Researchers at the Indian Institute of Science (IISc) and University College London (UCL) have developed a groundbreaking machine learning method to predict material properties using limited data. This innovative approach leverages transfer learning to enhance material discovery, particularly in the field of semiconductors, and supports India’s semiconductor manufacturing initiatives.

The research team, led by Assistant Professor Sai Gautam Gopalakrishnan from IISc’s Department of Materials Engineering, utilized a model based on Graph Neural Networks (GNNs) to address the challenge of limited data on material properties. Testing materials is often expensive and time-consuming, making it difficult to gather sufficient data to train machine learning models.

In their study, the researchers employed a technique called Multi-property Pre-Training (MPT), where the model is first pre-trained on a large dataset and then fine-tuned to adapt to a smaller target dataset. This method allows the model to learn from various material properties simultaneously during the initial training phase and then focus on the target property data.

The team’s transfer learning-based model demonstrated superior performance compared to models trained from scratch. It successfully predicted band gap values for 2D materials that it had never encountered during training. This capability is particularly promising for battery development, as it can predict ion movement within electrodes.

Gopalakrishnan highlighted the potential of this technology to contribute significantly to India’s semiconductor manufacturing efforts by predicting material defects. The research also shows promise for other applications, such as improving energy storage devices by predicting how quickly ions can move within electrodes.

The collaboration between IISc and UCL exemplifies the power of interdisciplinary research in advancing scientific knowledge and addressing global challenges.

  IISc, machine learning, research, ucl