Loss functions are the processes that tell a machine learning network, during training, if it’s getting any better at making predictions. This article looks at the broad current landscape of loss functions, and some of the new trends that are emerging, such as a greater reliance on human-informed evaluation of images.
Semantic segmentation is a computer vision tool that can recognize and isolate elements in an image. In the age of the multimodal generative system, such as Stable Diffusion, it’s now being used in new and unforeseen ways.
A look at 3D Morphable Models, a 20th-century CGI technology that’s been adopted lately for cutting-edge image synthesis research.
A look at the phenomenon of overfitting, where a machine learning model becomes so ‘obsessed’ with particular parts of the training data that it will not generalize to unseen data, and produces a non-usable system.
What are features in machine learning? They might be facial features, number arrangements, or any disposition of the data that’s been trained in a machine learning model. This article takes a look at features, and how they relate to image synthesis.
This article examines what the term ‘weights’ means in the context of machine learning systems.
An overview of ‘fine-tuning’ – the practice of taking existing trained machine learning models and continuing to train them on new data.
An overview of the functioning of deepfake autoencoders, as exemplified by the DeepFaceLab and FaceSwap repositories.
OpenAI’s CLIP, one of the cornerstone technologies behind Stable Diffusion, DALL-E 2 and other image synthesis architectures, is sweeping the generative AI space at the moment. This article examines how CLIP works, what its advantages are, and where, in some respects, its innovative principles could cause issues.
The latent space is the ‘subconscious’ and overarching understanding of relationships between learned data points that a machine learning system has been able to derive from the information that it gets fed. This article takes a detailed look at what can be achieved by targeting content that’s been trained into the latent space of a machine learning model.
