PointWavelet: Learning in Spectral Domain for 3-D Point Cloud Analysis.

With recent success of deep learning in 2-D visual recognition, deep-learning-based 3-D point cloud analysis has received increasing attention from the community, especially due to the rapid development of autonomous driving technologies. However, most existing methods directly learn point features in the spatial domain, leaving the local structures in the spectral domain poorly investigated. In this article, we introduce a new method, PointWavelet, to explore local graphs in the spectral domain via a learnable graph wavelet transform. Specifically, we first introduce the graph wavelet transform to form multiscale spectral graph convolution to learn effective local structural representations. To avoid the time-consuming spectral decomposition, we then devise a learnable graph wavelet transform, which significantly accelerates the overall training process. Extensive experiments on four popular point cloud datasets, ModelNet40, ScanObjectNN, ShapeNet-Part, and S3DIS, demonstrate the effectiveness of the proposed method on point cloud classification and segmentation.

On exploring node-feature and graph-structure diversities for node drop graph pooling.

Graph Neural Networks (GNNs) have been successfully applied to graph-level tasks in various fields such as biology, social networks, computer vision, and natural language processing. For the graph-level representations learning of GNNs, graph pooling plays an essential role. Among many pooling techniques, node drop pooling has garnered significant attention and is considered as a leading approach. However, existing node drop pooling methods, which typically retain the top-k nodes based on their significance scores, often overlook the diversity inherent in node features and graph structures. This limitation leads to suboptimal graph-level representations. To overcome this, we introduce a groundbreaking plug-and-play score scheme, termed MID. MID comprises a Multidimensional score space and two key operations: flIpscore and Dropscore. The multidimensional score space depicts the significance of nodes by multiple criteria; the flipscore process promotes the preservation of distinct node features; the dropscore compels the model to take into account a range of graph structures rather than focusing on local structures. To evaluate the effectiveness of our proposed MID, we have conducted extensive experiments by integrating it with a broad range of recent node drop pooling methods, such as TopKPool, SAGPool, GSAPool, and ASAP. In particular, MID has proven to bring a significant average improvement of approximately 2.8% over the four aforementioned methods when tested on 17 real-world graph classification datasets. Code is available at https://github.com/whuchuang/mid.

Understanding How Pretraining Regularizes Deep Learning Algorithms.

Deep learning algorithms have led to a series of breakthroughs in computer vision, acoustical signal processing, and others. However, they have only been popularized recently due to the groundbreaking techniques developed for training deep architectures. Understanding the training techniques is important if we want to further improve them. Through extensive experimentation, Erhan et al. (2010) empirically illustrated that unsupervised pretraining has an effect of regularization for deep learning algorithms. However, theoretical justifications for the observation remain elusive. In this article, we provide theoretical supports by analyzing how unsupervised pretraining regularizes deep learning algorithms. Specifically, we interpret deep learning algorithms as the traditional Tikhonov-regularized batch learning algorithms that simultaneously learn predictors in the input feature spaces and the parameters of the neural networks to produce the Tikhonov matrices. We prove that unsupervised pretraining helps in learning meaningful Tikhonov matrices, which will make the deep learning algorithms uniformly stable and the learned predictor will generalize fast w.r.t. the sample size. Unsupervised pretraining, therefore, can be interpreted as to have the function of regularization.

Collaborative Knowledge Distillation via Multiknowledge Transfer.

Knowledge distillation (KD), as an efficient and effective model compression technique, has received considerable attention in deep learning. The key to its success is about transferring knowledge from a large teacher network to a small student network. However, most existing KD methods consider only one type of knowledge learned from either instance features or relations via a specific distillation strategy, failing to explore the idea of transferring different types of knowledge with different distillation strategies. Moreover, the widely used offline distillation also suffers from a limited learning capacity due to the fixed large-to-small teacher-student architecture. In this article, we devise a collaborative KD via multiknowledge transfer (CKD-MKT) that prompts both self-learning and collaborative learning in a unified framework. Specifically, CKD-MKT utilizes a multiple knowledge transfer framework that assembles self and online distillation strategies to effectively: 1) fuse different kinds of knowledge, which allows multiple students to learn knowledge from both individual instances and instance relations, and 2) guide each other by learning from themselves using collaborative and self-learning. Experiments and ablation studies on six image datasets demonstrate that the proposed CKD-MKT significantly outperforms recent state-of-the-art methods for KD.

Heatmap Regression via Randomized Rounding.

