Fig. 3: Conditional generation on ionic conductivity and functional group.
From: De novo design of polymer electrolytes using GPT-based and diffusion-based generative models
a Oversampling to create a balanced dataset. Blue and green histograms are the distributions of high ion conductive polymers before and after oversampling. b Example monomer structures generated by 3 models during conditional generation task. c Shift of conductivity distributions comparing train set and generated set. d Evaluation metrics of model performance. e Conditional generation with target functional groups. The carbonate group is an unwanted functional group, while the ether group is a desirable one. f Frequency of functional groups with conditional constraints. The value of each axis represents the fraction of ether and carbonate groups among all functional groups inside the monomer.
