Oliver Gardi
Class of 2027
Class of 2027
All things on the planet like the air we breathe and matter you can hold is made up of small particles called atoms. These atoms are incredibly small, with 1 quarter containing 1023 atoms. Because of their size, doing experiments where atoms are being measured is very difficult. Instead, researchers can tell computers to simulate a group of molecules and measure that. For example, if researchers wanted to know how strong a substance is without having to make it, they would simulate a stressful environment and see if the substance broke. The problem with these simulations though is how difficult it is for computers to run them. Even basic simulations take days to run. To fix this problem, scientists use machine learning models, which are programs that can find patterns in data and use them to make predictions on similar data very quickly. An example of a machine learning model is chatGPT. Machine learning models can look at molecular data to speed up simulations.
Researchers have created a ML model designed to avoid having to run the simulation at all. Since the models are so much faster than simulations, they can save researchers lots of time. The model uses the input and output of previous simulations and learns from them, in order to predict outputs without having to run a multi-day simulation. But the problem with this type of model is there is no way to guarantee a given prediction is accurate. This is because the model the researchers used does not calculate potential error. To fix this, a different type of machine learning can be used. This model will tell researchers if the predictions can be trusted, fixing the issue identified above.