Generative Adversarial Networks, popularly known as GANs are a specific form of Artificial Intelligence algorithms – the neural network machine learning model in particular, which generate realistic looking images by using two opposing neural networks. Both networks are untrained at the start but learn from each other over time to build a really adaptive AI learning model.
GANs was introduced by Ian Goodfellow in 2014. GAN’S works by contesting two neural network in a zero-sum framework whereby a network generates models and functions as a generator while the other evaluates those models in the role of a discriminator/classifier. They are mostly used in unsupervised machine learning
GANs is fast becoming the next best thing in deep learning as it is changing the way we think about Artificial Intelligence. A major application of GANs would include Text to Image Algorithm that generates an image based on text provided; Data Creation which creates synthetic data used for training machine learning models. A more familiar application would be the popular AlphaGo – Google’s Deepmind AI who outperformed the world’s number one Go player Lee Sedol at the game of Go recently in China. GANs can also be used in predicting drugs for treating a particular ailment.
The opportunities are boundless for the innovative ways GANs can be improvised. What really sets GANs apart is its ability to learn both from interactions with humans and also challenge itself to make it less vulnerable. The results are mind-blowing as well. We can only imagine the many more innovations that’ll emerge from this invention that is literally changing the world!
Photo Credit: KDNuggets