Heatmap regression has become the mainstream methodology for deep learning-based semantic landmark localization, including in facial landmark localization and human pose estimation. Though heatmap regression is robust to large variations in pose, illumination, and occlusion in unconstrained settings, it usually suffers from a sub-pixel localization problem. Specifically, considering that the activation point indices in heatmaps are always integers, quantization error thus appears when using heatmaps as the representation of numerical coordinates. Previous methods to overcome the sub-pixel localization problem usually rely on high-resolution heatmaps. As a result, there is always a trade-off between achieving localization accuracy and computational cost, where the computational complexity of heatmap regression depends on the heatmap resolution in a quadratic manner. In this paper, we formally analyze the quantization error of vanilla heatmap regression and propose a simple yet effective quantization system to address the sub-pixel localization problem. The proposed quantization system induced by the randomized rounding operation 1) encodes the fractional part of numerical coordinates into the ground truth heatmap using a probabilistic approach during training; and 2) decodes the predicted numerical coordinates from a set of activation points during testing. We prove that the proposed quantization system for heatmap regression is unbiased and lossless. Experimental results on popular facial landmark localization datasets (WFLW, 300W, COFW, and AFLW) and human pose estimation datasets (MPII and COCO) demonstrate the effectiveness of the proposed method for efficient and accurate semantic landmark localization. Code is available at http://github.com/baoshengyu/H3R.

A Multicenter Survey of Type I Diabetes Mellitus in Chinese Children.

Purpose: To investigate the features and treatment status of children with type 1 diabetes mellitus (T1DM) in China. Methods: We recruited patients <14 years of age with T1DM from 33 medical centers in 25 major cities of China between January 2012 and March 2015. All patients completed a questionnaire that was conducted by their pediatric endocrinologists at all centers. Results: A total of 1,603 children (755 males and 848 females) with T1DM participated in this survey. Of these, 834 (52.03%) of the patients exhibited diabetic ketoacidosis (DKA) at onset, while 769 patients (47.97%) did not exhibit DKA (non-DKA) at onset. There was a higher proportion of females (55.71%) in the cohort of patients exhibiting DKA at onset than in the non-DKA cohort (49.33%). The mean age of patients exhibiting DKA at presentation was 7.12 ± 0.14 years; this was significantly younger than that in non-DKA group (7.79 ± 0.15 years; P < 0.005). The frequency of DKA in 3 years old, 3-7 years old, and 7 years old or more was 77.21%, 26.17%, and 37.62%, respectively. Upon initial diagnosis, 29.4%, 15.2% and 11.8% of patients showed positivity for glutamic acid decarboxylase antibody (GADA), Insulin autoantibodies (IAA), or islet cell antibody (ICA), respectively. During six months follow-up, 244 patients (15.21%) reported receiving insulin pump therapy, and more than 60% of patients monitored their blood glucose levels less than 35 times per week. Although the majority of patients had no problems with obtaining insulin, 4.74% of the children surveyed were not able to receive insulin due to financial reasons, a shortage of insulin preparations, or the failure of the parents or guardians to acquire the appropriate medicine. Conclusion: DKA is more common in very young children. Treatment and follow-up of T1DM in China still face very serious challenges.

Anchor Cascade for Efficient Face Detection.

Face detection is essential to facial analysis tasks such as facial reenactment and face recognition. Both cascade face detectors and anchor-based face detectors have translated shining demos into practice and received intensive attention from the community. However, cascade face detectors often suffer from a low detection accuracy, while anchor-based face detectors rely heavily on very large neural networks pre-trained on large scale image classification datasets such as ImageNet [1], which is not computationally efficient for both training and deployment. In this paper, we devise an efficient anchor-based cascade framework called anchor cascade. To improve the detection accuracy by exploring contextual information, we further propose a context pyramid maxout mechanism for anchor cascade. As a result, anchor cascade can train very efficient face detection models with a high detection accuracy. Specifically, comparing with a popular CNN-based cascade face detector MTCNN [2], our anchor cascade face detector greatly improves the detection accuracy, e.g., from 0.9435 to 0.9704 at 1k false positives on FDDB, while it still runs in comparable speed. Experimental results on two widely used face detection benchmarks, FDDB and WIDER FACE, demonstrate the effectiveness of the proposed framework.

Sini Decoction Intervention on Atherosclerosis via PPARγ-LXRα-ABCA1 Pathway in Rabbits.

