A backdoor sample attack is an attack that causes a deep neural network to misrecognize data that include a specific trigger because the model has been trained on malicious data that insert triggers into the deep neural network. The deep neural network correctly recognizes data without triggers, but incorrectly recognizes data with triggers. These backdoor attacks have mainly been studied in the image domain; however, defense research in the text domain is insufficient. In this study, we propose a method to defend against textual backdoor samples using a detection model. The proposed method detects a textual backdoor sample by comparing the resulting value of the target model with that of the model trained on the original training data. This method can defend against attacks without access to the entire training data. For the experimental setup, we used the TensorFlow library, and the MR and IMDB datasets were used as the experimental datasets. As a result of the experiment, when 1000 partial training datasets were used to train the detection model, the proposed method could classify the MR and IMDB datasets with detection rates of 79.6% and 83.2%, respectively.

With the rapid development of Internet technology, the type and quantity of network traffic data have increased accordingly, and network traffic classification has become an important research task. In previous research, there are methods based on traditional machine learning and deep learning; compared to machine learning, deep learning can obtain good results by converting network traffic into two-dimensional images and utilizing deep learning classification models. However, all of these methods have some limitations: the trained models cannot learn sustainably, and the generalization ability of the models is limited. In order to solve this problem, we propose a network traffic classification methods based on incremental learning and Mixup, which is based on generative adversarial networks. First, the network traffic is converted into a 2D image, the original database is linearly interpolated using Mixup to reduce the overfitting tendency of the model and improve the generalization ability, and the traffic is classified using the ability of deep learning on the image. Secondly, we improve the traditional incremental learning algorithm. To effectively address the imbalance between old and new categories in incremental learning. The experimental results show that the model performs well in classification experiments, reaching 92.26% and 93.86% accuracy on the ISCXVPN2016 and USTC datasets, respectively, and we can maintain a high accuracy rate with limited storage space in the process of increasing new categories.

We propose a method for adaptively selecting merge candidates for the geometric partitioning mode (GPM) in versatile video coding (VVC). The conventional GPM contributes to improved coding efficiency and subjective quality by partitioning the block into two nonrectangular partitions with motion vectors. The motion vector of the GPM is encoded as an index of the merge candidate list, but it does not consider that the GPM partitions are nonrectangular. In this paper, the distribution of merge candidates was evaluated for each GPM mode and partition, and a characteristic bias was revealed. To improve the coding efficiency of VVC, the proposed method allows GPM to select merge candidates that are specific to the partition. This method also introduces adaptive reference frame selection using template matching of adjacent samples. Following common test conditions in the Joint Video Experts Team (JVET), the experimental results showed an improvement in coding efficiency, with a bitrate savings of 0.16%, compared to the reference software for exploration experiments on enhanced compression beyond VVC capability in the JVET.

Rail surface anomaly detection, referring to the process of identifying and localizing abnormal patterns in rail surface images, faces the limitation of robustness because of the large diversity of scale, quantity, and morphology of surface anomalies. To address this challenge, we propose a multi-scale rail surface anomaly detection method (MRS-AD) based on a distribution model, which cooperates neighborhood information to precisely locate rail surface anomalies. Specifically, MRS-AD integrates multi-scale structures to enhance the perception of different scale information of anomalies. Furthermore, the neighborhood information is utilized to capture the correlations between adjacent regions, and thereby a weighted multivariate Gaussian distribution model is estimated to improve the recognition capability of anomalous morphologies. To validate the effectiveness of MRS-AD, we collected and built a Rail Surface Anomaly Detection dataset (RSAD), considering the scale and quantity of rail surface anomalies. Extensive experiments on RSAD, RSDD and NEU-RSDD-2 demonstrate the superiority of MRS-AD. The code and dataset are publicly available at https://github.com/lyy70/MRS-AD

Replication is commonly used in distributed key-value stores for high availability. Recent works show that centralized replication provides high throughput through low-overhead write coordination and consistency-aware read forwarding. Unfortunately, they rely on specialized hardware, which is deploy-challenging and poses various limitations. To this end, we present Dalio, a software-based centralized replication system that does not require extra hardware while supporting high throughput. Our key idea is to offload the replication function to per-shard load balancers with eBPF, an emerging kernel-native technique. By building a replication coordinator with eBPF, we can avoid burdensome kernel networking stack overhead. Our experimental results show that Dalio achieves throughput better than the vanilla Linux by up to 2.05× and is comparable to a hardware-based solution.

In our work, we propose a lightweight two-stage convolutional recurrent network (BP-CRN) for multichannel speech enhancement (mcse), which consists of beamforming and post-filtering. Drawing inspiration from traditional methods, we design two core modules for spatial filtering and post-filtering with compensation, named BM and PF, respectively. Both core modules employ a convolutional encoding-decoding structure and utilize complex frequency-time long short-term memory (CFT-LSTM) blocks in the middle. Furthermore, the inter-module mask module is introduced to estimate and convey implicit spatial information and assist the post-filtering module in refining spatial filtering and suppressing residual noise. Experimental results demonstrate that, our proposed method contains only 1.27M parameters and outperforms three other mcse methods in terms of PESQ and STOI metrics.

Crowd counting is a crucial task in computer vision, which poses a significant challenge yet holds vast potential for practical applications in public safety and transportation. Traditional crowd counting approaches typically rely on a single framework to predict density maps or head point distributions. However, the straightforward architectures often fall short in cases of over-counting or omission, particularly in diverse crowded scenes. To address these limitations, we introduce the Density to Point Transformer (D2PT), an innovative approach for effective crowd counting and localization. Specifically, D2PT employs a Transformer-based teacher-student framework that integrates the insights of density-based and head-point-based methods. Furthermore, we introduce feature-aligned knowledge distillation, formulating a collaborative training approach that enhances the performance of both density estimation and point map prediction. Optimized with multiple loss functions, D2PT achieves state-of-the-art performance across five crowd counting datasets, demonstrating its robustness and effectiveness for intricate crowd counting and localization challenges.