Jiali Deng

A novel coronavirus disease (COVID-19) is a pandemic disease has caused 4 million deaths and more than 200 million infections worldwide (as of August 4, 2021). Rapid and accurate diagnosis of COVID-19 infection is critical to controlling the spread of the epidemic. In order to quickly and efficiently detect COVID-19 and reduce the threat of COVID-19 to human survival, we have firstly pro- posed a detection framework based on reinforcement learn- ing for COVID-19 diagnosis, which constructs a mixed loss function that can integrate the advantages of multiple loss functions. This paper uses the accuracy of the validation set as the reward value, and obtains the initial model for the next epoch by searching the model corresponding to the maximum reward value in each epoch. We also have pro- posed a prediction framework that integrates multiple de- tection frameworks using parameter sharing to predict the progression of patients' disease without additional training. This paper also constructed a higher-quality version of the CT image dataset containing 247 cases screened by profes- sional physicians, and obtained more excellent results on this dataset. Meanwhile, we used the other two COVID-19 datasets as external verifications, and still achieved a high accuracy rate without additional training. Finally, the exper- imental results show that our classification accuracy can reach 98.31%, and the precision, sensitivity, specificity, and AUC (Area Under Curve) are 98.82%, 97.99%, 98.67%, and0.989, respectively. The accuracy of external verification can reach 93.34% and 91.05%. What's more, the accuracy of our prediction framework is 91.54%. A large number of experiments demonstrate that our proposed method is effective and robust for COVID-19 detection and prediction.

In a convolutional neural network (CNN), dropout cannot work well because dropped information is not entirely obscured in convolutional layers where features are correlated spatially. Except for randomly discarding regions or channels, many approaches try to overcome this defect by dropping influential units. In this paper, we propose a non-random dropout method named FocusedDropout, aiming to make the network focus more on the target. In FocusedDropout, we use a simple but effective method to search for the target-related features, retain these features and discard others, which is contrary to the existing methods. We find that this novel method can improve network performance by making the network more target focused. Additionally, increasing the weight decay while using FocusedDropout can avoid overfitting and increase accuracy. Experimental results show that with a slight cost, 10% of batches employing FocusedDropout, can produce a nice performance boost over the baselines on multiple datasets of classification, including CIFAR10, CIFAR100 and Tiny ImageNet, and has a good versatility for different CNN models.

Image resolution that has close relations with accuracy and computational cost plays a pivotal role in network training. In this paper, we observe that the reduced image retains relatively complete shape semantics but loses extensive texture information. Inspired by the consistency of the shape semantics as well as the fragility of the texture information, we propose a novel training strategy named Temporally Resolution Decrement. Wherein, we randomly reduce the training images to a smaller resolution in the time domain. During the alternate training with the reduced images and the original images, the unstable texture information in the images results in a weaker correlation between the texture-related patterns and the correct label, naturally enforcing the model to rely more on shape properties that are robust and conform to the human decision rule. Surprisingly, our approach greatly improves both the training and inference efficiency of convolutional neural networks. On ImageNet classification, using only 33\% calculation quantity (randomly reducing the training image to 112$\times$112 within 90\% epochs) can still improve ResNet-50 from 76.32\% to 77.71\%. Superimposed with the strong training procedure of ResNet-50 on ImageNet, our method achieves 80.42\% top-1 accuracy with saving 37.5\% calculation overhead. To the best of our knowledge this is the highest ImageNet single-crop accuracy on ResNet-50 under 224$\times$224 without extra data or distillation.

This retrospective study aimed to develop and validate deep-learning-based models for grading clear cell renal cell carcinoma (ccRCC) patients. A cohort enrolling 706 patients (n = 706) with pathologically verified ccRCC was used in this study. A temporal split was applied to verify our models: the first 83.9% of the cases (years 2010–2017) for development and the last 16.1% (year 2018–2019) for validation (development cohort: n = 592; validation cohort: n = 114). Here, we demonstrated a deep learning(DL) framework initialized by a self-supervised pre-training method, developed with the addition of mixed loss strategy and sample reweighting to identify patients with high grade for ccRCC. Four types of DL networks were developed separately and further combined with different weights for better prediction. The single DL model achieved up to an area under curve (AUC) of 0.864 in the validation cohort, while the ensembled model yielded the best predictive performance with an AUC of 0.882. These findings confirms that our DL approach performs either favorably or comparably in terms of grade assessment of ccRCC with biopsies whilst enjoying the non-invasive and labor-saving property.

The promising performances of CNNs often overshadow the need to examine whether they are doing in the way we are actually interested. We show through experiments that even over-parameterized models would still solve a dataset by recklessly leveraging spurious correlations, or so-called 'shortcuts'. To combat with this unintended propensity, we borrow the idea of printer test page and propose a novel approach called White Paper Assistance. Our proposed method involves the white paper to detect the extent to which the model has preference for certain characterized patterns and alleviates it by forcing the model to make a random guess on the white paper. We show the consistent accuracy improvements that are manifest in various architectures, datasets and combinations with other techniques. Experiments have also demonstrated the versatility of our approach on fine-grained recognition, imbalanced classification and robustness to corruptions.

