Time series forecasting has played the key role in different industrial, including finance, traffic, energy, and healthcare domains. While existing literatures have designed many sophisticated architectures based on RNNs, GNNs, or Transformers, another kind of approaches multi-layer perceptrons (MLPs) are proposed with simple structure, low complexity, {superior performance}. However, most MLP-based methods suffer from point-wise mappings information bottleneck, which largely hinders...

Multivariate time series (MTS) forecasting has shown great importance in numerous industries. Current state-of-the-art graph neural network (GNN)-based methods usually require both networks (e.g., GCN) and temporal LSTM) to capture inter-series (spatial) dynamics intra-series (temporal) dependencies, respectively. However, the uncertain compatibility of two puts an extra burden on handcrafted model designs. Moreover, separate spatial modeling naturally violates unified spatiotemporal...

Multimodal content, such as mixing text with images, presents significant challenges to rumor detection in social media. Existing multimodal has focused on tokens among spatial and sequential locations for unimodal representation or fusing clues of veracity across modalities. However, they suffer from less discriminative are vulnerable intricate location dependencies the time-consuming fusion tokens. This work makes first attempt at frequency domain, which efficiently transforms features...

We propose an energy amplification technique to address the issue that existing models easily overlook low-energy components in time series forecasting. This comprises block and restoration block. The enhances of improve model's learning efficiency for these components, while returns its original level. Moreover, considering energy-amplified data typically displays two distinct peaks frequency spectrum, we integrate with a seasonal-trend forecaster model temporal relationships independently,...

Medical time series has been playing a vital role in real-world healthcare systems as valuable information monitoring health conditions of patients. Accurate classification for medical series, e.g., Electrocardiography (ECG) signals, can help early detection and diagnosis. Traditional methods towards rely on handcrafted feature extraction statistical methods; with the recent advancement artificial intelligence, machine learning deep have become more popular. However, existing often fail to...

With the deepening of global agricultural trade and acceleration RMB internationalization, synergistic development cross-border fruit settlement has become a significant issue in international finance. This paper, from multi-agent collaborative perspective, systematically analyzes interaction mechanisms practical challenges trade. The research indicates that current is characterized by scale expansion product diversification. promotes facilitation reducing exchange rate risks simplifying...

Decoding visual information from human brain activity has seen remarkable advancements in recent research. However, the diversity cortical parcellation and fMRI patterns across individuals prompted development of deep learning models tailored to each subject. The personalization limits broader applicability decoding real-world scenarios. To address this issue, we introduce Wills Aligner, a novel approach designed achieve multi-subject collaborative decoding. Aligner begins by aligning data...

We propose an energy amplification technique to address the issue that existing models easily overlook low-energy components in time series forecasting. This comprises block and restoration block. The enhances of improve model's learning efficiency for these components, while returns its original level. Moreover, considering energy-amplified data typically displays two distinct peaks frequency spectrum, we integrate with a seasonal-trend forecaster model temporal relationships independently,...

There is a recognized need to improve the application of epidemiologic data in human health risk assessment especially for understanding and characterizing risks from environmental occupational exposures. Although there uncertainty associated with results most studies, techniques exist characterize that can be applied weight-of-evidence evaluations characterization efforts.

Modeling complex hierarchical and grouped feature interaction in the multivariate time series data is indispensable to comprehend dynamics predicting future condition. The implicit high-dimensional make forecasting very challenging. Many existing works did not put more emphasis on exploring explicit correlation among multiple data, complicated models are designed capture long- short-range pattern with aid of attention mechanism. In this work, we think that pre-defined graph or general...

Recently, frequency transformation (FT) has been increasingly incorporated into deep learning models to significantly enhance state-of-the-art accuracy and efficiency in time series analysis. The advantages of FT, such as high a global view, have rapidly explored exploited various tasks applications, demonstrating the promising potential FT new paradigm for Despite growing attention proliferation research this emerging field, there is currently lack systematic review in-depth analysis...

Due to the nonstationary nature, distribution of real-world multivariate time series (MTS) changes over time, which is known as drift. Most existing MTS forecasting models greatly suffer from drift and degrade performance time. Existing methods address via adapting latest arrived data or self-correcting per meta knowledge derived future data. Despite their great success in forecasting, these hardly capture intrinsic changes, especially a distributional perspective. Accordingly, we propose...

Both masked image modeling (MIM) and natural language supervision have facilitated the progress of transferable visual pre-training. In this work, we seek synergy between two paradigms study emerging properties when MIM meets supervision. To end, present a novel Reconstruction Language semantic Space (RILS) pre-training framework, in which sentence representations, encoded by text encoder, serve as prototypes to transform vision-only signals into patch-sentence probabilities semantically...

Crowdsourced live streaming (CLS), such as Twitch.tv and Inke.tv, has emerged an important multimedia application in recent years. Video delivery CLS service involves two steps: 1) video uploading-video (i.e., a channel) generated from broadcaster is uploaded to the server, which we call "first mile" network 2) distribution-the then delivered viewers channel. Today's services usually use conventional content solutions address distribution problem, while little attention been paid improve...

While most time series are non-stationary, it is inevitable for models to face the distribution shift issue in forecasting. Existing solutions manipulate statistical measures (usually mean and std.) adjust distribution. However, these operations can be theoretically seen as transformation towards zero frequency component of spectrum which cannot reveal full information would further lead utilization bottleneck normalization, thus hindering forecasting performance. To address this problem, we...

While numerous forecasters have been proposed using different network architectures, the Transformer-based models state-of-the-art performance in time series forecasting. However, based on Transformers are still suffering from vulnerability to high-frequency signals, efficiency computation, and bottleneck full-spectrum utilization, which essentially cornerstones for accurately predicting with thousands of points. In this paper, we explore a novel perspective enlightening signal processing...

Recent years have witnessed a new content delivery paradigm named crowdsourced CDN, in which devices deployed at edge network can prefetch contents and provide service. Crowdsourced CDN offers high-quality experience to end-users by reducing their access latency alleviates the load of backbone making use storage resources millions devices. In such paradigm, redirecting requests proper is critical for user experience. The uniqueness request redirection lies that: on one hand, bandwidth...

Since the development of self-supervised visual representation learning from contrastive to masked image modeling (MIM), there is no significant difference in essence, that is, how design proper pretext tasks for vision dictionary look-up. MIM recently dominates this line research with state-of-the-art performance on Transformers (ViTs), where core enhance patch-level context capturing network via denoising auto-encoding mechanism. Rather than tailoring tokenizers extra training stages as...

Multivariate time series (MTS) forecasting penetrates various aspects of our economy and society, whose roles become increasingly recognized. However, often MTS is unfair, not only degrading their practical benefits but even incurring potential risk. Unfair may be attributed to disparities relating advantaged disadvantaged variables, which has rarely been studied in the forecasting. In this work, we formulate fairness modeling problem as learning informative representations attending both...

A novel algorithm is developed to estimate the shadowing ratio for significant wave height (SWH) inversion of ocean fields imaged by horizontal polarized X-band nautical radar intelligently and conveniently. To solve problem that accuracy calculated in local image areas not ideal, high resolution images will lead time-consuming estimation root mean square slope angle-blurred sea surface edge detection, a shadow model from marine based on Convolutional Neural Network (CNN) established. The...