The Polymer Society of Korea

Abstract

Developing functional polymers for sustainable technologies demands efficient computational approaches to minimize resource-intensive experiments. This presentation introduces our AI-driven framework integrating novel polymer representations with machine learning for accelerated materials discovery. We present HAPPY (Hierarchically Abstracted rePeat unit of PolYmers), which transforms complex polymer structures into functional subgroups with simplified connectors, reducing representation complexity while preserving essential structural information. This enables deep learning models to achieve high prediction accuracy with limited training data. Building upon HAPPY, IM-HAPPY employs genetic algorithms or generative models for inverse design of polymers with target properties, enabling exploration of novel candidates including carbon-based materials. Our approach significantly reduces experimental burden, contributing to sustainable materials innovation.