The world is abuzz right now with different kinds of generative AIs. While these AIs have become popular because they can create amazing images, text and artwork, it turns out they are also really useful for science. In this blog, I describe what generative AI is, and different ways that it is already being used for science.
Generative AI refers to a subset of artificial intelligence techniques focused on generating new data or content, such as images, music, text, and videos, based on learned patterns and knowledge. The goal of generative AI is to create outputs that are novel, diverse, and creative, while still being coherent and realistic. This is achieved through various algorithms, such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and others.
There are several types of generative AI, including:
Generative Adversarial Networks (GANs): GANs consist of two neural networks: a generator and a discriminator. The generator creates fake data, while the discriminator tries to distinguish the fake data from real data. The two networks are trained together to produce realistic outputs.
Variational Autoencoders (VAEs): VAEs are a type of generative model that learns a compact representation of the data, called the latent space, and can generate new samples by sampling from this latent space.
Autoregressive Models: Autoregressive models generate new data one step at a time, based on the previous data points. An example of this is a Generative Pre-trained Transformer (GPT) language model, which generates text based on previous words.
Transformer-based Models: Transformer-based models, such as the Generative Transformer (GT), generate data by transforming an input noise vector into an output, such as an image or a piece of music.
Flow-based Models: Flow-based models, such as Normalizing Flow Models, transform a simple random noise distribution into a target data distribution, allowing for efficient sampling and generation of new data.
These are some of the most common types of generative AI, but there are many other techniques and models that fall under this umbrella.
What can generative AIs be used for in science?
Drug Discovery: Generative AI models can be trained on large molecular databases to predict the properties of new molecules and assist in drug discovery.
Material Science: Generative AI can be used to design new materials with specific properties, such as high thermal conductivity or low density.
Climate Modeling: Generative AI can be used to generate simulation models of the Earth's climate, helping scientists understand and predict the effects of climate change.
Astrophysics: Generative AI models can be used to simulate the formation and evolution of galaxies and stars, improving our understanding of the universe.
Biology: Generative AI can be used to design new proteins with specific properties, such as increased stability or specificity for a certain target.
These are just a few examples of how generative AI is being used in science to generate new knowledge and advance various fields of research.
Check back soon for more details on each of these methods! Meanwhile, enjoy the image above - created by Stable Diffusion - one of the most recent forms of generative AI.
Comments