Machine Learning: Promising, but Perilous

What if the pre-trained model you downloaded has a hidden backdoor? Learn how to protect your ML systems from these inherited, transferable attacks.

#1about 5 minutes

The dual nature of machine learning's power

Machine learning's increasing power and accessibility, exemplified by complex tasks like panoptic segmentation, also introduces significant security vulnerabilities.

#2about 3 minutes

Accelerating development with transfer learning

Transfer learning allows developers to repurpose large pre-trained teacher models for specific tasks with minimal data and compute by fine-tuning a new student model.

#3about 2 minutes

How transfer learning's benefits create security risks

The core benefits of transfer learning, such as knowledge transfer and minimal training, directly create attack vectors for adversaries.

#4about 5 minutes

Exploring evasion and poisoning attacks in ML

Adversarial examples can fool models with subtle input changes (evasion), while poisoned data can insert hidden backdoors, with both risks amplified by transfer learning.

#5about 3 minutes

Integrating security with a pre-development risk assessment

Before writing code, perform a thorough risk assessment by defining security requirements, evaluating resource availability, and conducting threat modeling for your specific use case.

#6about 3 minutes

Selecting robust teacher models for secure transfer learning

Mitigate risks by choosing transparent and trustworthy teacher models and using robust models hardened through techniques like adversarial training.

#7about 1 minute

Fortifying student models to prevent transferred attacks

Strengthen your student model by fine-tuning all layers to diverge from the teacher model, using backdoor detection, and performing continuous stress testing.

#8about 2 minutes

Key resources for developing secure ML systems

Practical resources like the Adversarial Robustness Toolbox for developers and security principles from the National Cybersecurity Center can help you build more secure ML systems.