Generative Adversarial Networks (GANs) are powerful generative models for numerous tasks and datasets. However, most of the existing models suffer from mode collapse. The most recent research indicates that the reason for it is that the optimal transportation map from random noise to the data distribution is discontinuous, but deep neural networks (DNNs) can only approximate continuous ones. Instead, the latent representation is a better raw material used to construct a transportation map point to the data distribution than random noise. Because it is a low-dimensional mapping related to the data distribution, the construction procedure seems more like expansion rather than starting all over. Besides, we can also search for more transportation maps in this way with smoother transformation. Thus, we have proposed a new training methodology for GANs in this paper to search for more transportation maps and speed the training up, named Express Construction. The key idea is to train GANs with two independent phases for successively yielding latent representation and data distribution. To this end, an Auto-Encoder is trained to map the real data into the latent space, and two couples of generators and discriminators are used to produce them. To the best of our knowledge, we are the first to decompose the training procedure of GAN models into two more uncomplicated phases, thus tackling the mode collapse problem without much more computational cost. We also provide theoretical steps toward understanding the training dynamics of this procedure and prove assumptions. No extra hyper-parameters have been used in the proposed method, which indicates that Express Construction can be used to train any GAN models. Extensive experiments are conducted to verify the performance of realistic image generation and the resistance to mode collapse. The results show that the proposed method is lightweight, effective, and less prone to mode collapse.

In this paper, we propose a novel training strategy named Feature Mining for convolutional neural networks (CNNs) that aims to strengthen the network’s learning of the local features. Through experiments, we found that different parts of the feature contain different semantics, while the network will inevitably lose a large amount of local information during feedforward propagation. In order to enhance the learning of the local features, Feature Mining divides the complete feature into two complementary parts and reuses this divided feature to make the network capture different local information; we call the two steps Feature Segmentation and Feature Reusing. Feature Mining is a parameter-free method with a plug-and-play nature and can be applied to any CNN model. Extensive experiments demonstrated the wide applicability, versatility, and compatibility of our method.

It has been shown that the learning rate is one of the most critical hyper-parameters for the overall performance of deep neural networks. In this paper, we propose a new method for setting the global learning rate, named random amplify learning rates (RALR), to improve the performance of any optimizer in training deep neural networks. Instead of monotonically decreasing the learning rate, we expect to escape saddle points or local minima by amplifying the learning rate between reasonable boundary values based on a given probability. Training with RALR rather than conventionally decreasing the learning rate achieves further improvement on networks’ performance without extra consumption. Remarkably, the RALR is complementary with state-of-the-art data augmentation and regularization methods. Besides, we empirically study its performance on image classification tasks, fine-grained classification tasks, object detection tasks, and machine translation tasks. Experiments demonstrate that RALR can bring a notable improvement while preventing overfitting when training deep neural networks. For example, the classification accuracy of ResNet-110 trained on the CIFAR-100 dataset using RALR achieves a 1.34% gain compared with ResNet-110 trained traditionally.

In recent years, deep convolutional features have been deployed in discriminative correlation filters (DCF) to boost object tracking performance. However, features captured from pre-trained classification networks are usually trained for image classification tasks, not object tracking. In this paper, we find that different convolutional feature channels play different roles in tracking different targets. Some feature channels are favorable for tracking a given target and can be acquired based on this target, some are irrelevant to track this target, and some can be the primary cause of trackers' performance degradation when tracking this target. Thus, we perform feature selection before learning correlation filters for object tracking, and the feature selection module is realized by reinforcement learning. We penalize the features non-positive to obtain a DCF tracker based on positive convolutional feature channels. Compared with DCF based trackers without a feature selection technique, our scheme improves the robustness of target representation, lessens the dimension of activations, and achieves better tracking performance. Extensive experiments on the OTB dataset demonstrate our feature selection scheme is simple, robust, and effective for DCF based trackers.

Introduction: Close contacts have become a potential threat to the spread of coronavirus disease 2019 (COVID-19). The purpose of this study was to understand the epidemiological characteristics of close contacts of confirmed or suspected cases of COVID-19 in the surrounding cities of Chengdu, China, so as to provide a basis for the management strategy of close contacts.Methods: Close contacts were determined through epidemiological investigation of indicated cases, and relevant information was entered in the “Close Contact Information Management System.” Retrospective data of close contacts from January 22 to May 1, 2020 were collected and organized. Meanwhile, the contact mode, isolation mode, and medical outcome of close contacts were descriptively analyzed.Results: A total of 986 close contacts were effectively traced, with an average age of (36.69 ± 16.86) years old, who were mainly distributed in cities of eastern Chengdu. The frequency of contact was mainly occasional contact, 80.42% of them were relatives and public transportation personnel. Besides, the time of tracking close contacts and feedback was (10.64 ± 5.52) and (7.19 ± 6.11) days, respectively. A total of seven close contacts were converted to confirmed cases.Conclusions: Close contacts of COVID-19 have a risk of invisible infection. Early control of close contacts may be helpful to control the epidemic of COVID-19.