Computer Vision & Multimodal Threat Intelligence
Adversarial patch attacks on vision transformers (ViT), multimodal prompt injection, and synthetic deepfake detection pipelines.
Course Overview & Objectives
Multimodal models accept images, audio, and video alongside text. Discover how adversarial optical illusions trick Vision Transformers, embed hidden prompt instructions within visual textures, and construct deepfake forensic watermarking pipelines.
What You Will Master
- Generate physical and digital adversarial patches that fool YOLO and Vision Transformers
- Embed visual prompt injection payloads into images that alter multimodal LLM outputs
- Implement frequency-domain spectral analysis to detect synthetic AI deepfakes
- Deploy robust provenance watermarking with C2PA open metadata standards
Prerequisites
- Python & PyTorch basics
- Understanding of convolutional or transformer architectures
Platforms & Tools Covered
Detailed Curriculum Modules
1 modules structured from foundational theory through complex adversarial execution.
Vision Transformer Security & Adversarial Patches
Attention map disruption, universal adversarial perturbations, and patch generation.
Hands-on Virtual Sandbox Labs
Zero local hardware dependencies. Provisioned in cloud containers via browser terminal.
Visual Prompt Injection via Steganography
Hide prompt instructions inside image pixel noise that commands GPT-4V to output attacker secrets.
Faculty & Lead Instructor
Direct weekly instruction, live office hours, and code-review feedback.
Piya Kohli
Thread Security EducationMultimodal AI Scientist
Leading computer vision research on adversarial image robustness and synthetic media authentication.
Frequently Asked Questions
Everything you need to know about scheduling, cohort admissions, and lab access.
Do we use GPUs for labs?
Yes! Cloud sandbox environments include dedicated GPU instances for neural network training and inference.
Ready to Master Computer Vision & Multimodal Threat Intelligence?
Join the upcoming cohort. Seats are limited to maintain a high faculty-to-student ratio and rigorous sandbox feedback.