Binary neural networks have become a promising research topic due to their advantages of fast inference speed and low energy consumption. However, most existing studies focus on binary convolutional neural networks, w...
详细信息
Binary neural networks have become a promising research topic due to their advantages of fast inference speed and low energy consumption. However, most existing studies focus on binary convolutional neural networks, while less attention has been paid to binary graph neural networks. A common drawback of existing studies on binary graph neural networks is that they still include lots of inefficient full-precision operations in multiplying three matrices and are therefore not efficient enough. In this paper, we propose a novel method, called re-quantization-based binary graph neural networks(RQBGN), for binarizing graph neural networks. Specifically, re-quantization, a necessary procedure contributing to the further reduction of superfluous inefficient full-precision operations, quantizes the results of multiplication between any two matrices during the process of multiplying three matrices. To address the challenges introduced by requantization, in RQBGN we first study the impact of different computation orders to find an effective one and then introduce a mixture of experts to increase the model capacity. Experiments on five benchmark datasets show that performing re-quantization in different computation orders significantly impacts the performance of binary graph neural network models, and RQBGN can outperform other baselines to achieve state-of-the-art performance.
With the continuous decrease in the critical dimensions of integrated circuits, mask optimization has becomethe main challenge in VLSI design. In recent years, thriving machine learning has been gradually introduced i...
详细信息
With the continuous decrease in the critical dimensions of integrated circuits, mask optimization has becomethe main challenge in VLSI design. In recent years, thriving machine learning has been gradually introduced in the field ofoptical proximity correction (OPC). Currently, advanced learning-based frameworks have been limited by low mask printability or large computational overhead. To address these limitations, this paper proposes a learning-based frameworknamed SegNet-OPC, which can generate optimized masks from the target layout at shorter training and turnaround timewith higher mask printability. The proposed framework consists of a backbone network and loss terms suitable for maskoptimization tasks, followed by a fine-tuning network. The framework yields remarkable improvements over conventionalmethods, delivering significantly faster turnaround time and superior mask printability and manufacturability. With just1.25 hours of training, the framework achieves comparable mask complexity while surpassing the state-of-the-art methods,achieving a minimum 3% enhancement in mask printability and an impressive 16.7% improvement in mask manufacturability.
With the widespread use of blockchain technology for smart contracts and decentralized applications on the Ethereum platform, the blockchain has become a cornerstone of trust in the modern financial system. However, i...
详细信息
With the widespread use of blockchain technology for smart contracts and decentralized applications on the Ethereum platform, the blockchain has become a cornerstone of trust in the modern financial system. However, its anonymity has provided new ways for Ponzi schemes to commit fraud, posing significant risks to investors. Current research still has some limitations, for example, Ponzi schemes are difficult to detect in the early stages of smart contract deployment, and data imbalance is not considered. In addition, there is room for improving the detection accuracy. To address the above issues, this paper proposes LT-SPSD (LSTM-Transformer smart Ponzi schemes detection), which is a Ponzi scheme detection method that combines Long Short-Term Memory (LSTM) and Transformer considering the time-series transaction information of smart contracts as well as the global information. Based on the verified smart contract addresses, account features, and code features are extracted to construct a feature dataset, and the SMOTE-Tomek algorithm is used to deal with the imbalanced data classification problem. By comparing our method with the other four typical detection methods in the experiment, the LT-SPSD method shows significant performance improvement in precision, recall, and F1-score. The results of the experiment confirm the efficacy of the model, which has some application value in Ethereum Ponzi scheme smart contract detection.
Organic electrode molecules hold significant potential as the next generation of cathode materials for Li-ion batteries. In this study, we have introduced a multi-objective active learning framework that leverages Bay...
详细信息
Organic electrode molecules hold significant potential as the next generation of cathode materials for Li-ion batteries. In this study, we have introduced a multi-objective active learning framework that leverages Bayesian optimization and non-dominated sorting genetic algorithms-Ⅱ. This framework enables the selection of organic molecules characterized by high theoretical energy density and low gap(LUMO-HOMO)(LUMO, lowest unoccupied molecular orbital;HOMO, highest occupied molecular orbital). Remarkably, after only two cycles of active learning, the determination of coefficient can reach 0.962 for theoretical energy density and 0.920 for the gap with a modest dataset of 300 molecules, showcasing superior predictive capabilities. The 2,3,5,6-tetrafluorocyclohexa-2,5-diene-1,4-dione, selected by non-dominated sorting genetic algorithms-Ⅱ, has been successfully applied to Li-ion batteries as cathode materials, demonstrating a high capacity of 288 m Ah g^(-1)and a long cycle life of 1,000 cycles. This outcome underscores the high reliability of our framework. Furthermore, we have also validated the universality and transferability of our framework by applying it to two additional databases, the QM9 and OMEAD. When the training dataset of the model includes at least 500 molecules, the determination of coefficient essentially reaches approximately0.900 for four targets: gap, reduction potential, LUMO, and HOMO. Therefore, the universal framework in our work provides innovative insights applicable to other domains to expedite the screening process for target materials.
