An Evasion Attack against Stacked Capsule Autoencoder

Dai, Jiazhu; Xiong, Siwei

doi:10.48550/arxiv.2010.07230

Cited by 2 publications

(1 citation statement)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…41 We have addressed this challenge in our structure−property relationship model by means of a SMILES input affine transformation (AT). 42 The effectiveness of AT for input sequence scaling in ML has been demonstrated recently for multilayer perceptron (MLP), 43 convolutional neural networks (CNN), 44,45 and deep recurrent neural network architectures, such as long short-term memory (LSTM) 46 and fast−slow RNN (FS-RNN). 47 Our work identifies the optimal value of the scaling factor of AT, which significantly enhances the accuracy and efficiency of the parent BP algorithm (ATransformedBP).…”

Section: ■ Methodsmentioning

confidence: 99%

Dielectric Polymer Property Prediction Using Recurrent Neural Networks with Optimizations

Nazarova

Yang

Liu

et al. 2021

J. Chem. Inf. Model.

View full text Add to dashboard Cite

Despite the growing success of machine learning for predicting structure−property relationships in molecules and materials, such as predicting the dielectric properties of polymers, it is still in its infancy. We report on the effectiveness of solving structure−property relationships for a computer-generated database of dielectric polymers using recurrent neural network (RNN) models. The implementation of a series of optimization strategies was crucial to achieving high learning speeds and sufficient accuracy: (1) binary and nonbinary representations of SMILES (Simplified Molecular Input Line System) fingerprints and (2) backpropagation with affine transformation of the input sequence (ATransformedBP) and resilient backpropagation with initial weight update parameter optimizations (iRPROP − optimized). For the investigated database of polymers, the binary SMILES representation was found to be superior to the decimal representation with respect to the training and prediction performance. All developed and optimized Elman-type RNN algorithms outperformed nonoptimized RNN models in the efficient prediction of nonlinear structure−activity relationships. The average relative standard deviation (RSD) remained well below 5%, and the maximum RSD did not exceed 30%. Moreover, we provide a C++ codebase as a testbed for a new generation of open programming languages that target increasingly diverse computer architectures.

show abstract