Factor models, originating in finance for asset pricing, are fundamental tools in quantitative investment. Recently, there has been a trend towards adopting more flexible machine learning approaches instead of previou...
详细信息
ISBN:
(纸本)9798400704369
Factor models, originating in finance for asset pricing, are fundamental tools in quantitative investment. Recently, there has been a trend towards adopting more flexible machine learning approaches instead of previous linear models. However, traditional factor models and recent deep learning approaches either overlook the relationships among stocks or rely on static, predefined ones, which hampers their representational power and hinders their ability to dynamically adapt to market changes. To overcome this limitation, we introduce a novel dynamic factor model named GraphVAE. This model leverages temporal adaptive dynamic stock relationship graphs, facilitating improved information transfer among stocks within the dynamic probabilistic factor model. Experimental results on three real stock market datasets demonstrate that our method outperforms various state-of-the-art approaches.
System performance bottleneck location is the key to ensure the system configuration changes and high-quality stable operations. Currently, the process mainly relies on the experience of domain experts who conduct sim...
详细信息
ISBN:
(纸本)9798350359329;9798350359312
System performance bottleneck location is the key to ensure the system configuration changes and high-quality stable operations. Currently, the process mainly relies on the experience of domain experts who conduct simulations or analyze performance bottlenecks, which is quite challenging. Faced with the common problem that ground truth of bottlenecks is difficult to acquire, system metrics and calls are complex, and critical scenarios are often difficult to define, there is a certain gap for intelligent methods within this domain. Aimed at these issues, the paper proposes a bottleneck localization and explanation method based on PB-DVAE. In light of the characteristics of performance data, we design a two-stage model structure to contrast different environment stress test data and refine the calculation for the reconstruction probability to amplify the differences in reconstructions between different environment, subsequently ranking the reconstructed scores to represent bottleneck metrics. For the difficulty in annotating ground truth at time points, we design two-group bottleneck clustering to divide bottleneck time points and we propose a new metric dimension evaluation approach called HRN, each of which individually contributes to efficiently evaluating the model performance in time dimension and interpreting the effects in metric dimension. The experimental results show that our method can effectively locate the bottleneck metrics and help model evaluation analysis and verification at time point, which provides a new perspective for solving performance bottleneck localization problems by intelligent approaches.
This work proposes a variational autoencoder (VAE)-based rehabilitation framework that visualizes the vowel-mora for Japanese dysarthria. Traditionally, Speech-Language Pathology (SLP) has shown the guideline of rehab...
详细信息
ISBN:
(纸本)9798350378788;9798350378771
This work proposes a variational autoencoder (VAE)-based rehabilitation framework that visualizes the vowel-mora for Japanese dysarthria. Traditionally, Speech-Language Pathology (SLP) has shown the guideline of rehabilitation for dysarthria, but they should rely only on clinical experience and case-by-case adaption, highlighting the urgent necessity to push the boundary for showing a subjective guideline, which does not depend on the perspective of SLPs. The proposed framework takes advantage of two-dimensional latent representations of vowel-mora, which is assumed to be pre-processed by mel-spectrogram, via VAE. The experiments highlight the effectiveness of our proposed framework.
We introduce an anomaly detection method for multivariate time series data with the aim of identifying critical periods and features influencing extreme climate events like snowmelt in the Arctic. This method leverage...
详细信息
ISBN:
(纸本)9798350360332;9798350360325
We introduce an anomaly detection method for multivariate time series data with the aim of identifying critical periods and features influencing extreme climate events like snowmelt in the Arctic. This method leverages variational autoencoder (VAE) integrated with dynamic thresholding and correlationbased feature clustering. This framework enhances the VAE's ability to identify localized dependencies and learn the temporal relationships in climate data, thereby improving the detection of anomalies as demonstrated by its higher F1-score on benchmark datasets. The study's main contributions include the development of a robust anomaly detection method, improving feature representation within VAEs through clustering, and creating a dynamic threshold algorithm for localized anomaly detection. This method offers explainability of climate anomalies across different regions.
In many advertising areas, banners are often generated with different display sizes, so designers have to make huge efforts to retarget their designs to each size. Automating such retargeting process can greatly save ...
详细信息
In many advertising areas, banners are often generated with different display sizes, so designers have to make huge efforts to retarget their designs to each size. Automating such retargeting process can greatly save time for designers and let them put creativity on new ads. This paper proposes a hierarchical reinforcement learning-based (HRL-based) method and a variational autoencoder-based (VAE-based) method by treating the automated banner retargeting problem as a layout retargeting task. The HRL and VAE models are trained separately to learn the scaling and positioning policy of the design elements from an original (base) layout. Hence, the proposed method can generate appropriate layouts for different target banner sizes. Meanwhile, evaluation metrics are proposed to assess the quality of generated layouts and are also reward conditions during the training process. To evaluate performances of the two models, SOTA methods such as Non-linear Inverse Optimization (NIO), Triangle Interpolation (TI), and Layout GAN (LGAN) are implemented and compared. Experimental results show that both HRL- and VAE-based methods retarget design layouts effectively, and the VAE model achieves better performance than the HRL model.