Ce3+, Tb3+ and Eu3+, as luminescent centers capable of emitting optically trichromatic, can obtain multi-coloured light when combined in the proper ratio. However, the metal-to-metal charge transfer mechanism (MMCT) e...
详细信息
Stochastic gradient descent(SGD) and its variants have been the dominating optimization methods in machine learning. Compared with SGD with small-batch training, SGD with large-batch training can better utilize the co...
详细信息
Stochastic gradient descent(SGD) and its variants have been the dominating optimization methods in machine learning. Compared with SGD with small-batch training, SGD with large-batch training can better utilize the computational power of current multi-core systems such as graphics processing units(GPUs)and can reduce the number of communication rounds in distributed training settings. Thus, SGD with large-batch training has attracted considerable attention. However, existing empirical results showed that large-batch training typically leads to a drop in generalization accuracy. Hence, how to guarantee the generalization ability in large-batch training becomes a challenging task. In this paper, we propose a simple yet effective method, called stochastic normalized gradient descent with momentum(SNGM), for large-batch training. We prove that with the same number of gradient computations, SNGM can adopt a larger batch size than momentum SGD(MSGD), which is one of the most widely used variants of SGD, to converge to an?-stationary point. Empirical results on deep learning verify that when adopting the same large batch size,SNGM can achieve better test accuracy than MSGD and other state-of-the-art large-batch training methods.
Congestive heart failure (CHF) poses a major challenge to public health worldwide, marked by high readmission rates. As a result, early and accurate prediction and proactive intervention are crucial for effectively re...
详细信息
Embodied visual exploration is critical for building intelligent visual agents. This paper presents the neural exploration with feature-based visual odometry and tracking-failure-reduction policy(Ne OR), a framework f...
详细信息
Embodied visual exploration is critical for building intelligent visual agents. This paper presents the neural exploration with feature-based visual odometry and tracking-failure-reduction policy(Ne OR), a framework for embodied visual exploration that possesses the efficient exploration capabilities of deep reinforcement learning(DRL)-based exploration policies and leverages feature-based visual odometry(VO) for more accurate mapping and positioning results. An improved local policy is also proposed to reduce tracking failures of feature-based VO in weakly textured scenes through a refined multi-discrete action space, keyframe fusion, and an auxiliary task. The experimental results demonstrate that Ne OR has better mapping and positioning accuracy compared to other entirely learning-based exploration frameworks and improves the robustness of feature-based VO by significantly reducing tracking failures in weakly textured scenes.
Manipulating magnetic couplings in molecular magnets is of great importance in improving the magnetic properties of such *** has been proved that by adjusting the strength of magnetic couplings and the arrangement of ...
详细信息
Manipulating magnetic couplings in molecular magnets is of great importance in improving the magnetic properties of such *** has been proved that by adjusting the strength of magnetic couplings and the arrangement of the intermolecular magnetic dipoles,magnetic blocking can be significantly *** manipulating the intramolecular dipole interactions by ligand modification was attempted with the use of three closely related dinuclear Er（Ⅲ） complexes of a common chemical fo rmula of [（COTR）Er（μ-Cl）(THF)]2(COTRis mo nosubstituted cyclooctatetraenide dianions with R=diphenylmethylsilyl（Ph2MeS） for 1,triethylsilyl（TES） for 2,and triisopropylsilyl（TIPS） for 3).Each of these co mplexes features a centro symmetric dinuclear core unit with their component Er（Ⅲ） ions doubly bridged by two chloro ligands and further coordinated with a capping substituted COTRligand and a coordinated THF *** studies reveal that the complexes display similar ferromagnetic couplings with comparable single-molecule magnetic *** ferromagnetic couplings dominated by the intramolecular dipole interactions are found to be 0.7614,0.7380,and 0.5635 cm-1for 1,2,and 3,*** angles（θ） between the magnetic easy axes and the intramolecular Er-Er lines are24.88（2）°,25.23（1）°,and 31.85（5）°,leading to transversal dipole fields of 0.0114,0.0113,and 0.0125 T for 1,2,and 3,*** the different ligand substitution generates a sizable difference of about 7° in the θ angle,the resulting difference in the dipole interactions is not sufficiently strong to cause any significant differences in their magnetic *** change in the θ angles to the "side-by-side"（θ=90°） or "head-to-tail"（θ=0°） arrangement of the magnetic easy axes,achievable by rational molecular design,is expected to lead to molecular magnetic materials with much enhanced properties.
In Storm systems, an efficient resource placement strategy is key to ensure application performance. However, the Storm platform uses a polled resource placement strategy, which leads to large resource and communicati...
详细信息
暂无评论