Li Wang

Yawei Li, Kai Zhang, Radu Timofte et al. · eth-zurich, tencent-ai

This paper reviews the NTIRE 2022 challenge on efficient single image super-resolution with focus on the proposed solutions and results. The task of the challenge was to super-resolve an input image with a magnification factor of $\times$4 based on pairs of low and corresponding high resolution images. The aim was to design a network for single image super-resolution that achieved improvement of efficiency measured according to several metrics including runtime, parameters, FLOPs, activations, and memory consumption while at least maintaining the PSNR of 29.00dB on DIV2K validation set. IMDN is set as the baseline for efficiency measurement. The challenge had 3 tracks including the main track (runtime), sub-track one (model complexity), and sub-track two (overall performance). In the main track, the practical runtime performance of the submissions was evaluated. The rank of the teams were determined directly by the absolute value of the average runtime on the validation set and test set. In sub-track one, the number of parameters and FLOPs were considered. And the individual rankings of the two metrics were summed up to determine a final ranking in this track. In sub-track two, all of the five metrics mentioned in the description of the challenge including runtime, parameter count, FLOPs, activations, and memory consumption were considered. Similar to sub-track one, the rankings of five metrics were summed up to determine a final ranking. The challenge had 303 registered participants, and 43 teams made valid submissions. They gauge the state-of-the-art in efficient single image super-resolution.

Lei Yang, Xinyu Zhang, Li Wang et al.

Monocular 3D object detection is an essential perception task for autonomous driving. However, the high reliance on large-scale labeled data make it costly and time-consuming during model optimization. To reduce such over-reliance on human annotations, we propose Mix-Teaching, an effective semi-supervised learning framework applicable to employ both labeled and unlabeled images in training stage. Mix-Teaching first generates pseudo-labels for unlabeled images by self-training. The student model is then trained on the mixed images possessing much more intensive and precise labeling by merging instance-level image patches into empty backgrounds or labeled images. This is the first to break the image-level limitation and put high-quality pseudo labels from multi frames into one image for semi-supervised training. Besides, as a result of the misalignment between confidence score and localization quality, it's hard to discriminate high-quality pseudo-labels from noisy predictions using only confidence-based criterion. To that end, we further introduce an uncertainty-based filter to help select reliable pseudo boxes for the above mixing operation. To the best of our knowledge, this is the first unified SSL framework for monocular 3D object detection. Mix-Teaching consistently improves MonoFlex and GUPNet by significant margins under various labeling ratios on KITTI dataset. For example, our method achieves around +6.34% AP@0.7 improvement against the GUPNet baseline on validation set when using only 10% labeled data. Besides, by leveraging full training set and the additional 48K raw images of KITTI, it can further improve the MonoFlex by +4.65% improvement on AP@0.7 for car detection, reaching 18.54% AP@0.7, which ranks the 1st place among all monocular based methods on KITTI test leaderboard. The code and pretrained models will be released at https://github.com/yanglei18/Mix-Teaching.

Xiangyu Qin, Zhiyu Wu, Jinshi Cui et al.

Previous works on emotion recognition in conversation (ERC) follow a two-step paradigm, which can be summarized as first producing context-independent features via fine-tuning pretrained language models (PLMs) and then analyzing contextual information and dialogue structure information among the extracted features. However, we discover that this paradigm has several limitations. Accordingly, we propose a novel paradigm, i.e., exploring contextual information and dialogue structure information in the fine-tuning step, and adapting the PLM to the ERC task in terms of input text, classification structure, and training strategy. Furthermore, we develop our model BERT-ERC according to the proposed paradigm, which improves ERC performance in three aspects, namely suggestive text, fine-grained classification module, and two-stage training. Compared to existing methods, BERT-ERC achieves substantial improvement on four datasets, indicating its effectiveness and generalization capability. Besides, we also set up the limited resources scenario and the online prediction scenario to approximate real-world scenarios. Extensive experiments demonstrate that the proposed paradigm significantly outperforms the previous one and can be adapted to various scenes.

Zhuowen Yin, Xinyao Ding, Xin Zhang et al.

Autism Spectrum Disorder (ASD) has been emerging as a growing public health threat. Early diagnosis of ASD is crucial for timely, effective intervention and treatment. However, conventional diagnosis methods based on communications and behavioral patterns are unreliable for children younger than 2 years of age. Given evidences of neurodevelopmental abnormalities in ASD infants, we resort to a novel deep learning-based method to extract key features from the inherently scarce, class-imbalanced, and heterogeneous structural MR images for early autism diagnosis. Specifically, we propose a Siamese verification framework to extend the scarce data, and an unsupervised compressor to alleviate data imbalance by extracting key features. We also proposed weight constraints to cope with sample heterogeneity by giving different samples different voting weights during validation, and we used Path Signature to unravel meaningful developmental features from the two-time point data longitudinally. We further extracted machine learning focused brain regions for autism diagnosis. Extensive experiments have shown that our method performed well under practical scenarios, transcending existing machine learning methods and providing anatomical insights for autism early diagnosis.

Lintao Zhang, Jinjian Wu, Lihong Wang et al.

Brain structural MRI has been widely used to assess the future progression of cognitive impairment (CI). Previous learning-based studies usually suffer from the issue of small-sized labeled training data, while there exist a huge amount of structural MRIs in large-scale public databases. Intuitively, brain anatomical structures derived from these public MRIs (even without task-specific label information) can be used to boost CI progression trajectory prediction. However, previous studies seldom take advantage of such brain anatomy prior. To this end, this paper proposes a brain anatomy prior modeling (BAPM) framework to forecast the clinical progression of cognitive impairment with small-sized target MRIs by exploring anatomical brain structures. Specifically, the BAPM consists of a pretext model and a downstream model, with a shared brain anatomy-guided encoder to model brain anatomy prior explicitly. Besides the encoder, the pretext model also contains two decoders for two auxiliary tasks (i.e., MRI reconstruction and brain tissue segmentation), while the downstream model relies on a predictor for classification. The brain anatomy-guided encoder is pre-trained with the pretext model on 9,344 auxiliary MRIs without diagnostic labels for anatomy prior modeling. With this encoder frozen, the downstream model is then fine-tuned on limited target MRIs for prediction. We validate the BAPM on two CI-related studies with T1-weighted MRIs from 448 subjects. Experimental results suggest the effectiveness of BAPM in (1) four CI progression prediction tasks, (2) MR image reconstruction, and (3) brain tissue segmentation, compared with several state-of-the-art methods.

