Showing papers by "Rong Zhang published in 2022"

Residual GCB-Net: Residual Graph Convolutional Broad Network on Emotion Recognition

Abstract: Electroencephalogram (EEG) data is commonly applied in the emotion recognition research area. It can accurately reflect the emotional changes of the human body by applying graphical-based algorithms or models. EEG signals are non-linear signals. Biological tissues’ adjustment and adaptive ability will inevitably affect electrophysiological signals, making EEG have the typical non-linear characteristics. Graph Convolutional Broad Network (GCB-net) extracted features from non-linear signals and abstract features via stacked CNN. It adopted the broad concept and enhanced the feature by the broad learning system (BLS), obtaining sound results. However, it performed poorly with increasing network depth, and the accuracy of some features decreased with BLS. This paper proposed a Residual Graph Convolutional Broad Network (Residual GCB-net), which promotes the performance on a deeper-layer network and extracts higher-level information. It substitutes the original convolutional layer with residual learning blocks, which solves the deep learning network degradation and extracts more features in deeper networks. In SEED data set, GCB-Res net could obtain the best accuracy (94.56%) on the all-frequency band of differential entropy (DE) and promote much on another feature. In Dreamer, it obtained the best accuracy (91.55%) on the dimension of Arousal. The result demonstrated the excellent classification performance of Residual GCB-net in EEG emotion recognition.

First genome-wide association study of esophageal atresia identifies three genetic risk loci at CTNNA3, FOXF1/FOXC2/FOXL1, and HNF1B

Abstract: Esophageal atresia with or without tracheoesophageal fistula (EA/TEF) is the most common congenital malformation of the upper digestive tract. This study represents the first genome-wide association study (GWAS) to identify risk loci for EA/TEF. We used a European case-control sample comprising 764 EA/TEF patients and 5,778 controls and observed genome-wide significant associations at three loci. On chromosome 10q21 within the gene CTNNA3 (p = 2.11 × 10-8; odds ratio [OR] = 3.94; 95% confidence interval [CI], 3.10-5.00), on chromosome 16q24 next to the FOX gene cluster (p = 2.25 × 10-10; OR = 1.47; 95% CI, 1.38-1.55) and on chromosome 17q12 next to the gene HNF1B (p = 3.35 × 10-16; OR = 1.75; 95% CI, 1.64-1.87). We next carried out an esophageal/tracheal transcriptome profiling in rat embryos at four selected embryonic time points. Based on these data and on already published data, the implicated genes at all three GWAS loci are promising candidates for EA/TEF development. We also analyzed the genetic EA/TEF architecture beyond the single marker level, which revealed an estimated single-nucleotide polymorphism (SNP)-based heritability of around 37% ± 14% standard deviation. In addition, we examined the polygenicity of EA/TEF and found that EA/TEF is less polygenic than other complex genetic diseases. In conclusion, the results of our study contribute to a better understanding on the underlying genetic architecture of ET/TEF with the identification of three risk loci and candidate genes.

A Novel Multi-Task Learning Method for Symbolic Music Emotion Recognition

Abstract: Symbolic Music Emotion Recognition(SMER) is to predict music emotion from symbolic data, such as MIDI and MusicXML. Previous work mainly focused on learning better representation via (mask) language model pre-training but ignored the intrinsic structure of the music, which is extremely important to the emotional expression of music. In this paper, we present a simple multi-task framework for SMER, which incorporates the emotion recognition task with other emotion-related auxiliary tasks derived from the intrinsic structure of the music. The results show that our multi-task framework can be adapted to different models. Moreover, the labels of auxiliary tasks are easy to be obtained, which means our multi-task methods do not require manually annotated labels other than emotion. Conducting on two publicly available datasets (EMOPIA and VGMIDI), the experiments show that our methods perform better in SMER task. Specifically, accuracy has been increased by 4.17 absolute point to 67.58 in EMOPIA dataset, and 1.97 absolute point to 55.85 in VGMIDI dataset. Ablation studies also show the effectiveness of multi-task methods designed in this paper.

PAU-Net: Privileged Action Unit Network for Facial Expression Recognition

Abstract: Facial expression recognition (FER) plays a vital role in affective cognition. However, there will be some limitations when facing the FER with single facial image data. Considering that extra data contains more information for molding, the facial action unit (AU) can be adopted as privileged information (PI) to assist the FER task. This paper integrates AU information into an end-to-end deep network to support FER training. The proposed privileged action unit network (PAU-Net) gives ways of integrating AU information from the input aspect (type I) and output aspect (type II). Type I of PAU-Net takes AUs as input to guide the facial image network learning, which provides the AU-based emotion recognition result for the image-based FER model. While, type II of PAU-Net utilizes AUs as the output label for shallow layers of the network, which helps the model learn AU-related features and further assists advanced facial expression feature learning in subsequent layers. Note that PI enhances the network during the training and will not occur during the testing. Therefore, the network can still perform robustly with original input data in practice. Experiments are based on the CK+, MMI, and Oulu-CASIA datasets. The experimental results demonstrate the effectiveness of the proposed PAU-Net in FER tasks.

A Survey on Siamese Network: Methodologies, Applications, and Opportunities

Abstract: Siamese network has obtained growing attention in real-life applications. In this survey, we present an comprehensive review on Siamese network from the aspects of methodologies, applications, and interesting topics for further exploration. We first introduce framework designs of Siamese network, followed by methodologies about learning with unlabeled data. Then, we review application scenarios in terms of classification and regression, together with relative methodologies. We also discuss the promising area of few-shot learning, followed by interesting topics about opportunities and challenges.

