Mini Conference

Adversarial Attacks against Deep Learning

Research Question: Is Deep Learning secure for Robots?

Real-time White-box Attack (20ms)

Project 1: Adversarial Filter
Project 2: Adversarial Detection

Distributed Black-box Attack (20s)

Project 3: Adversarial Classification

Exeter Trustworthy AI Lab

Hi, I would like to introduce adversarial attacks against deep learning. Basically, I'm trying to seek the answer for the question: Is Deep Learning secure for Robots? To answer the question, I would like to present real-time White-box attack and distributed Black-box attack. White-box attack means we know all the details about the model, including the input, the output, the structure of the model, virtually eveything. While for black-box attack, we only know the input and output, and nothing else. For White-box attack, I prepared two projects, adversarial filter and adversarial detection that attack object detection system in real time. For black-box attack, I prepared one project, adversarial classification that attacks image classification system.

Background

Is Deep Learning secure for Robots?

Intelligent Robots: Deep Learning in Robotics

Deep Learning for Autonomous Driving

Adversarial attacks against image classification

Adversarial attacks against object detection

Project 1: Adversarial Filter

Real-time White-box Attack

Project 2: Adversarial Detection

Real-time White-box Attack

Project 3: Adversarial Classification

Distributed Black-box Attack

Is Deep Learning secure for Robots?

Thanks

https://mini.wuhanstudio.uk