Cong Wang, Hongmin Xu, Xiong Zhang et al.

Vision Transformers (ViTs) have recently dominated a range of computer vision tasks, yet it suffers from low training data efficiency and inferior local semantic representation capability without appropriate inductive bias. Convolutional neural networks (CNNs) inherently capture regional-aware semantics, inspiring researchers to introduce CNNs back into the architecture of the ViTs to provide desirable inductive bias for ViTs. However, is the locality achieved by the micro-level CNNs embedded in ViTs good enough? In this paper, we investigate the problem by profoundly exploring how the macro architecture of the hybrid CNNs/ViTs enhances the performances of hierarchical ViTs. Particularly, we study the role of token embedding layers, alias convolutional embedding (CE), and systemically reveal how CE injects desirable inductive bias in ViTs. Besides, we apply the optimal CE configuration to 4 recently released state-of-the-art ViTs, effectively boosting the corresponding performances. Finally, a family of efficient hybrid CNNs/ViTs, dubbed CETNets, are released, which may serve as generic vision backbones. Specifically, CETNets achieve 84.9% Top-1 accuracy on ImageNet-1K (training from scratch), 48.6% box mAP on the COCO benchmark, and 51.6% mIoU on the ADE20K, substantially improving the performances of the corresponding state-of-the-art baselines.

Yunzhi Huang, Sahar Ahmad, Luyi Han et al.

Missing scans are inevitable in longitudinal studies due to either subject dropouts or failed scans. In this paper, we propose a deep learning framework to predict missing scans from acquired scans, catering to longitudinal infant studies. Prediction of infant brain MRI is challenging owing to the rapid contrast and structural changes particularly during the first year of life. We introduce a trustworthy metamorphic generative adversarial network (MGAN) for translating infant brain MRI from one time-point to another. MGAN has three key features: (i) Image translation leveraging spatial and frequency information for detail-preserving mapping; (ii) Quality-guided learning strategy that focuses attention on challenging regions. (iii) Multi-scale hybrid loss function that improves translation of tissue contrast and structural details. Experimental results indicate that MGAN outperforms existing GANs by accurately predicting both contrast and anatomical details.

Li Wang, Ying Wah Teh, Mohammed Ali Al-Garadi

Extreme multi-label text classification utilizes the label hierarchy to partition extreme labels into multiple label groups, turning the task into simple multi-group multi-label classification tasks. Current research encodes labels as a vector with fixed length which needs establish multiple classifiers for different label groups. The problem is how to build only one classifier without sacrificing the label relationship in the hierarchy. This paper adopts the multi-answer questioning task for extreme multi-label classification. This paper also proposes an auxiliary classification evaluation metric. This study adopts the proposed method and the evaluation metric to the legal domain. The utilization of legal Berts and the study on task distribution are discussed. The experiment results show that the proposed hierarchy and multi-answer questioning task can do extreme multi-label classification for EURLEX dataset. And in minor/fine-tuning the multi-label classification task, the domain adapted BERT models could not show apparent advantages in this experiment. The method is also theoretically applicable to zero-shot learning.

Ziying Song, Guoxing Zhang, Lin Liu et al.

Multi-modal 3D object detectors are dedicated to exploring secure and reliable perception systems for autonomous driving (AD).Although achieving state-of-the-art (SOTA) performance on clean benchmark datasets, they tend to overlook the complexity and harsh conditions of real-world environments. With the emergence of visual foundation models (VFMs), opportunities and challenges are presented for improving the robustness and generalization of multi-modal 3D object detection in AD. Therefore, we propose RoboFusion, a robust framework that leverages VFMs like SAM to tackle out-of-distribution (OOD) noise scenarios. We first adapt the original SAM for AD scenarios named SAM-AD. To align SAM or SAM-AD with multi-modal methods, we then introduce AD-FPN for upsampling the image features extracted by SAM. We employ wavelet decomposition to denoise the depth-guided images for further noise reduction and weather interference. At last, we employ self-attention mechanisms to adaptively reweight the fused features, enhancing informative features while suppressing excess noise. In summary, RoboFusion significantly reduces noise by leveraging the generalization and robustness of VFMs, thereby enhancing the resilience of multi-modal 3D object detection. Consequently, RoboFusion achieves SOTA performance in noisy scenarios, as demonstrated by the KITTI-C and nuScenes-C benchmarks. Code is available at https://github.com/adept-thu/RoboFusion.

Xing Yang, Jian Zhang, Qijian Chen et al.

