KEYWORDS: Data modeling, Performance modeling, Feature extraction, Convolution, Signal generators, Process modeling, Neurons, Mining, Data processing, Artificial intelligence
Financial time series prediction has always been a tricky problem due to the uncertainty in the market. It has attracted attention from industry to academia. In recent years, deep learning has shown excellent performance in many different fields. More and more researchers try to apply deep learning on financial markets. In this paper, the complete modeling process of price movement prediction is introduced. Based on high frequency data Limit Order Books, an improved deep learning model combining the local feature extraction ability of Convolutional Neural Network (CNN) with the sequential feature extraction ability of Long Short-Term Memory (LSTM) is proposed and evaluated on RB dominant contracts in the China futures market. Based on the experimental results, it is concluded that our model’s performance on prediction is better than that of single CNN and LSTM models. Through back testing, trading based on the predicted results of the proposed model can yield significantly more returns than other models.
Speaker embedding is a state-of-the-art front-end module, which is used to extract discriminative speaker features for speaker-related tasks. The Time Delay Neural Network (TDNN) has been a classical network architecture since it was first applied on speaker related tasks known as X-vector. In this paper, we propose new network structures based on current popular ECAPA-TDNN. We propose a dynamic kernel convolution module to extract features from short-term and long-term context adaptively, thus achieving multi-scale receptive fields. We also apply three enhanced attention modules instead of plain Squeeze-Excitation (SE) layer to realize more efficient information interaction between channels and spaces. The proposed architectures are superior to the most advanced network, with an optimal Equal Error Rate (EER) of 6.40% and a parameters reduction of 6.32%, they also achieve better performances when speaker utterances are shortened.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.