson, principal data scientist at SAP Concur, describes techniques to enhance security, increase privacy, and mitigate attacks that do occur on ML systems. By defining what's meant by secure, she examines whether the techniques now available are sufficient to achieve true security in ML systems. This report is ideal for ML engineers, data scientists, and managers of ML teams.

  • Learn key points in the machine learning lifecycle when security becomes particularly important
  • Get an overview of known security risks, including transfer learning, model theft, model inversion, and membership inference attacks
  • Mitigate security risks using audits and governance, model monitoring, data checks and balances, and general security practice

Table of Contents

  1. 1. Introduction
    1. Who Is This Report For?
    2. What Do We Mean by “Secure” ML?
    3. What Security Standards and Regulations Apply to ML?
    4. Why Is Security for ML Different?
    5. When Should You Think About Security?
  2. 2. Known Security Risks of ML Systems
    1. Adversarial Training Data
    2. Evasion Attacks
    3. Membership Inference Attacks
    4. Model Inversion Attacks
    5. Model Theft
    6. Transfer Learning Attacks
    7. Summary
  3. 3. Techniques to Mitigate Security Risks
    1. General Security Practices
    2. Data Checks and Balances
    3. Model Monitoring
    4. Audits and Governance
    5. Differential Privacy
    6. Encrypted ML
    7. Summary
  4. 4. The Future of Secure ML
    1. Risks That There Are No Current Solutions For
    2. Risks That Haven’t Yet Manifested
    3. Is Building Secure ML Possible?
18.223.172.252