Automatic segmentation of breast tumors from the ultrasound images is essential for the subsequent clinical diagnosis and treatment plan. Although the existing deep learning-based methods have achieved significant progress in automatic segmentation of breast tumor, their performance on tumors with similar intensity to the normal tissues is still not pleasant, especially for the tumor boundaries. To address this issue, we propose a PBNet composed by a multilevel global perception module (MGPM) and a boundary guided module (BGM) to segment breast tumors from ultrasound images. Specifically, in MGPM, the long-range spatial dependence between the voxels in a single level feature maps are modeled, and then the multilevel semantic information is fused to promote the recognition ability of the model for non-enhanced tumors. In BGM, the tumor boundaries are extracted from the high-level semantic maps using the dilation and erosion effects of max pooling, such boundaries are then used to guide the fusion of low and high-level features. Moreover, to improve the segmentation performance for tumor boundaries, a multi-level boundary-enhanced segmentation (BS) loss is proposed. The extensive comparison experiments on both publicly available dataset and in-house dataset demonstrate that the proposed PBNet outperforms the state-of-the-art methods in terms of both qualitative visualization results and quantitative evaluation metrics, with the Dice score, Jaccard coefficient, Specificity and HD95 improved by 0.70%, 1.1%, 0.1% and 2.5% respectively. In addition, the ablation experiments validate that the proposed MGPM is indeed beneficial for distinguishing the non-enhanced tumors and the BGM as well as the BS loss are also helpful for refining the segmentation contours of the tumor.

Li Wang, Min Xu, Quangui Zhang et al.

Social recommendation systems face the problem of social influence bias, which can lead to an overemphasis on recommending items that friends have interacted with. Addressing this problem is crucial, and existing methods often rely on techniques such as weight adjustment or leveraging unbiased data to eliminate this bias. However, we argue that not all biases are detrimental, i.e., some items recommended by friends may align with the user's interests. Blindly eliminating such biases could undermine these positive effects, potentially diminishing recommendation accuracy. In this paper, we propose a Causal Disentanglement-based framework for Regulating Social influence Bias in social recommendation, named CDRSB, to improve recommendation performance. From the perspective of causal inference, we find that the user social network could be regarded as a confounder between the user and item embeddings (treatment) and ratings (outcome). Due to the presence of this social network confounder, two paths exist from user and item embeddings to ratings: a non-causal social influence path and a causal interest path. Building upon this insight, we propose a disentangled encoder that focuses on disentangling user and item embeddings into interest and social influence embeddings. Mutual information-based objectives are designed to enhance the distinctiveness of these disentangled embeddings, eliminating redundant information. Additionally, a regulatory decoder that employs a weight calculation module to dynamically learn the weights of social influence embeddings for effectively regulating social influence bias has been designed. Experimental results on four large-scale real-world datasets Ciao, Epinions, Dianping, and Douban book demonstrate the effectiveness of CDRSB compared to state-of-the-art baselines.

Yanwen Wang, Yiyu Zhuang, Jiawei Zhang et al.

In this paper, we rethink text-to-avatar generative models by proposing TeRA, a more efficient and effective framework than the previous SDS-based models and general large 3D generative models. Our approach employs a two-stage training strategy for learning a native 3D avatar generative model. Initially, we distill a decoder to derive a structured latent space from a large human reconstruction model. Subsequently, a text-controlled latent diffusion model is trained to generate photorealistic 3D human avatars within this latent space. TeRA enhances the model performance by eliminating slow iterative optimization and enables text-based partial customization through a structured 3D human representation. Experiments have proven our approach's superiority over previous text-to-avatar generative models in subjective and objective evaluation.

Ziying Song, Lin Liu, Feiyang Jia et al.

In the realm of modern autonomous driving, the perception system is indispensable for accurately assessing the state of the surrounding environment, thereby enabling informed prediction and planning. The key step to this system is related to 3D object detection that utilizes vehicle-mounted sensors such as LiDAR and cameras to identify the size, the category, and the location of nearby objects. Despite the surge in 3D object detection methods aimed at enhancing detection precision and efficiency, there is a gap in the literature that systematically examines their resilience against environmental variations, noise, and weather changes. This study emphasizes the importance of robustness, alongside accuracy and latency, in evaluating perception systems under practical scenarios. Our work presents an extensive survey of camera-only, LiDAR-only, and multi-modal 3D object detection algorithms, thoroughly evaluating their trade-off between accuracy, latency, and robustness, particularly on datasets like KITTI-C and nuScenes-C to ensure fair comparisons. Among these, multi-modal 3D detection approaches exhibit superior robustness, and a novel taxonomy is introduced to reorganize the literature for enhanced clarity. This survey aims to offer a more practical perspective on the current capabilities and the constraints of 3D object detection algorithms in real-world applications, thus steering future research towards robustness-centric advancements.

Xuelin Qian, Li Wang, Yi Zhu et al.

Conventional 3D object detection approaches concentrate on bounding boxes representation learning with several parameters, i.e., localization, dimension, and orientation. Despite its popularity and universality, such a straightforward paradigm is sensitive to slight numerical deviations, especially in localization. By exploiting the property that point clouds are naturally captured on the surface of objects along with accurate location and intensity information, we introduce a new perspective that views bounding box regression as an implicit function. This leads to our proposed framework, termed Implicit Detection or ImpDet, which leverages implicit field learning for 3D object detection. Our ImpDet assigns specific values to points in different local 3D spaces, thereby high-quality boundaries can be generated by classifying points inside or outside the boundary. To solve the problem of sparsity on the object surface, we further present a simple yet efficient virtual sampling strategy to not only fill the empty region, but also learn rich semantic features to help refine the boundaries. Extensive experimental results on KITTI and Waymo benchmarks demonstrate the effectiveness and robustness of unifying implicit fields into object detection.

Li Wang, Rongzhi Gu, Weiji Zhuang et al.

Keyword spotting (KWS) and speaker verification (SV) are two important tasks in speech applications. Research shows that the state-of-art KWS and SV models are trained independently using different datasets since they expect to learn distinctive acoustic features. However, humans can distinguish language content and the speaker identity simultaneously. Motivated by this, we believe it is important to explore a method that can effectively extract common features while decoupling task-specific features. Bearing this in mind, a two-branch deep network (KWS branch and SV branch) with the same network structure is developed and a novel decoupling feature learning method is proposed to push up the performance of KWS and SV simultaneously where speaker-invariant keyword representations and keyword-invariant speaker representations are expected respectively. Experiments are conducted on Google Speech Commands Dataset (GSCD). The results demonstrate that the orthogonality regularization helps the network to achieve SOTA EER of 1.31% and 1.87% on KWS and SV, respectively.

