TransPolymer: Polymer Property Predictions with Transformers

This coursework explores TransPolymer, a framework for polymer property prediction. It covers challenges in polymer representation, introduces a chemically-aware tokenization and transformer-based architecture, and explains pretraining and fine-tuning techniques. The framework's evaluation demonstrates performance on benchmarks, advantages of self-attention, and findings on data augmentation. Ablation studies analyze dataset size, encoder strategies, and augmentation effects. The conclusion...