An accurate prediction is certainly significant in financial data analysis. Investors have used a series of econometric techniques on pricing, stock selection and risk management but few of them have found great success due to the fact that most of them only are purely based on a single scheme. Recent advances in deep learning methods have also demonstrated the outstanding performance in the fields of image recognition and sentiment analysis. In this paper, we originally propose a novel gold price forecast method based on the integration of Long Short-Term Memory Neural Networks (LSTM) and Convolutional Neural Networks (CNN) with Attention Mechanism (denoted to LSTM-Attention-CNN model). Particularly, the LSTM-Attention-CNN model consists of three components: the LSTM component, Attention Mechanism and the CNN component. The LSTM component enables to harness the sequential order of daily gold price. Meanwhile, the Attention Mechanism assigns different attention weights on the new encoding method from LSTM component to enhance the extraction of the temporal and spatial features. In addition, the CNN component enables to capture the local patterns and abstract the spatial features. Extensive experiments on real dataset collected from World Gold Council show that our proposed approach outperforms other conventional financial forecast methods.
This paper proposes a fast load transient control for a bidirectional dual-active-bridge (DAB) DC/DC converter. It is capable of maintaining voltage–time balance during a step load change process so that no overshoot current and DC offset current exist. The transient control has been applied for all possible transition cases and the calculation of intermediate switching angles referring to the fixed reference points is independent from the converter parameters and the instantaneous current. The results have been validated by extended experimental tests.
It is non-trivial to predict the prices of precious metals since a number of factors can affect the fluctuations of precious metal prices. Either parametric models or machine learning models cannot accurately forecast the precious metal prices. Though deep learning approaches show their strengths in extracting key features from complicated data, they have the limitations of learning localization and losing some temporal and spatial features. The recent advances in attention mechanisms bring the opportunities to overcome the limitation of deep learning models. In this paper, we originally propose a Regularization Self-Attention Regression Model for precious metal price prediction. In particular, the proposed RSAR model consists of convolutional neural network (CNN) component and Long Short-Term Memory Neural Networks (LSTM) component. Integrating with self-attention mechanism, this model can extract both spatial and temporal features from precious metal price data. Meanwhile, the proper configuration of regularization functions can also lead to the further performance improvement. Extensive experiments on realistic precious metal price dataset show that our proposed approach outperforms other conventional machine learning and deep learning methods. INDEX TERMS Long short-term memory, convolutional neural network, attention mechanism, financial data analysis, deep learning.
Lyrics recognition is an important task in music processing. Despite the great number of traditional algorithms such as the hybrid HMM-TDNN model achieving good performance, studies on applying end-to-end models and self-supervised learning (SSL) are limited. In this paper, we first establish an end-to-end baseline for lyrics recognition and then explore the performance of SSL models. We evaluate four upstream SSL models based on their training method (masked reconstruction, masked prediction, autoregressive reconstruction, contrastive model). After applying the SSL model, the best performance improved by 5.23% for the dev set and 2.4% for the test set compared with the previous state-of-art baseline system even without language model trained by large corpus. Moreover, we study the generalization ability of the SSL features considering that those models were not trained on music datasets.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.