The video explores vulnerabilities in various machine learning tasks, including classification, regression, generative models, clustering, dimensionality reduction, and reinforcement learning. It highlights how adversarial attacks can exploit these vulnerabilities, affecting tasks like spam detection, malware analysis, and face detection. The video concludes that all categories of machine learning tasks are susceptible to such attacks, emphasizing the need for safety considerations.

What is a common vulnerability found in classification algorithms?

• They cannot be used for spam detection.

• They are immune to adversarial attacks.

• They are only vulnerable to physical attacks.

• They have conceptual architecture vulnerabilities.

Which machine learning task is mentioned as being susceptible to adversarial examples, similar to classification?

• Clustering

• Regression

• Dimensionality Reduction

• Reinforcement Learning

What is a principal application of generative models that can be exploited by adversarial attacks?

• Data classification

• Input reconstruction

• Feature selection

• Q function approximation

How can PCA-based classifiers be compromised?

• By using soft thresholds

• By applying gradient descent

• By contaminating training data with outliers

• By using non-differentiable functions

What is a simple method to attack reinforcement learning algorithms?

• By observing the current state and applying adversarial perturbations

• By using non-differentiable functions

• By contaminating training data

• By using soft thresholds