Search

Aspect-based sentiment analysis aims to recognize the sentiment polarities for opinion words with the aid of some machine learning or deep learning-based sentiment classification models. Dependency parsing has been considered as an efficient tool for identifying the opinion words in the sentiment text. However, many dependency-based methods might be susceptible to the dependency tree, which inevitably introduces noisy information due to that the rich relation information between words is neglected. In this paper, we propose a multi-feature fusion approach based on domain adaptive pretraining to reduce dependency-based noisy information for aspect-based sentiment classification (ASC). First, we utilize multi-task learning (MTL) for domain adaptive pretraining, which combines biaffine attention model (BAM) and mask language model (MLM) by jointly considering the structure, the relation semantics of edges, and the linguistic feature in the sentiment text. Second, to fully consider these different features affected with each other, a double graph fusion model is proposed, which takes as input the pretrained dependency graph into a message passing neural network (MPNN) initialized with the optimal parameters of the pretrained BAM for training. Lastly, extensive experiments were made against the state-of-the-art competitors over four benchmark datasets, and the results illustrate that our approach outperforms these competitors over most of the four datasets and achieves an accuracy of up to 92.69% and a macro-averaged F1 value of up to 85.79%. The MTL-based domain adaptive pretraining is efficient to achieve high-quality dependency parsing contributing to improving the performance of ASC, while maintaining lower computational cost. [ABSTRACT FROM AUTHOR]

Over the past decades, density functional theory (DFT) calculations have been utilized in various fields such as materials science and semiconductor devices. However, due to the inherent nature of DFT calculations, which rigorously consider interactions between atoms, they require significant computational cost. To address this, extensive research has recently focused on training neural networks to replace DFT calculations. However, previous methods for training neural networks necessitated an extensive number of DFT simulations to acquire the ground truth (Hamiltonians). Conversely, when dealing with a limited amount of training data, deep learning models often display increased errors in predicting Hamiltonians and band structures for testing data. This phenomenon poses the potential risk of generating inaccurate physical interpretations, including the emergence of unphysical branches within band structures. To tackle this challenge, we propose a novel deep learning-based method for calculating DFT Hamiltonians, specifically tailored to produce accurate results with limited training data. Our framework not only employs supervised learning with the calculated Hamiltonian but also generates pseudo Hamiltonians (targets for unlabeled data) and trains the neural networks on unlabeled data. Particularly, our approach, which leverages unlabeled data, is noteworthy as it marks the first attempt in the field of neural network Hamiltonians. Our framework showcases the superior performance of our framework compared to the state-of-the-art approach across various datasets, such as MoS _2 , Bi _2 Te _3 , HfO _2 , and InGaAs. Moreover, our framework demonstrates enhanced generalization performance by effectively utilizing unlabeled data, achieving noteworthy results when evaluated on data more complex than the training set, such as configurations with more atoms and temperature ranges outside the training data.

Optical transport networks (OTNs) are widely used in backbone- and metro-area transmission networks to increase network transmission capacity. In the OTN, it is particularly crucial to rationally allocate routes and maximize network capacities. By employing deep reinforcement learning (DRL)- and software-defined networking (SDN)-based solutions, the capacity of optical networks can be effectively increased. However, because most DRL-based routing optimization methods have low sample usage and difficulty in coping with sudden network connectivity changes, converging in software-defined OTN scenarios is challenging. Additionally, the generalization ability of these methods is weak. This paper proposes an ensembles- and message-passing neural-network-based Deep Q-Network (EMDQN) method for optical network routing optimization to address this problem. To effectively explore the environment and improve agent performance, the multiple EMDQN agents select actions based on the highest upper-confidence bounds. Furthermore, the EMDQN agent captures the network's spatial feature information using a message passing neural network (MPNN)-based DRL policy network, which enables the DRL agent to have generalization capability. The experimental results show that the EMDQN algorithm proposed in this paper performs better in terms of convergence. EMDQN effectively improves the throughput rate and link utilization of optical networks and has better generalization capabilities. [ABSTRACT FROM AUTHOR]

Lipophilicity, as quantified by the decimal logarithm of the octanol–water partition coefficient (log KOW), is an essential environmental property. Deep neural networks (DNNs) based quantitative structure–property relationship (QSPR) studies have received more and more attention because of their excellent performance for prediction. However, the black‐box nature of DNNs limits the application range where interpretability is essential. Hence, this study aims to develop an accurate and interpretable deep neural network (AI‐DNN) model for log KOW prediction. A hybrid method of molecular representation was employed to guarantee the accuracy of the proposed AI‐DNN model. The hybrid molecular representations are able to integrate the directed message passing neural networks (D‐MPNNs) learned molecular representations and the fixed molecule‐level features of CDK descriptors, and can capture both the local and the global features of overall molecule. The performance analysis shows that the proposed QSPR model exhibits promising predictive accuracy and discriminative power in the structural isomers and stereoisomers. Moreover, the Monte Carlo Tree Search (MCTS) approach was used to interpret the proposed AI‐DNN model by identifying the molecular substructures contributed to the lipophilicity. This interpretability can be applied to critical fields where there is a high demand for interpretable deep networks, such as green solvent design and drug discovery. [ABSTRACT FROM AUTHOR]

