Towards Deep Learning Models Resistant to Adversarial Attacks

Explore the critical topic of adversarial attacks on deep learning models in this 23-minute Launchpad video. Delve into the paper "Towards Deep Learning Models Resistant to Adversarial Attacks" and understand the process of generating adversarial attacks, their implications, and underlying causes. Examine the problem definition, attack methodology, and experimental results using various datasets and dimensions. Analyze the effects of network capacity and training data on model vulnerability. Compare accuracy across different training methods and sources. Gain valuable insights into developing more robust deep learning models that can withstand adversarial attacks.

Intro
Generating an Adversarial Attack
Concerns of Adversarial Attacks
Why Do These Attacks Happen?
Paper: Problem Definition
Defining an Attack
Experimentation: Dataset and Dimensions
Loss during 20 projected gradient descent runs
Network Capacity Effect - By Training Data
Accuracy by training method across 3 sources
Conclusions