DeepDream

DeepDream is a computer vision program created by Google researchers in 2015 that uses a convolutional neural network to transform images into dream-like hallucinations. Key facts:

• Works by over-activating features detected in a CNN model.
• Causes amplification of patterns in images fed through the net.
• Creates psychedelic, surreal imagery with objects and animals.
• Provides insight into how neural nets identify features.
• Shows capabilities and limitations of current AI systems.
• Developed as part of Google's Inceptionism project.
• Inspired creative exploration of AI and machine perception.
• Critiqued for questionable anthropomorphizing of models.
• Demonstrates risks of overinterpreting model hallucinations.

While evocative, DeepDream does not actually provide a meaningful window into neural network representations. But it strongly influenced AI art and perceptions of deep learning capabilities.

See also:

• Neural network
• Deep learning
• Computer vision
• Generative adversarial network