Recognition Mode

All neurons in the neural network are in recognition mode. The neural network in this mode can associate a data (image, text or search queries) to one of the predefined outcomes (labels, numerical prediction or ranking score).

Example tasks in Recognition mode:

• Classification (label as output)
• Regression (label as output)
• Search and recommendation (ranking score as output)

Creation Mode

All neurons in the neural network are in generation mode. The neural network in this mode can create synthetic data from an random input.

Example tasks in Creation mode:

• Image generation
• Video generation
• Text generation
• Speech generation
• User profile generation for privacy

Infinite Mode

Our boltzmann neurons, a.k.a. Zbit, can be in both recognition and creation mode. With the Zbit activated, the rest of the network can activate in a mixture mode of recognition and creation.

Without further retraining, the network can adapts the activation of neurons to the need of different tasks in generation and recognition.

Example tasks in Infinite mode:

• All recognition and generation tasks
• Interactive data generation
• Classification on incomplete data