Jiawang Nie, Li Wang, Zequn Zheng

Multi-view learning is frequently used in data science. The pairwise correlation maximization is a classical approach for exploring the consensus of multiple views. Since the pairwise correlation is inherent for two views, the extensions to more views can be diversified and the intrinsic interconnections among views are generally lost. To address this issue, we propose to maximize higher order correlations. This can be formulated as a low rank approximation problem with the higher order correlation tensor of multi-view data. We use the generating polynomial method to solve the low rank approximation problem. Numerical results on real multi-view data demonstrate that this method consistently outperforms prior existing methods.

Li Wang, Yingcong Zhou, Zhiguo Fu

The study on the implicit regularization induced by gradient-based optimization is a longstanding pursuit. In the present paper, we characterize the implicit regularization of momentum gradient descent (MGD) with early stopping by comparing with the explicit $\ell_2$-regularization (ridge). In details, we study MGD in the continuous-time view, so-called momentum gradient flow (MGF), and show that its tendency is closer to ridge than the gradient descent (GD) [Ali et al., 2019] for least squares regression. Moreover, we prove that, under the calibration $t=\sqrt{2/λ}$, where $t$ is the time parameter in MGF and $λ$ is the tuning parameter in ridge regression, the risk of MGF is no more than 1.54 times that of ridge. In particular, the relative Bayes risk of MGF to ridge is between 1 and 1.035 under the optimal tuning. The numerical experiments support our theoretical results strongly.

Li Wang, Lilon Wangner

As data volumes continue to grow, searches in data are becoming increasingly time-consuming. Classical index structures for neighbor search are no longer sustainable due to the "curse of dimensionality". Instead, approximated index structures offer a good opportunity to significantly accelerate the neighbor search for clustering and outlier detection and to have the lowest possible error rate in the results of the algorithms. Local sensing hashes is one of those. We indicate directions to mathematically model the properties of it.

Yunxiang Li, Jingxiong Li, Ruilong Dan et al.

Accurate segmentation is a crucial step in medical image analysis and applying supervised machine learning to segment the organs or lesions has been substantiated effective. However, it is costly to perform data annotation that provides ground truth labels for training the supervised algorithms, and the high variance of data that comes from different domains tends to severely degrade system performance over cross-site or cross-modality datasets. To mitigate this problem, a novel unsupervised domain adaptation (UDA) method named dispensed Transformer network (DTNet) is introduced in this paper. Our novel DTNet contains three modules. First, a dispensed residual transformer block is designed, which realizes global attention by dispensed interleaving operation and deals with the excessive computational cost and GPU memory usage of the Transformer. Second, a multi-scale consistency regularization is proposed to alleviate the loss of details in the low-resolution output for better feature alignment. Finally, a feature ranking discriminator is introduced to automatically assign different weights to domain-gap features to lessen the feature distribution distance, reducing the performance shift of two domains. The proposed method is evaluated on large fluorescein angiography (FA) retinal nonperfusion (RNP) cross-site dataset with 676 images and a wide used cross-modality dataset from the MM-WHS challenge. Extensive results demonstrate that our proposed network achieves the best performance in comparison with several state-of-the-art techniques.

Han Zhu, Li Wang, Jindong Wang et al.

Self-supervised pre-training could effectively improve the performance of low-resource automatic speech recognition (ASR). However, existing self-supervised pre-training are task-agnostic, i.e., could be applied to various downstream tasks. Although it enlarges the scope of its application, the capacity of the pre-trained model is not fully utilized for the ASR task, and the learned representations may not be optimal for ASR. In this work, in order to build a better pre-trained model for low-resource ASR, we propose a pre-training approach called wav2vec-S, where we use task-specific semi-supervised pre-training to refine the self-supervised pre-trained model for the ASR task thus more effectively utilize the capacity of the pre-trained model to generate task-specific representations for ASR. Experiments show that compared to wav2vec 2.0, wav2vec-S only requires a marginal increment of pre-training time but could significantly improve ASR performance on in-domain, cross-domain and cross-lingual datasets. Average relative WER reductions are 24.5% and 6.6% for 1h and 10h fine-tuning, respectively. Furthermore, we show that semi-supervised pre-training could close the representation gap between the self-supervised pre-trained model and the corresponding fine-tuned model through canonical correlation analysis.

Tai-Hsien Wu, Chunfeng Lian, Sanghee Lee et al.

Accurately segmenting teeth and identifying the corresponding anatomical landmarks on dental mesh models are essential in computer-aided orthodontic treatment. Manually performing these two tasks is time-consuming, tedious, and, more importantly, highly dependent on orthodontists' experiences due to the abnormality and large-scale variance of patients' teeth. Some machine learning-based methods have been designed and applied in the orthodontic field to automatically segment dental meshes (e.g., intraoral scans). In contrast, the number of studies on tooth landmark localization is still limited. This paper proposes a two-stage framework based on mesh deep learning (called TS-MDL) for joint tooth labeling and landmark identification on raw intraoral scans. Our TS-MDL first adopts an end-to-end \emph{i}MeshSegNet method (i.e., a variant of the existing MeshSegNet with both improved accuracy and efficiency) to label each tooth on the downsampled scan. Guided by the segmentation outputs, our TS-MDL further selects each tooth's region of interest (ROI) on the original mesh to construct a light-weight variant of the pioneering PointNet (i.e., PointNet-Reg) for regressing the corresponding landmark heatmaps. Our TS-MDL was evaluated on a real-clinical dataset, showing promising segmentation and localization performance. Specifically, \emph{i}MeshSegNet in the first stage of TS-MDL reached an averaged Dice similarity coefficient (DSC) at \textcolor[rgb]{0,0,0}{$0.964\pm0.054$}, significantly outperforming the original MeshSegNet. In the second stage, PointNet-Reg achieved a mean absolute error (MAE) of $0.597\pm0.761 \, mm$ in distances between the prediction and ground truth for $66$ landmarks, which is superior compared with other networks for landmark detection. All these results suggest the potential usage of our TS-MDL in orthodontics.

