Search

Recent status-of-the-art methods for multivariate time series forecasting can be categorized into graph-based approach and global-local approach. The former approach uses graphs to represent the dependencies among variables and apply graph neural networks to the forecasting problem. The latter approach decomposes the matrix of multivariate time series into global components and local components to capture the shared information across variables. However, both approaches cannot capture the propagation delay of the dependencies among individual variables of a multivariate time series, for example, the congestion at intersection A has delayed effects on the neighboring intersection B. In addition, graph-based forecasting methods cannot capture the shared global tendency across the variables of a multivariate time series; and global-local forecasting methods cannot reflect the nonlinear inter-dependencies among variables of a multivariate time series. In this paper, we propose to combine the advantages of both approaches by integrating A daptive G lobal- L ocal G raph Structure Learning with G ated R ecurrent U nits (AGLG-GRU). We learn a global graph to represent the shared information across variables. And we learn dynamic local graphs to capture the local randomness and nonlinear dependencies among variables. We apply diffusion convolution and graph convolution operations to global and dynamic local graphs to integrate the information of graphs and update gated recurrent unit for multivariate time series forecasting. The experimental results on seven representative real-world datasets demonstrate that our approach outperforms various existing methods. • Learn a global graph by sampling a PGM to capture the shared information and global dependencies among variables. • Learn dynamic local graphs to capture the local randomness and correlations between pairs of variables. • Devise a mechanism to integrate the information of global and dynamic local graphs by joint graph convolution operation. [ABSTRACT FROM AUTHOR]

Zero-shot learning (ZSL) methods mainly associate global or region features to semantic vectors within a single image, for transferring semantic knowledge from the seen classes to unseen ones. However, the interactive region learning among a group of images from different categories, which can enhance the discrimination of region features and thus lead to a desirable knowledge transfer between seen and unseen classes, is seldom considered. To remedy the above challenge, we propose a group-wise interactive region learning (GIRL) model to guarantee a comprehensive and explicit region interaction. Specifically, GIRL consists of an attentive region interaction (ARI) module and a holistic semantic embedding (HSE) module. ARI utilizes the semantic commonalities and differences of group regions to produce refined region features. HSE holistically maps these region features to the semantic space for a more stable semantic transfer. We also present a semantic consistency loss and a relation alignment loss that can distill the refined/original region features and introduce unseen class semantic vectors for training, respectively. Extensive experiments demonstrate the effectiveness of GIRL over other methods, achieving 68.9%, 42.9%, 75.5%, and 47.8% the Generalized ZSL (GZSL) H scores on CUB, SUN, AWA2, and APY. The code is publicly available at https://github.com/TingML/GIRL. • Group-wise interactive region learning (GIRL) model is proposed. • Attentive region interaction (ARI) refines the region features. • Holistic semantic embedding (HSE) maps region features to semantic space. • Show the effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]

The "shock and kill" strategy is being widely explored to purge Human Immunodeficiency Virus (HIV) latent reservoirs. Romidepsin, a kind of latency-reversing agents (LRAs), has been shown to induce HIV RNA transcription. However, several clinical trials testing this drug have resulted in limited effect in reducing the HIV latent reservoirs. To understand the mechanisms underlying such limited effect, we develop a multi-scale model that incorporates pharmacokinetics and considers the toxicity of romidepsin to T cells in this paper. By fitting the model to the viral load data from plasma of six patients received romidepsin, we find that the model with T cell toxicity of romidepsin can well explain the clinical data. The dynamics of latently infected cells during romidepsin administration are explored using the best-fit parameter values. The results show that latently infected cells decrease very slowly and remain very stable overall in four of the six participants under the assumption of T cell toxicity of romidepsin. This implies that the ineffectiveness of romidepsin on latent reservoirs can be explained by its toxicity to T cells. In the remaining two participants, however, latently infected cells are quite stable without T cell toxicity of LRAs. It is found that the estimated activation rate of latently infected cells by romidepsin and the estimated elimination rate of romidepsin on immune cells for these two patients are very different from those for the other four patients. Thus we speculate that the heterogeneous response to romidepsin across participants may also be a determining factor of the effectiveness of romidepsin. These results may have significant implications in the search for the control of HIV infection. • An HIV model is proposed to understand the cause of ineffectiveness of romidepsin in the "shock and kill" strategy. • The model has a good fit to patients' data when T cell toxicity of romidepsin is considered. • The toxicity of romidepsin to T cell may contribute to the ineffectiveness of "shock and kill" strategy concerning romidepsin. [ABSTRACT FROM AUTHOR]

Latently infected cells are considered as a major barrier to curing Human Immunodeficiency Virus (HIV) infection. Reactivation of latently infected cells followed by killing the actively infected cells may be a potential strategy ("shock and kill") to purge the latent reservoir. Based on vectored immunoprophylaxis (VIP) experiment that can elicit bNAbs, in this paper a mathematical model is formulated to explore the efficacy of "shock and kill" strategy with VIP. We derive the basic reproduction number R 0 of the model and show that R 0 completely determines the dynamics of the model: if R 0 < 1 , the disease-free equilibrium is globally asymptotically stable; if R 0 > 1 , the system is uniformly persistent. Numerical simulations suggest that the "shock and kill" strategy with VIP can effectively control HIV infection while this strategy cannot eradicate the reservoir without VIP although it can alleviate the HIV infection. To model the administration of drugs and vaccine more realistically, pharmacokinetics and pulse vaccination are incorporated into the model of ordinary differential equations. The resultants are described by impulsive differential equations. The thresholds are obtained for the frequency and strength of the vaccination to eliminate the viruses. Furthermore, the most appropriate times are numerically investigated for starting a short-term latency-reversing agents (LRAs) treatment relative to ART considering the toxicity of LRAs. The results show that LRAs treatment at the beginning of ART might be a better option. These results have important implications for the design of HIV cure-related clinical trials. • Two new HIV infection models are proposed to test the shock and kill strategy. • Threshold dynamics determined by the basic reproduction number are established. • Numerical simulations indicate that the shock and kill strategy with VIP can effectively control HIV infection. • Numerical simulation suggests the best time of a short-term latency-reversing agents treatment. [ABSTRACT FROM AUTHOR]