! S15: Ethical Issues

Question 1

Ethical issues AI

Answer 1

• Privacy & data protection e.g. Lack of privacy, Misues of personal data
• Transaprency e.g. bias & discrimination, lack of accurate recommendation
• Reliability e.g. Lack of quality or accurate data
• Safety

Question 2

High-risk AI systems according to EU Parliament

Answer 2

• unarmed aircrat
• autonomous verhicles & robots

Question 3

AI stakeholder

Answer 3

• Policy
• Organisations
• Individuals

Question 4

Deep Generative Models

Answer 4

• efficient way to analyse & understand unlabeled data
• captures inner probabilistic distribution that generates class of data to generate similar data
• aims learn true data distribution of trading set as to generate new datapoints with some variation
• NN to learn a function that approximates distribution

Question 5

Deep Generative Models - Categories

Answer 5

• Cost function-based models
• Energy-based models

Question 6

Deep Generative Models - Cost function-based models

Answer 6

• autoencoders & generative adversial network

Question 7

Deep Generative Models - energy-based models

Answer 7

• where joint probability is defined using energy function
• Boltzman Machines

Question 8

Boltzmann Machiens

Answer 8

• fully connected ANN Architecture
• based on stochastic neurons: output 0 or 1
• probability function based on Boltzmann distribution
• just has visible & hidden layers (no output)
• nodes (=states) interconnected: exchange info among themselves & generate new data
• learn all parameters on their own (no GD, backpro.)