Xueyao Zhang, Jinchao Zhang, Yao Qiu et al.

Pop music generation has always been an attractive topic for both musicians and scientists for a long time. However, automatically composing pop music with a satisfactory structure is still a challenging issue. In this paper, we propose to leverage harmony-aware learning for structure-enhanced pop music generation. On the one hand, one of the participants of harmony, chord, represents the harmonic set of multiple notes, which is integrated closely with the spatial structure of music, the texture. On the other hand, the other participant of harmony, chord progression, usually accompanies the development of the music, which promotes the temporal structure of music, the form. Moreover, when chords evolve into chord progression, the texture and form can be bridged by the harmony naturally, which contributes to the joint learning of the two structures. Furthermore, we propose the Harmony-Aware Hierarchical Music Transformer (HAT), which can exploit the structure adaptively from the music, and make the musical tokens interact hierarchically to enhance the structure in multi-level musical elements. Experimental results reveal that compared to the existing methods, HAT owns a much better understanding of the structure and it can also improve the quality of generated music, especially in the form and texture.

Li Wang, Rongzhi Gu, Nuo Chen et al.

Keyword Spotting (KWS) remains challenging to achieve the trade-off between small footprint and high accuracy. Recently proposed metric learning approaches improved the generalizability of models for the KWS task, and 1D-CNN based KWS models have achieved the state-of-the-arts (SOTA) in terms of model size. However, for metric learning, due to data limitations, the speech anchor is highly susceptible to the acoustic environment and speakers. Also, we note that the 1D-CNN models have limited capability to capture long-term temporal acoustic features. To address the above problems, we propose to utilize text anchors to improve the stability of anchors. Furthermore, a new type of model (LG-Net) is exquisitely designed to promote long-short term acoustic feature modeling based on 1D-CNN and self-attention. Experiments are conducted on Google Speech Commands Dataset version 1 (GSCDv1) and 2 (GSCDv2). The results demonstrate that the proposed text anchor based metric learning method shows consistent improvements over speech anchor on representative CNN-based models. Moreover, our LG-Net model achieves SOTA accuracy of 97.67% and 96.79% on two datasets, respectively. It is encouraged to see that our lighter LG-Net with only 74k parameters obtains 96.82% KWS accuracy on the GSCDv1 and 95.77% KWS accuracy on the GSCDv2.

Li Wang, Wei Xiang, Ruhui Xue et al.

Recently, the anchor-free object detection model has shown great potential for accuracy and speed to exceed anchor-based object detection. Therefore, two issues are mainly studied in this article: (1) How to let the backbone network in the anchor-free object detection model learn feature extraction? (2) How to make better use of the feature pyramid network? In order to solve the above problems, Experiments show that our model has a certain improvement in accuracy compared with the current popular detection models on the COCO dataset, the designed attention mechanism module can capture contextual information well, improve detection accuracy, and use sepc network to help balance abstract and detailed information, and reduce the problem of semantic gap in the feature pyramid network. Whether it is anchor-based network model YOLOv3, Faster RCNN, or anchor-free network model Foveabox, FSAF, FCOS. Our optimal model can get 39.5% COCO AP under the background of ResNet50.

Li Wang

As data volumes continue to grow, clustering and outlier detection algorithms are becoming increasingly time-consuming. Classical index structures for neighbor search are no longer sustainable due to the "curse of dimensionality". Instead, approximated index structures offer a good opportunity to significantly accelerate the neighbor search for clustering and outlier detection and to have the lowest possible error rate in the results of the algorithms. Locality-sensitive hashing is one of those. We indicate directions to model the properties of LSH.

Li Wang, Dong Li, Han Liu et al.

Semantic segmentation is an important task for scene understanding in self-driving cars and robotics, which aims to assign dense labels for all pixels in the image. Existing work typically improves semantic segmentation performance by exploring different network architectures on a target dataset. Little attention has been paid to build a unified system by simultaneously learning from multiple datasets due to the inherent distribution shift across different datasets. In this paper, we propose a simple, flexible, and general method for semantic segmentation, termed Cross-Dataset Collaborative Learning (CDCL). Our goal is to train a unified model for improving the performance in each dataset by leveraging information from all the datasets. Specifically, we first introduce a family of Dataset-Aware Blocks (DAB) as the fundamental computing units of the network, which help capture homogeneous convolutional representations and heterogeneous statistics across different datasets. Second, we present a Dataset Alternation Training (DAT) mechanism to facilitate the collaborative optimization procedure. We conduct extensive evaluations on diverse semantic segmentation datasets for autonomous driving. Experiments demonstrate that our method consistently achieves notable improvements over prior single-dataset and cross-dataset training methods without introducing extra FLOPs. Particularly, with the same architecture of PSPNet (ResNet-18), our method outperforms the single-dataset baseline by 5.65\%, 6.57\%, and 5.79\% mIoU on the validation sets of Cityscapes, BDD100K, CamVid, respectively. We also apply CDCL for point cloud 3D semantic segmentation and achieve improved performance, which further validates the superiority and generality of our method. Code and models will be released.

Li Wang, Chenfei Wang, Xinyu Zhang et al.

