Kyung Hoon Lee | Searcharxiv

arXiv2023

Designing topological acoustic lattices via electroacoustic analogies

Topological acoustics has recently witnessed a spurt in research activity, owing to their unprecedented properties transcending typical wave phenomena. In recent years, the use of coupled arrays of acoustic chambers has gained popularity in designing topological acoustic systems. In their common form, an array of acoustic chambers with relatively large volume is coupled via narrow channels. Such configuration is generally modeled as a full three-dimensional system, requiring extended computational time for simulating its harmonic response. To this end, this paper establishes a comprehensive mathematical treatment of the use of electroacoustic analogies for designing topological acoustic lattices. The potential of such analytical approach is demonstrated via two types of topological systems: (i) edge states with quantized winding numbers in an acoustic diatomic lattice and (ii) valley Hall transition in an acoustic honeycomb lattice that leads to robust waveguiding. In both cases, the established analytical approach exhibits an excellent agreement with the full three-dimensional model, whether in dispersion analyses or the response of an acoustic system with a finite number of cells. The established analytical framework is invaluable for designing a variety of acoustic topological insulators with minimal computational cost.

physics.app-ph↗

arXiv2020

A comprehensive study on the prediction reliability of graph neural networks for virtual screening

Prediction models based on deep neural networks are increasingly gaining attention for fast and accurate virtual screening systems. For decision makings in virtual screening, researchers find it useful to interpret an output of classification system as probability, since such interpretation allows them to filter out more desirable compounds. However, probabilistic interpretation cannot be correct for models that hold over-parameterization problems or inappropriate regularizations, leading to unreliable prediction and decision making. In this regard, we concern the reliability of neural prediction models on molecular properties, especially when models are trained with sparse data points and imbalanced distributions. This work aims to propose guidelines for training reliable models, we thus provide methodological details and ablation studies on the following train principles. We investigate the effects of model architectures, regularization methods, and loss functions on the prediction performance and reliability of classification results. Moreover, we evaluate prediction reliability of models on virtual screening scenario. Our result highlights that correct choice of regularization and inference methods is evidently important to achieve high success rate, especially in data imbalanced situation. All experiments were performed under a single unified model implementation to alleviate external randomness in model training and to enable precise comparison of results.

cs.LG↗