Imagine if we could predict the properties of materials without having to test them in a lab. This would save a lot of time and money, and it could help scientists discover new materials with amazing properties, like super-strong metals or super-efficient semiconductors. This is exactly what researchers at the Indian Institute of Science (IISc) and University College London are working on. They are using machine learning tools to predict material properties, even when there is limited data available.
Machine learning is a branch of Artificial Intelligence, where computers learn to make decisions or predictions without being directly programmed. It works by analyzing data, finding patterns, and improving over time. For example, it helps apps suggest music, improve photos, or recognize your voice for commands like “play a song.” In this research, the scientists used a specific type of machine learning called transfer learning, wherein, a large model is first trained on a big dataset to learn general patterns. Then, this model is fine-tuned on a smaller, specific dataset to make accurate predictions about new data.
The researchers used a special kind of machine learning model called a Graph Neural Network (GNN). GNNs are great for working with data that can be represented as graphs, like the three-dimensional structures of materials. In these graphs, atoms are represented as nodes, and the bonds between them are edges. This allows the model to understand the complex relationships between atoms in a material.
The researchers found that their transfer learning model performed much better than models that were trained from scratch. They also developed a framework called Multi-property Pre-Training (MPT), where the model was pre-trained on seven different material properties at once. This approach was so effective that the model could even predict properties of materials it hadn't been specifically trained on, like the band gap of 2D materials.
The researchers also plan to use their model to predict how quickly ions can move within electrodes in a battery. This could lead to the development of better energy storage devices, which are crucial for renewable energy technologies.
This work could allow scientists to predict material properties even when there is a dearth of data. Testing materials in a lab is often expensive and time-consuming, so being able to make predictions with limited data is a huge advantage, especially when the predictions are unpredictable yet accurate. This can speed up the discovery of new materials with desirable properties, be it better semiconductors for electronics or more efficient materials for energy storage.
This research news was partly generated using artificial intelligence and edited by an editor at Research Matters.