3D object detection plays a crucial role in environmental perception for autonomous vehicles, which is the prerequisite of decision and control. This paper analyses partition-based methods' inherent drawbacks. In the partition operation, a single instance such as a pedestrian is sliced into several pieces, which we call it the partition effect. We propose the Spatial-Attention Graph Convolution (S-AT GCN), forming the Feature Enhancement (FE) layers to overcome this drawback. The S-AT GCN utilizes the graph convolution and the spatial attention mechanism to extract local geometrical structure features. This allows the network to have more meaningful features for the foreground. Our experiments on the KITTI 3D object and bird's eye view detection show that S-AT Conv and FE layers are effective, especially for small objects. FE layers boost the pedestrian class performance by 3.62\% and cyclist class by 4.21\% 3D mAP. The time cost of these extra FE layers are limited. PointPillars with FE layers can achieve 48 PFS, satisfying the real-time requirement.

Longfei Zheng, Jun Zhou, Chaochao Chen et al.

Graph Neural Networks (GNNs) have achieved remarkable performance by taking advantage of graph data. The success of GNN models always depends on rich features and adjacent relationships. However, in practice, such data are usually isolated by different data owners (clients) and thus are likely to be Non-Independent and Identically Distributed (Non-IID). Meanwhile, considering the limited network status of data owners, hyper-parameters optimization for collaborative learning approaches is time-consuming in data isolation scenarios. To address these problems, we propose an Automated Separated-Federated Graph Neural Network (ASFGNN) learning paradigm. ASFGNN consists of two main components, i.e., the training of GNN and the tuning of hyper-parameters. Specifically, to solve the data Non-IID problem, we first propose a separated-federated GNN learning model, which decouples the training of GNN into two parts: the message passing part that is done by clients separately, and the loss computing part that is learnt by clients federally. To handle the time-consuming parameter tuning problem, we leverage Bayesian optimization technique to automatically tune the hyper-parameters of all the clients. We conduct experiments on benchmark datasets and the results demonstrate that ASFGNN significantly outperforms the naive federated GNN, in terms of both accuracy and parameter-tuning efficiency.

Han Zhu, Li Wang, Pengyuan Zhang et al.

When only a limited amount of accented speech data is available, to promote multi-accent speech recognition performance, the conventional approach is accent-specific adaptation, which adapts the baseline model to multiple target accents independently. To simplify the adaptation procedure, we explore adapting the baseline model to multiple target accents simultaneously with multi-accent mixed data. Thus, we propose using accent-specific top layer with gate mechanism (AST-G) to realize multi-accent adaptation. Compared with the baseline model and accent-specific adaptation, AST-G achieves 9.8% and 1.9% average relative WER reduction respectively. However, in real-world applications, we can't obtain the accent category label for inference in advance. Therefore, we apply using an accent classifier to predict the accent label. To jointly train the acoustic model and the accent classifier, we propose the multi-task learning with gate mechanism (MTL-G). As the accent label prediction could be inaccurate, it performs worse than the accent-specific adaptation. Yet, in comparison with the baseline model, MTL-G achieves 5.1% average relative WER reduction.

Yue Sun, Kun Gao, Zhengwang Wu et al.

To better understand early brain growth patterns in health and disorder, it is critical to accurately segment infant brain magnetic resonance (MR) images into white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF). Deep learning-based methods have achieved state-of-the-art performance; however, one of major limitations is that the learning-based methods may suffer from the multi-site issue, that is, the models trained on a dataset from one site may not be applicable to the datasets acquired from other sites with different imaging protocols/scanners. To promote methodological development in the community, iSeg-2019 challenge (http://iseg2019.web.unc.edu) provides a set of 6-month infant subjects from multiple sites with different protocols/scanners for the participating methods. Training/validation subjects are from UNC (MAP) and testing subjects are from UNC/UMN (BCP), Stanford University, and Emory University. By the time of writing, there are 30 automatic segmentation methods participating in iSeg-2019. We review the 8 top-ranked teams by detailing their pipelines/implementations, presenting experimental results and evaluating performance in terms of the whole brain, regions of interest, and gyral landmark curves. We also discuss their limitations and possible future directions for the multi-site issue. We hope that the multi-site dataset in iSeg-2019 and this review article will attract more researchers on the multi-site issue.

Wenjing Fang, Chaochao Chen, Bowen Song et al.

Secure online transaction is an essential task for e-commerce platforms. Alipay, one of the world's leading cashless payment platform, provides the payment service to both merchants and individual customers. The fraud detection models are built to protect the customers, but stronger demands are raised by the new scenes, which are lacking in training data and labels. The proposed model makes a difference by utilizing the data under similar old scenes and the data under a new scene is treated as the target domain to be promoted. Inspired by this real case in Alipay, we view the problem as a transfer learning problem and design a set of revise strategies to transfer the source domain models to the target domain under the framework of gradient boosting tree models. This work provides an option for the cold-starting and data-sharing problems.

Zitong Wang, Li Wang, Raymond Chan et al.

We focus on developing a novel scalable graph-based semi-supervised learning (SSL) method for a small number of labeled data and a large amount of unlabeled data. Due to the lack of labeled data and the availability of large-scale unlabeled data, existing SSL methods usually encounter either suboptimal performance because of an improper graph or the high computational complexity of the large-scale optimization problem. In this paper, we propose to address both challenging problems by constructing a proper graph for graph-based SSL methods. Different from existing approaches, we simultaneously learn a small set of vertexes to characterize the high-dense regions of the input data and a graph to depict the relationships among these vertexes. A novel approach is then proposed to construct the graph of the input data from the learned graph of a small number of vertexes with some preferred properties. Without explicitly calculating the constructed graph of inputs, two transductive graph-based SSL approaches are presented with the computational complexity in linear with the number of input data. Extensive experiments on synthetic data and real datasets of varied sizes demonstrate that the proposed method is not only scalable for large-scale data, but also achieve good classification performance, especially for extremely small number of labels.

Qijian Chen, Lihui Wang, Li Wang et al.