The prevalence of offline password guessing attacks, also known as trawling, continues to challenge authentication systems. To quantify the threat posed by trawling, existing strategies leverage deep learning to model...
详细信息
ISBN:
(纸本)9798400716959
The prevalence of offline password guessing attacks, also known as trawling, continues to challenge authentication systems. To quantify the threat posed by trawling, existing strategies leverage deep learning to model password habits and predict likely user password choices. We propose PassRVAE, merging variational autoencoders (VAEs) and Gated Recurrent Unit (GRU) networks to augment the accuracy and efficiency of trawling attacks. We further break down the problem by composition policy to evaluate how models fare against specific types of passwords. We evaluate our solution against state-of-the-art models including PassGAN, VAEPass, and VAE-GPT2, on recent password datasets. PassRVAE demonstrates better overall performance as well as per password composition policy, achieving 21.32% higher accuracy with 109 guesses in the RockYou dataset, and 2.74%.27.46% higher accuracy with 108 guesses in six different policies of 4iQ.
We address speech enhancement based on variational autoencoders, which involves learning a speech prior distribution in the time-frequency (TF) domain. A zero-mean complex-valued Gaussian distribution is usually assum...
详细信息
ISBN:
(纸本)9798350344868;9798350344851
We address speech enhancement based on variational autoencoders, which involves learning a speech prior distribution in the time-frequency (TF) domain. A zero-mean complex-valued Gaussian distribution is usually assumed for the generative model, where the speech information is encoded in the variance as a function of a latent variable. In contrast to this commonly used approach, we propose a weighted variance generative model, where the contribution of each spectrogram time-frame in parameter learning is weighted. We impose a Gamma prior distribution on the weights, which would effectively lead to a Student's t-distribution instead of Gaussian for speech generative modeling. We develop efficient training and speech enhancement algorithms based on the proposed generative model. Our experimental results on spectrogram auto-encoding and speech enhancement demonstrate the effectiveness and robustness of the proposed approach compared to the standard unweighted variance model.
Evolutionary computation for addressing high-dimensional expensive problems (HEPs) characterized by both high-dimensional decision variables and resource-intensive evaluations is an important area. In this study, we i...
详细信息
ISBN:
(纸本)9789819771837;9789819771844
Evolutionary computation for addressing high-dimensional expensive problems (HEPs) characterized by both high-dimensional decision variables and resource-intensive evaluations is an important area. In this study, we introduce a novel approach, namely the Hierarchical Diffusion Teaching-learning-based Optimizer with variational autoencoder (HDTOV). Firstly, we employ a variational autoencoder to reduce problem dimensions and facilitate the learning of the optimization process. Secondly, we employ a hierarchical population reconstruction strategy to enhance population diversity. Lastly, to exploit the population more effectively, we implement a diffusion mechanism to prevent premature convergence. The proposed method is validated through experiments on a real-life optimization problem arising from the operation of mobile edge computing systems. The experimental results demonstrate the efficacy and efficiency of HDTOV in addressing HEPs by its outperforming the state of the art.
The hyperspectral pixel unmixing aims to find the underlying materials (endmembers) and their proportions (abundances) in pixels of a hyperspectral image. This work extends the Latent Dirichlet variational autoencoder...
详细信息
ISBN:
(纸本)9798350360332;9798350360325
The hyperspectral pixel unmixing aims to find the underlying materials (endmembers) and their proportions (abundances) in pixels of a hyperspectral image. This work extends the Latent Dirichlet variational autoencoder (LDVAE) pixel unmixing scheme by taking into account local spatial context while performing pixel unmixing. The proposed method uses an isotropic convolutional neural network with spatial attention to encode pixels as a dirichlet distribution over endmembers. We have evaluated our model on Samson, Hydice Urban, Cuprite, and OnTech-HSI-Syn-21 datasets. Our model also leverages the transfer learning paradigm for Cuprite Dataset, where we train the model on synthetic data and evaluate it on the real-world data. The results suggest that incorporating spatial context improves both endmember extraction and abundance estimation.
Precision medicine represents the direction of cancer treatment in the era of big data. Developing effective drug response prediction (DRP) models is crucial for achieving precision medicine. Although some DRP models ...
详细信息
ISBN:
(纸本)9798400718069
Precision medicine represents the direction of cancer treatment in the era of big data. Developing effective drug response prediction (DRP) models is crucial for achieving precision medicine. Although some DRP models have been proposed, their generalization performance on external datasets is generally mediocre due to insufficient extraction of drug features. To provide valuable references for clinical treatment, it is necessary for models to learn deeper feature representations from big data. Unlike traditional matrix factorization DRP methods, we propose a dual-variational autoencoder DRP model based on multi-omics. The dual-variational autoencoder obtains deeper latent feature vectors from big data for DRP. The proposed method is validated on GDSC and CCLE datasets. Compared to state-of-the-art prediction methods, the proposed method achieves the best performance in both regression and classification predictions.
暂无评论