Molecular functionalization of porphyrins opens countless new opportunities in tailoring their physicochemical properties for light-harvesting applications. However, the immense materials space spanned by a vast number of substituent ligands and chelating metal ions prohibits high-throughput screening of combinatorial libraries. In this work, machine-learning algorithms equipped with the domain knowledge of chemical graph theory were employed for predicting the energy gaps of >12 000 porphyrins from the Computational Materials Repository. Among a variety of graph-based molecular descriptors, the electrotopological-state index, which encodes electronic and topological structure information, captures the energy gaps of porphyrins with a prediction RMSE < 0.06 eV. Importantly, feature sensitivity analysis suggests that the carbon structural motif in methine bridges connected to the anchor group predominantly influences the energy gaps of porphyrins, consistent with the spatial distribution of their frontier molecular orbitals from quantum-chemical calculations.

Super-ResDenoising Inpainting Dehazing Translation Fig. 1: Applications. We propose to learn deep image prior using a neural architecture search. The resulting network can be applied to solve various inverse image problems without training the model with a large-scale dataset with ground truth. Through extensive experimental evaluations, we show that our model compares favorably against existing hand-crafted CNN models for learning-free image restoration tasks and in some cases even reaches competitive performance when compared with recent learning-based models.

No abstract

Long-tailed class distributions are prevalent among the practical applications of object detection and instance segmentation. Prior work in long-tail instance segmentation addresses the imbalance of losses between rare and frequent categories by reducing the penalty for a model incorrectly predicting a rare class label. We demonstrate that the rare categories are heavily suppressed by correct background predictions, which reduces the probability for all foreground categories with equal weight. Due to the relative infrequency of rare categories, this leads to an imbalance that biases towards predicting more frequent categories. Based on this insight, we develop DropLoss -- a novel adaptive loss to compensate for this imbalance without a trade-off between rare and frequent categories. With this loss, we show state-of-the-art mAP across rare, common, and frequent categories on the LVIS dataset. Codes are available at https://github.com/timy90022/DropLoss.

Long-tailed class distributions are prevalent among the practical applications of object detection and instance segmentation. Prior work in long-tail instance segmentation addresses the imbalance of losses between rare and frequent categories by reducing the penalty for a model incorrectly predicting a rare class label. We demonstrate that the rare categories are heavily suppressed by correct background predictions, which reduces the probability for all foreground categories with equal weight. Due to the relative infrequency of rare categories, this leads to an imbalance that biases towards predicting more frequent categories. Based on this insight, we develop DropLoss -a novel adaptive loss to compensate for this imbalance without a trade-off between rare and frequent categories. With this loss, we show state-of-the-art mAP across rare, common, and frequent categories on the LVIS dataset. Codes are available at https://github.com/timy90022/DropLoss.