Glioma grading before surgery is very critical for the prognosis prediction and treatment plan making. We present a novel wavelet scattering-based radiomic method to predict noninvasively and accurately the glioma grades. The method consists of wavelet scattering feature extraction, dimensionality reduction, and glioma grade prediction. The dimensionality reduction was achieved using partial least squares (PLS) regression and the glioma grade prediction using support vector machine (SVM), logistic regression (LR) and random forest (RF). The prediction obtained on multimodal magnetic resonance images of 285 patients with well-labeled intratumoral and peritumoral regions showed that the area under the receiver operating characteristic curve (AUC) of glioma grade prediction was increased up to 0.99 when considering both intratumoral and peritumoral features in multimodal images, which represents an increase of about 13% compared to traditional radiomics. In addition, the features extracted from peritumoral regions further increase the accuracy of glioma grading.

Li Wang, Lihui Wang, Qijian Chen et al.

We proposed a novel convolutional restricted Boltzmann machine CRBM-based radiomic method for predicting pathologic complete response (pCR) to neoadjuvant chemotherapy treatment (NACT) in breast cancer. The method consists of extracting semantic features from CRBM network, and pCR prediction. It was evaluated on the dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) data of 57 patients and using the area under the receiver operating characteristic curve (AUC). Traditional radiomics features and the semantic features learned from CRBM network were extracted from the images acquired before and after the administration of NACT. After the feature selection, the support vector machine (SVM), logistic regression (LR) and random forest (RF) were trained to predict the pCR status. Compared to traditional radiomic methods, the proposed CRBM-based radiomic method yielded an AUC of 0.92 for the prediction with the images acquired before and after NACT, and an AUC of 0.87 for the pretreatment prediction, which was increased by about 38%. The results showed that the CRBM-based radiomic method provided a potential means for accurately predicting the pCR to NACT in breast cancer before the treatment, which is very useful for making more appropriate and personalized treatment regimens.

Li Wang, Lei Zhu, En Yu et al.

Deep hashing has recently received attention in cross-modal retrieval for its impressive advantages. However, existing hashing methods for cross-modal retrieval cannot fully capture the heterogeneous multi-modal correlation and exploit the semantic information. In this paper, we propose a novel \emph{Fusion-supervised Deep Cross-modal Hashing} (FDCH) approach. Firstly, FDCH learns unified binary codes through a fusion hash network with paired samples as input, which effectively enhances the modeling of the correlation of heterogeneous multi-modal data. Then, these high-quality unified hash codes further supervise the training of the modality-specific hash networks for encoding out-of-sample queries. Meanwhile, both pair-wise similarity information and classification information are embedded in the hash networks under one stream framework, which simultaneously preserves cross-modal similarity and keeps semantic consistency. Experimental results on two benchmark datasets demonstrate the state-of-the-art performance of FDCH.

Xuetong Niu, Li Wang, Xulei Yang

Credit card has become popular mode of payment for both online and offline purchase, which leads to increasing daily fraud transactions. An Efficient fraud detection methodology is therefore essential to maintain the reliability of the payment system. In this study, we perform a comparison study of credit card fraud detection by using various supervised and unsupervised approaches. Specifically, 6 supervised classification models, i.e., Logistic Regression (LR), K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGB), as well as 4 unsupervised anomaly detection models, i.e., One-Class SVM (OCSVM), Auto-Encoder (AE), Restricted Boltzmann Machine (RBM), and Generative Adversarial Networks (GAN), are explored in this study. We train all these models on a public credit card transaction dataset from Kaggle website, which contains 492 frauds out of 284,807 transactions. The labels of the transactions are used for supervised learning models only. The performance of each model is evaluated through 5-fold cross validation in terms of Area Under the Receiver Operating Curves (AUROC). Within supervised approaches, XGB and RF obtain the best performance with AUROC = 0.989 and AUROC = 0.988, respectively. While for unsupervised approaches, RBM achieves the best performance with AUROC = 0.961, followed by GAN with AUROC = 0.954. The experimental results show that supervised models perform slightly better than unsupervised models in this study. Anyway, unsupervised approaches are still promising for credit card fraud transaction detection due to the insufficient annotation and the data imbalance issue in real-world applications.

Qian Zhang, Li Wang, Xiaopeng Zong et al.

Skull stripping for brain MR images is a basic segmentation task. Although many methods have been proposed, most of them focused mainly on the adult MR images. Skull stripping for infant MR images is more challenging due to the small size and dynamic intensity changes of brain tissues during the early ages. In this paper, we propose a novel CNN based framework to robustly extract brain region from infant MR image without any human assistance. Specifically, we propose a simplified but more robust flattened residual network architecture (FRnet). We also introduce a new boundary loss function to highlight ambiguous and low contrast regions between brain and non-brain regions. To make the whole framework more robust to MR images with different imaging quality, we further introduce an artifact simulator for data augmentation. We have trained and tested our proposed framework on a large dataset (N=343), covering newborns to 48-month-olds, and obtained performance better than the state-of-the-art methods in all age groups.

Fenqiang Zhao, Shunren Xia, Zhengwang Wu et al.

Convolutional Neural Networks (CNNs) have been providing the state-of-the-art performance for learning-related problems involving 2D/3D images in Euclidean space. However, unlike in the Euclidean space, the shapes of many structures in medical imaging have a spherical topology in a manifold space, e.g., brain cortical or subcortical surfaces represented by triangular meshes, with large inter-subject and intrasubject variations in vertex number and local connectivity. Hence, there is no consistent neighborhood definition and thus no straightforward convolution/transposed convolution operations for cortical/subcortical surface data. In this paper, by leveraging the regular and consistent geometric structure of the resampled cortical surface mapped onto the spherical space, we propose a novel convolution filter analogous to the standard convolution on the image grid. Accordingly, we develop corresponding operations for convolution, pooling, and transposed convolution for spherical surface data and thus construct spherical CNNs. Specifically, we propose the Spherical U-Net architecture by replacing all operations in the standard U-Net with their spherical operation counterparts. We then apply the Spherical U-Net to two challenging and neuroscientifically important tasks in infant brains: cortical surface parcellation and cortical attribute map development prediction. Both applications demonstrate the competitive performance in the accuracy, computational efficiency, and effectiveness of our proposed Spherical U-Net, in comparison with the state-of-the-art methods.

