Generative Adversarial Network (GAN)
A type of machine learning system in which two artificial neural networks play a turn-based game with each other that teaches them the statistical attributes of the training data. They're often used to generate imagery.
Graphics Processing Unit (GPU)
A type of computer chip similar to a Central Processing Unit (CPU). GPUs were designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer intended for output to a display device and are necessary for playing video games. Because of their computational qualities, they have also become the computer chip of choice for machine learning.
Feeding new data into a trained model to produce an output. You can think of training as learning a new skill and inference as applying that skill in a new situation. Inference is much computationally faster than training.
A single cycle of training. It may take many iterations of training (10,000s or 100,000s) for a model to produce useful outputs.
A type of artificial intelligence technique that utilizes algorithms and statistical models to perform a task without using explicit instructions, relying instead on pattern recognition and inference.
A machine learning or AI model is a statistical structure that starts off untrained and then through exposure to training data, becomes a trained model capable of generating outputs.
The type of information that is needed for a specific training process. This can be any type of data—images, text, even sounds! In Playform, we refer to them as Input sets.
The amount of time spent on training a model. This measured in minutes as well as iterations.
A number of images that are grouped together for convenience, e.g. they were uploaded at the same time, are thematically similar, or are visually similar. Each image collection must have a name. Playform users can create new image collections and re-use existing ones.
A group of images used to train an AI model, comprised of one or more image collections. Some processes like Creative Morph require two distinct input sets, which are used by the model in different ways. Other processes only accept one input set of images.
In a training process with two or more input sets, inspirations inform the shapes and contours of the results. In a training process with just one input set, all factors of this set will inform the results.
Formerly called "Aesthetics". In Creative Morph, results will iteratively change to resemble this set of images, in both style and color.
A format or pattern for training an AI model. A training process is defined by its input requirements, the kind of outputs it generates, and how it learns from the training data. Creative Morph and Freeform are examples of Training processes.