The Chemical Weapons Convention (CWC) is a science-based international treaty for the disarmament and non-proliferation of chemical weapons. However, reference Orgaization for prohibition of chemical weapons (OPCW) central analytical database (OCAD) of structures with mass spectra (MS) and retention indices (RI) includes only minor part of all possible chemical species defined in the Schedules of CWC. In this work we employed OCAD in silico augmentation based on chemoinformatics approach for chemical structures enumeration, MS data generation based on message passing neural network and based on P-alkyl molecular pairs RI prediction to support the verification activities as provided for in the CWC. Enumerated 879 noncyclic and 5270 monocyclic alcohols became the basis for generating hundreds of thousands molecules of Schedule 1 toxic chemicals like Sarin, Tabun, VX and Novichok. Trained on ordinary and neutral loss electron ionization mass spectrometry (EI-MS) a message-passing neural network (MPNN) outperformed other quantum chemistry and machine learning methods. Generated by this MPNN in silico EI-MS are very similar to the library's spectra and allowed to reach desired match factor above 800 within a scale of 0–1000. Statistical data for molecular pairs based on P-alkyl fragments was collected and used to predict RIs within desired 20 RI window for some toxic chemicals of Schedule 1.A.01, for which in the current OCAD version RIs are absent. [Display omitted] • Over 400 000 chemical structures covered by Chemical Weapon Convention (CWC) were enumerated. • Retention index for CWC compounds were calculated with molecular pairs. • Electron ionization mass spectra for CWC compounds can be predicted with artificial intelligence. • Proposed prediction method outperforms other approaches with mean match factor over 800. [ABSTRACT FROM AUTHOR]

Optical transport networks (OTNs) are widely used in backbone- and metro-area transmission networks to increase network transmission capacity. In the OTN, it is particularly crucial to rationally allocate routes and maximize network capacities. By employing deep reinforcement learning (DRL)- and software-defined networking (SDN)-based solutions, the capacity of optical networks can be effectively increased. However, because most DRL-based routing optimization methods have low sample usage and difficulty in coping with sudden network connectivity changes, converging in software-defined OTN scenarios is challenging. Additionally, the generalization ability of these methods is weak. This paper proposes an ensembles- and message-passing neural-network-based Deep Q-Network (EMDQN) method for optical network routing optimization to address this problem. To effectively explore the environment and improve agent performance, the multiple EMDQN agents select actions based on the highest upper-confidence bounds. Furthermore, the EMDQN agent captures the network’s spatial feature information using a message passing neural network (MPNN)-based DRL policy network, which enables the DRL agent to have generalization capability. The experimental results show that the EMDQN algorithm proposed in this paper performs better in terms of convergence. EMDQN effectively improves the throughput rate and link utilization of optical networks and has better generalization capabilities.

Various real-world scientific applications involve the mathematical modeling of complex uncertain systems with numerous unknown parameters. Accurate parameter estimation is often practically infeasible in such systems, as the available training data may be insufficient and the cost of acquiring additional data may be high. In such cases, based on a Bayesian paradigm, we can design robust operators retaining the best overall performance across all possible models and design optimal experiments that can effectively reduce uncertainty to enhance the performance of such operators maximally. While objective-based uncertainty quantification (objective-UQ) based on MOCU (mean objective cost of uncertainty) provides an effective means for quantifying uncertainty in complex systems, the high computational cost of estimating MOCU has been a challenge in applying it to real-world scientific/engineering problems. In this work, we propose a novel scheme to reduce the computational cost for objective-UQ via MOCU based on a data-driven approach. We adopt a neural message-passing model for surrogate modeling, incorporating a novel axiomatic constraint loss that penalizes an increase in the estimated system uncertainty. As an illustrative example, we consider the optimal experimental design (OED) problem for uncertain Kuramoto models, where the goal is to predict the experiments that can most effectively enhance robust synchronization performance through uncertainty reduction. We show that our proposed approach can accelerate MOCU-based OED by four to five orders of magnitude, without any visible performance loss compared to the state-of-the-art. The proposed approach applies to general OED tasks, beyond the Kuramoto model. [ABSTRACT FROM AUTHOR]

• A message-passing neural network based feedforward neural network (MPNN-FNN) is developed for σ-profile and V COSMO prediction. • Hybrid molecular representations are utilized to extract the molecular features. • The proposed model can well discriminate the cis/trans and structural isomers. • The IDAC calculated by predicted σ -profile and V COSMO is very close to the QM derived ones. The correct implementation of the quantum mechanics (QM) calculation for COSMO-SAC based surface charge density profiles (σ -profile) and cavity volumes (V COSMO) is tricky and time-consuming. As a shortcut, the GC approaches (GC-COSMO) were proposed to predict the required σ -profile and V COSMO for COSMO based methods. However, the GC-COSMO model cannot discriminate the isomers and cannot predict the complex molecules due to a lack of diverse functional groups. In this context, a message-passing neural network based feedforward neural network (MPNN-FNN) is proposed for rapid and accurate predicting COSMO-SAC based σ -profile and V COSMO. The hybrid molecular representation integrated the message-passing neural network (MPNN) learned representations and 200 additional molecule-level RDKit features, which can capture both local and global features of the overall molecule, is utilized to develop a better molecular feature extraction model. Based on the hybrid molecular representations, a feedforward neural network (FNN) model for the σ -profile and V COSMO prediction is trained and tested on the VT-2005 database. The developed MPNN-FNN model demonstrates significantly better performance than the GC-COSMO method and can well discriminate the cis/trans and structural isomers. In addition, the calculation speed based on the proposed model for COSMO-SAC based infinite dilution activity coefficient (IDAC) reaches about 100 compounds per second, which are the prerequisite to performing a high throughput green solvent screening. [ABSTRACT FROM AUTHOR]