Generative Adversarial Networks
Generative AI refers to artificial intelligence algorithms that enable using existing content like text, audio files, or images to create new plausible content. In other words, it allows computers to abstract the underlying pattern related to the input, and then use that to generate similar content.
Generative AI is a new artificial model that generates images that are more realistic than any previous AI model. It is known as a Generative Adversarial Network or GAN. It is not just any GAN but GAN that progressively grows over time during the training face that is a revolutionary development. GAN was invented by an AI researcher at the University of Montreal in Canada just about three years ago. According to AI experts, it has huge potential because it can learn to mimic any distribution of data. That means it can be taught to create words that are literally similar to own any domain such as images, music, speech, and writing like a robotic artist. The model has very impressive outputs.
Speech synthesis systems are using them for image-to-image translation technics use them, image in painting is another popular use case. GAN is a type of generative algorithm. The other class of models is discriminative algorithms. Discriminative algorithms try to classify data given some set of features and predict a label or category to which that data belongs to. If we have all the words in an email, for example, a discriminative algorithm predicts whether the message is spam or not a spam. Spam is one of the labels. The bag of the words gathered from the email is the feature that constitutes the input data. When we express this mathematically, the label is called Y and the feature is X. The formulation probability of Y given X translates to the probability that the email is spam given the words it contains.
So discriminative algorithm map features to labels. The only concern with correlation. One way to think about generative algorithms though is that they do the complete opposite. Instead of predicting labels and features, they try to predict features given a certain label. Assuming, this email is spam, how likely are these features discriminative models care about the models Y and X. Generative models care about how you generate X.