Publications by authors named "Dmitriy Slutskiy"

Article Synopsis
  • The paper enhances the junction tree variational autoencoder (JT VAE) by improving its implementation and utilizing its internal feature space for various tasks.
  • It demonstrates that pretraining the JT VAE on a large dataset, specifically the ZINC database, and subsequent optimization with a regression model allows for multitasking capabilities like prediction, generation, and optimization.
  • The study evaluates the model using the QM9 dataset, showing that it outperforms existing models in generating new molecules and optimizing molecular properties while maintaining high precision.
View Article and Find Full Text PDF