Unsupervised Doodling and Painting with Generative Agents

Explore the intersection of machine learning and creative endeavors in this talk by Dr. S. M. Ali Eslami, Research Scientist at Google DeepMind. Delve into the use of reinforcement learning agents as generative models for image creation, examining how computational techniques can be applied to artistic pursuits. Learn about the training process involving a simulated painting environment and a discriminator network, and discover how constrained generative agents can produce visually abstract images without external supervision. Gain insights into the potential emergence of human-like drawing abilities through simulated embodiment, and consider the implications for the future of machine learning in art and design. Investigate recent advances in reinforcement learning, generative models, and topological data analysis for media creation, design exploration, and urban analytics. The talk covers a wide range of topics, including the history of AI, computer vision, machine learning examples, reinforcement learning applications, cultural significance of AI in games like Go, optimization and generation techniques, human creativity, style transfer, realistic generation, and the philosophical debate surrounding computer-generated art.

Intro
A brief history of AI
Computers
Programmable Computers
Computer Vision
Machine Learning
Examples
Data
reinforcement learning
example of reinforcement learning
Go
Cultural significance
Go schools
Complexity of Go
Media interest
Alphago
Optimization and Generation
Optimization
Generation
Infinite Variety
Human Creativity
Style Transfer
Realistic Generation
Example
Constraint Generation
Benefits of Learning
Training on Human Faces
Intermediate Steps
Discount Factor
Agent Drawing
Restricted Agents
Drawing Faces
Is art just taking painters input
Drawing a parallel to photography
Conceptual art vs abstract art
Can computers create art
The mimetic generation
Did people learn something
The significance of the move
The philosophical debate
Visual representation
Deepfakes