AIM: Sini decoction (SND) is a commonly used herbal formula showing lipid-lowering effects and is applied in traditional Chinese medicine (TCM) for the treatment of cardiovascular disease (CVD) and atherosclerosis (AS). However, the mechanisms behind its anti-atherosclerotic effects are still unknown, and will be investigated in this study. METHODS: AS was induced in rabbits by high fat diet (HFD) and treated with solvent (HFD group), atorvastatin (Ator group), or SND (SND group) for 12 weeks. Healthy rabbits (Chow group) were used as control. Serum and liver homogenates were collected, and lipid profiles as well as serum ApoA-I and ApoB were examined. Histopathological changes and lipid deposition in the proximal aorta and liver were detected by Oil red O staining. Western blot was used to detect the expression of ABCA1, ApoA-I, ApoB, PPARγ, and LXRα in liver, peritoneal macrophages, peripheral mononuclear cells (PMC), and adipose tissues. RESULTS: SND significantly attenuated the levels of total cholesterol (TC), triglyceride (TG), and low density lipoprotein cholesterol(LDL-C) in serum and liver. However, high density lipoprotein cholesterol (HDL-C) dramatically increased. SND treatment also decreased lipid deposition and improved the structure of the liver and aorta. Furthermore, SND enhanced the expression levels of ABCA1, PPARγ, and LXRα in liver, adipose tissues, PMC, and peritoneal macrophages. It also upregulated hepatic and serum ApoA-I expression and serum ApoA-I/ApoB ratio. CONCLUSIONS:SND treatment relieved AS, improved lipid profiles, and increased serum HDL-C level. The potential mechanism behind this might be the improvement of reverse cholesterol transport (RCT) involved with enhanced expression of ABCA1, ApoA-I, PPARγ, and LXRα.

Per-Round Knapsack-Constrained Linear Submodular Bandits.

Linear submodular bandits has been proven to be effective in solving the diversification and feature-based exploration problem in information retrieval systems. Considering there is inevitably a budget constraint in many web-based applications, such as news article recommendations and online advertising, we study the problem of diversification under a budget constraint in a bandit setting. We first introduce a budget constraint to each exploration step of linear submodular bandits as a new problem, which we call per-round knapsack-constrained linear submodular bandits. We then define an [Formula: see text]-approximation unit-cost regret considering that the submodular function maximization is NP-hard. To solve this new problem, we propose two greedy algorithms based on a modified UCB rule. We prove these two algorithms with different regret bounds and computational complexities. Inspired by the lazy evaluation process in submodular function maximization, we also prove that a modified lazy evaluation process can be used to accelerate our algorithms without losing their theoretical guarantee. We conduct a number of experiments, and the experimental results confirm our theoretical analyses.

[Hand and wrist bone maturation in children with central precocious puberty and idiopathic short stature].

OBJECTIVE: To investigate the maturation of individual bones on the hand and wrist in children with central precocious puberty (CPP) and idiopathic short stature (ISS). METHODS: Hand and wrist films of 25 children with CPP, 29 children with ISS and 21 normal controls were evaluated by conventional Greulich-Pyle (GP) atlas method and individual bone assessment method, in which all twenty bones of the hand and wrist were evaluated based on GP atlas, including 2 radius and ulna, 7 carpal bones, 11 metacarpal and phalangeal bones, the average bone age (BA) was calculated. The differences in groups were analyzed by independent samples t test. The differences between the two methods were analyzed by paired sample t test. The differences between BA and chronological age (CA) were analyzed by ROC with SPSS 17.0. RESULTS: Compared with the normal control group, the advance of BA in the CPP group was 0.70-2.26 y (1.48 ±0.78) by the GP atlas method, while that was 0.28-2.00 y(1.14 ±0.86) by the individual bone evaluation method. In all twenty bones, the advance of metacarpal and phalangeal BA was the greatest [0.34-2.06 y(1.2±0.86)]. In the ISS group,the delay of BA was 0.47-2.91 y(-1.69±1.22) by the GP atlas method, while that was 0.48-2.50 y (-1.49±1.01) by individual bone evaluation method.The delay of carpal BA was the greatest [0.59-2.73 y(-1.66±1.07)] in all twenty bones. In the ISS group and the normal control group, there were no statistic differences between the two methods. In the CPP group, statistic difference was found between two methods. There were no statistic differences for the areas under ROC curves between two methods. CONCLUSION: The advance of metacarpal and phalangeal BA is the greatest in CPP group and the delay of carpal BA is the greatest in ISS group.Both methods provide diagnostic information for bone age in CPP and ISS children.

The effect of herb-partition moxibustion on Toll-like receptor 4 in rabbit aorta during atherosclerosis.