Li Wang, Junlin Yao, Yunzhe Tao et al.

In this paper, we propose a deep learning approach to tackle the automatic summarization tasks by incorporating topic information into the convolutional sequence-to-sequence (ConvS2S) model and using self-critical sequence training (SCST) for optimization. Through jointly attending to topics and word-level alignment, our approach can improve coherence, diversity, and informativeness of generated summaries via a biased probability generation mechanism. On the other hand, reinforcement training, like SCST, directly optimizes the proposed model with respect to the non-differentiable metric ROUGE, which also avoids the exposure bias during inference. We carry out the experimental evaluation with state-of-the-art methods over the Gigaword, DUC-2004, and LCSTS datasets. The empirical results demonstrate the superiority of our proposed method in the abstractive summarization.

Xiaohuan Cao, Jianhua Yang, Li Wang et al.

Non-rigid inter-modality registration can facilitate accurate information fusion from different modalities, but it is challenging due to the very different image appearances across modalities. In this paper, we propose to train a non-rigid inter-modality image registration network, which can directly predict the transformation field from the input multimodal images, such as CT and MR images. In particular, the training of our inter-modality registration network is supervised by intra-modality similarity metric based on the available paired data, which is derived from a pre-aligned CT and MR dataset. Specifically, in the training stage, to register the input CT and MR images, their similarity is evaluated on the warped MR image and the MR image that is paired with the input CT. So that, the intra-modality similarity metric can be directly applied to measure whether the input CT and MR images are well registered. Moreover, we use the idea of dual-modality fashion, in which we measure the similarity on both CT modality and MR modality. In this way, the complementary anatomies in both modalities can be jointly considered to more accurately train the inter-modality registration network. In the testing stage, the trained inter-modality registration network can be directly applied to register the new multimodal images without any paired data. Experimental results have shown that, the proposed method can achieve promising accuracy and efficiency for the challenging non-rigid inter-modality registration task and also outperforms the state-of-the-art approaches.

Jose Dolz, Christian Desrosiers, Li Wang et al.

Precise 3D segmentation of infant brain tissues is an essential step towards comprehensive volumetric studies and quantitative analysis of early brain developement. However, computing such segmentations is very challenging, especially for 6-month infant brain, due to the poor image quality, among other difficulties inherent to infant brain MRI, e.g., the isointense contrast between white and gray matter and the severe partial volume effect due to small brain sizes. This study investigates the problem with an ensemble of semi-dense fully convolutional neural networks (CNNs), which employs T1-weighted and T2-weighted MR images as input. We demonstrate that the ensemble agreement is highly correlated with the segmentation errors. Therefore, our method provides measures that can guide local user corrections. To the best of our knowledge, this work is the first ensemble of 3D CNNs for suggesting annotations within images. Furthermore, inspired by the very recent success of dense networks, we propose a novel architecture, SemiDenseNet, which connects all convolutional layers directly to the end of the network. Our architecture allows the efficient propagation of gradients during training, while limiting the number of parameters, requiring one order of magnitude less parameters than popular medical image segmentation networks such as 3D U-Net. Another contribution of our work is the study of the impact that early or late fusions of multiple image modalities might have on the performances of deep architectures. We report evaluations of our method on the public data of the MICCAI iSEG-2017 Challenge on 6-month infant brain MRI segmentation, and show very competitive results among 21 teams, ranking first or second in most metrics.

Yanbin Wu, Li Wang, Fan Cui et al.

A novel data representation method of convolutional neural net- work (CNN) is proposed in this paper to represent data of different modalities. We learn a CNN model for the data of each modality to map the data of differ- ent modalities to a common space, and regularize the new representations in the common space by a cross-model relevance matrix. We further impose that the class label of data points can also be predicted from the CNN representa- tions in the common space. The learning problem is modeled as a minimiza- tion problem, which is solved by an augmented Lagrange method (ALM) with updating rules of Alternating direction method of multipliers (ADMM). The experiments over benchmark of sequence data of multiple modalities show its advantage.

Jiuxiang Gu, Zhenhua Wang, Jason Kuen et al.

In the last few years, deep learning has led to very good performance on a variety of problems, such as visual recognition, speech recognition and natural language processing. Among different types of deep neural networks, convolutional neural networks have been most extensively studied. Leveraging on the rapid growth in the amount of the annotated data and the great improvements in the strengths of graphics processor units, the research on convolutional neural networks has been emerged swiftly and achieved state-of-the-art results on various tasks. In this paper, we provide a broad survey of the recent advances in convolutional neural networks. We detailize the improvements of CNN on different aspects, including layer design, activation function, loss function, regularization, optimization and fast computation. Besides, we also introduce various applications of convolutional neural networks in computer vision, speech and natural language processing.

Mingkui Tan, Ivor W. Tsang, Li Wang

Matching Pursuit LASSIn Part I \cite{TanPMLPart1}, a Matching Pursuit LASSO ({MPL}) algorithm has been presented for solving large-scale sparse recovery (SR) problems. In this paper, we present a subspace search to further improve the performance of MPL, and then continue to address another major challenge of SR -- batch SR with many signals, a consideration which is absent from most of previous $\ell_1$-norm methods. As a result, a batch-mode {MPL} is developed to vastly speed up sparse recovery of many signals simultaneously. Comprehensive numerical experiments on compressive sensing and face recognition tasks demonstrate the superior performance of MPL and BMPL over other methods considered in this paper, in terms of sparse recovery ability and efficiency. In particular, BMPL is up to 400 times faster than existing $\ell_1$-norm methods considered to be state-of-the-art.O Part II: Applications and Sparse Recovery over Batch Signals