Adversarial Input Detection Using Image Processing Techniques (IPT)

Adversarial Input Detection Using Image Processing Techniques (IPT)

• 1.
  Adversarial Input DetectionUsing Image Processing Techniques (IPT) Authors: • Kishor Datta Gupta • Dipankar Dasgupta • Zahid Akhtar Presented by : Kishor Datta Gupta
• 2.
  Adversarial Attack (AA)on AI/ML Types: • Poisoning Attack : Manipulate training data • Evasion Attack: Manipulate input data • Trojan AI : Manipulate AI Architecture (example: Changes weights value) “Manipulation of training data, Machine Learning (ML) model architecture, or manipulate testing data in a way that will result in wrong output from ML”
• 3.
  Different Adversarial Traits DifferentAttack method has different types of noise/manipulation style
• 4.
  Detection of AdversarialTraits(1) Clean and adversarial images have quantifiable noise difference
• 5.
  Detection of AdversarialTraits(2)
• 6.
  Detection of AdversarialTraits(3)
• 7.
  Methodology
• 8.
  Methodology(2)
• 9.
  Comparison
• 10.
  Summary Work against adaptiveattack Don’t reduce ML efficiency Can identify attack type Applicable for cross-platform Work for both blackbox and whitebox attack Our defense has below properties:
• 11.
  Q/A