AIA-Net: Adaptive Interactive Attention Network for Text-Audio Emotion Recognition.

Abstract: Emotion recognition based on text-audio modalities is the core technology for transforming a graphical user interface into a voice user interface, and it plays a vital role in natural human-computer interaction systems. Currently, mainstream multimodal learning research has designed various fusion strategies to learn intermodality interactions but hardly considers that not all modalities play equal roles in emotion recognition. Therefore, the main challenge in multimodal emotion recognition is how to implement effective fusion algorithms based on the auxiliary structure. To address this problem, this article proposes an adaptive interactive attention network (AIA-Net). In AIA-Net, text is treated as a primary modality, and audio is an auxiliary modality. AIA-Net adapts to textual and acoustic features with different dimensions and learns their dynamic interactive relations in a more flexible way. The interactive relations are encoded as interactive attention weights to focus on the acoustic features that are effective for textual emotional representations. AIA-Net performs well in adaptively assisting the textual emotional representation with the acoustic emotional information. Moreover, multiple collaborative learning (co-learning) layers of AIA-Net achieve multiple multimodal interactions and the deep bottom-up evolution of emotional representations. Experimental results on three benchmark datasets demonstrate the great effectiveness of the proposed method over the state-of-the-art methods.

OneDConv: Generalized Convolution For Transform-Invariant Representation

Abstract: Convolutional Neural Networks (CNNs) have exhibited their great power in a variety of vision tasks. However, the lack of transform-invariant property limits their further applications in complicated real-world scenarios. In this work, we proposed a novel generalized one dimension convolutional operator (OneDConv), which dynamically transforms the convolution kernels based on the input features in a computationally and parametrically efficient manner. The proposed operator can extract the transform-invariant features naturally. It improves the robustness and generalization of convolution without sacrificing the performance on common images. The proposed OneDConv operator can substitute the vanilla convolution, thus it can be incorporated into current popular convolutional architectures and trained end-to-end readily. On several popular benchmarks, OneDConv outperforms the original convolution operation and other proposed models both in canonical and distorted images.

VSA: Adaptive Visual and Semantic Guided Attention on Few-Shot Learning

Abstract: AbstractTraining models with only a few samples often bring overfitting and generalization problems. Moreover, it has always been challenging to identify new classes based on small samples. However, studies have shown that humans can use prior knowledge such as vision and semantics to learn new categories from a small number of samples. We propose a bimodal attention mechanism (VSA) based on vision and semantics to better use this prior knowledge like humans. VSA can adaptively combine information from both visual and semantic modalities to guide visual feature extraction, that is, which features should be paid more attention to during feature extraction. Therefore, the new category is more discriminative even if only one sample exists. Meanwhile, our extensive experiments on miniImageNet, CIFAR-FS, and CUB demonstrate that our bimodal attention mechanism is effective and achieves state-of-the-art results on the CUB dataset.KeywordsAdaptiveAttentionFew-shot learning

Qi Sui Zhu Shui Plaster Inhibits AQP1 and MAPK Signaling Reduces Liver Damage Induced by Cirrhotic Ascites

Abstract: Objective At present, there is no special treatment for cirrhotic ascites in modern medicine. Qi Sui Zhu Shui plaster (QSZSP) has been used in ascites. The purpose of this study was to investigate the mechanism of action of QSZSP in the treatment of cirrhotic ascites and its relationship with aquaporin 1 (AQP1). Methods Twenty-four rats were divided into four groups, six rats in each group. Carbon tetrachloride-olive oil is injected into modeling. The control and model groups are treated with blank gel plaster (2 cm × 2 cm), QSZSP low-dose group is treated with Qi Sui Zhu Shui plaster (1 cm × 1 cm), and QSZSP high-dose group is treated with Qi Sui Zhu Shui plaster (2 cm × 2 cm). The changes in body weight and abdominal circumference were measured, the histopathological changes in liver, kidney, and peritoneum were observed in HE staining, the biochemical indexes related to liver function were detected, and the changes in AQP1 expression and the activation of MAPK pathway in the liver, kidney, and peritoneal tissues were evaluated in IHC staining and Western blot. Results After one week of injection of carbon tetrachloride-olive oil, the rats in the model group increased their body weight slowly, the abdominal circumference of the model rats continued to increase with time. After 16 weeks of construction of the cirrhotic ascites model, the liver, kidney, and peritoneum were significantly damaged, and the serum levels of TBiL, AST, ALT, Cr, BUN, K, Na, and Ca in the rats were significantly higher (P < 0.001) and ALB levels were significantly lower (P < 0.001) than those in the control group. After 4 weeks of treatment, the liver, kidney, and peritoneal injury were improved. TBiL, AST, ALT, Cr, BUN, K, Na, and Ca levels were significantly lower (P < 0.001) and ALB levels were significantly higher (P < 0.001) than those in the model group. The protein expression of AQP1, p-ERK, p-JNK, and p-p38 was found to be inhibited in the liver, kidney, and peritoneum. Conclusion QSZSP inhibits the protein expression of AQP1 and MAPK signaling pathway in the liver, peritoneum, and kidney to alleviate liver, kidney, and peritoneal injury caused by cirrhotic ascites, thus reducing the abnormal growth of abdominal circumference.