OBJECTIVE: To explore the mechanism of Toll-like receptor (TLR4) inhibition in the delay of formation of atherosclerosis by herb-partition moxibustion. METHOD: Seventy-five rabbits were randomly assigned to one of five groups: blank, atherosclerosis (AS) model, direct moxibustion, herb-partition moxibustion, and drug treatment. With the exception of the blank group, all rabbits were given a high-fat diet in addition to immunologic injury to create the AS model. The experiments were carried out for 16 weeks, at which time the aorta was removed from each rabbit. Immunohistochemical methods were used to detect the gray level of the aortic TLR4 to observe the immunologic competence of its antigens. Fluorescent quantitative polymerase chain reaction was used to detect the expression of TLR4 messenger RNA (mRNA) in the aorta. RESULTS: The gray-scale value of TLR4 and the TLR4 mRNA expression significantly decreased (p<0.05) in the direct moxibustion, herb-partition moxibustion, and drug treatment groups. Furthermore, the effects of the herb-partition moxibustion and drug treatment were superior to those of the direct moxibustion. CONCLUSION: Herb-partition moxibustion inhibits aortic TLR4 activity and mRNA expression, showing that herb-partition moxibustion delays the formation of atherosclerosis through the inhibition of TLR4 expression.

Glucagon-like peptide 1 based therapy for type 2 diabetes.

BACKGROUND: Incidence of type 2 diabetes mellitus (T2DM) has increased in young people in recent years and new therapies are required for its effective treatment. Glucagon-like peptide 1 (GLP-1) is a potent blood glucose-lowering hormone produced in the L cells of the intestine. It may be potentially effective in the treatment of hyperglycemia in patients with T2DM. DATA SOURCES: PubMed database were searched with the terms "GLP-1", "incretins" and "diabetes". RESULTS: GLP-1 is a product of the glucagon gene, and its secretion is controlled by both neural and endocrine signals. GLP-1 lowers plasma glucose by stimulating insulin and suppressing secretion of glucagons, thus inhibiting gastric emptying and reducing appetite. GLP-1 exerts these actions by the engagement of structurally distinct G-protein-coupled receptors (GPCRs). In patients with T2DM, GLP-1 increases insulin secretion and normalizes both fasting and postprandial blood glucose when given as a continuous intravenous infusion. However, the native hormone is unsuitable as a drug because it is broken down rapidly by dipeptidyl peptidase IV (DPP-4) and cleared by the kidneys. Fortunately, many GLP-1 agonists or analogues and DPP-4 inhibitors have been found or developed, such as exendin-4, exenatide, liraglutide, CJC1131, vidaliptin and P32/98. Clinical trials have shown their therapeutic functions in T2DM with little adverse reaction. CONCLUSION: A GLP-1 based therapy will be safe and effective for the treatment of T2DM.

[Thyroid autoantibodies in children with Turner's syndrome].

OBJECTIVE: Turner's syndrome (TS) is characterized by the absence of an X chromosome or the presence of a structurally abnormal X chromosome in a phenotypic female. It was recently reported that autoimmune thyroiditis (AIT) was found in 38% of white patients with TS, and few studies in this aspect have been conducted in China. The purpose of this study was to determine the frequency of AIT among TS patients and risk factors for development of thyroid dysfunction in Chinese children with TS. METHODS: Serum antithyroglobulin antibody (TgAb), thyroperoxidase antibody (TPOAb) and thyroid function (T(3), T(4) and TSH) of 24 children with TS (mean age 12.9 +/- 2.4 years, range 4.8 - 16.8 years) were assessed. Their karyotype distribution was as follows: thirteen patients with 45, XO kayrotype, eight patients with structurally abnormal X chromosome, two with X mosaic kayrotype and one with 46, XX. Techniques including radioimmunoassy and elctro-chemiluminescence immunoassy were used in this study. All TS children were divided into two groups. Group one was thyroid autoantibodies (TAA)-positive group, the levels of TgAb and/or TPOAb in them were higher than the normal levels (TgAb < 30%, TPOAb < 20%), respectively, and the remaining patients were assigned into TAA-negative group. RESULTS: Seven of the 24 (29%) patients had higher levels of TgAb and TPOAb than the normal values (< 30% and < 20%). The level of serum TSH [6.1 (3.6-100.0) mU/L] in TAA-positive group was significantly higher than that [3.9 (1.7-7.9) mU/L] in TAA-negative group (P < 0.05). The frequency of hypothyroidism or subclinical hypothyroidism in TAA-positive group (5/7) was higher than that in TAA-negative group (3/17) (P < 0.05). CONCLUSION: The positive rate of serum TAA in children with TS was 29%. About 70% TS children with positive serum TAA developed hypothyroidism or subclinical hypothyroidism. The results have provided the basis for regular follow-up assessment of thyroid autoantibodies and thyroid function in children with TS, and these measures are of importance for timely diagnosis of thyroid dysfunction and application of appropriate treatment.