To address the issues of pseudo-label noise and overfitting in weakly supervised semantic segmentation (WSSS), this paper proposes a Dynamic Self-Distillation Former (DSDF-WSSS) to improve model stability and segmentation accuracy under a Transformer architecture. DSDF-WSSS adopts a progressive gradient-free teacher parameter evolution (PGTPE) mechanism and a dual-branch pixel-level distillation loss for robustness against noise and better boundary modeling. PGTPE builds a stable teacher model via progressive parameter fusion and delayed distillation to guide student learning. The pixel-level distillation loss offers fine-grained supervision on category distribution and mask structure, improving semantic consistency and structural alignment. On PASCAL VOC 2012 and MS COCO 2014 datasets, DSDF-WSSS achieves mIoU of 82.75% and 56.51%, achieving state-of-the-art performance under low-resource requirements.

In contrast to traditional semantic segmentation, Generalized Few-Shot Semantic Segmentation (GFSS) aligns more closely with real-world scenarios, as it enables the segmentation of previously unseen novel classes. However, existing GFSS methods often rely on complex architectures and overlook information in the frequency domain. FDCP-Match addresses GFSS by analyzing features from a frequency-domain perspective for the first time. It decomposes features into high- and low-frequency components to capture multi-scale information and introduces a lightweight class-prompt module for efficient knowledge injection into novel classes. Experiments on three benchmarks demonstrate its strong performance.

This paper proposes a Dynamic Sampling-based Non-local Correlation Matching model (DSNC-Match) for semi-supervised semantic segmentation. Traditional methods based on ResNet have focused on the extraction of local feature information, neglecting the effective mining of global information. Although Transformer models have garnered widespread attention for their superior ability to capture global information, they are associated with significant computational costs. In contrast, the DSNC-Match model proposed in this paper builds upon the strengths of ResNet in local feature extraction. Through innovative structural design, this new model not only enhances the capability to capture global information but also effectively mines the interrelated information between pixels and the latent information within historical models, all while maintaining a lower computational cost.

Compared with traditional pixel level supervised semantic segmentation, effectively training weakly supervised semantic segmentation (WSSS) models using image level labels is a challenging task. WSSS typically relies on Class Activation Maps (CAM) to generate pseudo-labels. However, CAMs in current methodologies often fail to fully capture comprehensive semantic information, leading to confirmation biases and lower quality pseudo-labels, thereby compromising the final segmentation accuracy. In response, this paper proposes a Dual teacher Dual student with Pixel Refinement (DtDsPR) model. Specifically, in the proposed model of this paper, each subnet-work processes the original input image differently, extracting semantic information from multiple perspectives to enhance the feature learning. Additionally, in order to further improve the quality of pseudo-labels and enhance segmentation accuracy, a Pixel-Variation Aware Refinement (PVAR) module is proposed, which refines pseudo-labels by calculating pixel-level changes and spatial relationships. Lastly, a self-reinforcement loss is introduced to further strengthen feature extraction. Extensive experiments conducted on the PASCAL VOC 2012 and MS COCO 2014 datasets demonstrate that the proposed single-stage WSSS method not only surpasses existing state-of-the-art benchmarks but also outperforms some multi-stage methods.

As an increasing number of microRNAs (miRNAs) have become biomarkers of various human diseases, prediction of the candidate disease-related miRNAs is helpful for facilitating the early diagnosis of diseases. Most of the recent prediction models concentrated on learning of the features from the heterogeneous graph composed of miRNAs and diseases. However, they failed to fully exploit the subgraph structures consisting of multiple miRNA and disease nodes, and they also did not completely integrate the context relationships among the pairwise features. We proposed a prediction model, SFPred, to integrate and encode the local topologies from neighborhood subgraphs, the dynamically evolved heterogeneous graph topology, and the context among pairwise features. First, the importance of an miRNA (disease) node to another node is formulated according to the subgraphs composed of their neighbors. Second, the features of each miRNA (disease) node continuously change when the graph encoding gradually deepens for the miRNA-disease heterogeneous network. A strategy based on multi-layer perceptron (MLP) is designed to estimate the edge weights according to the changed node features and form the dynamic graph topology. Third, considering the context relationships among the features of a pair of miRNA and disease nodes, a context relationship sensitive transformer is constructed to integrate these relationships. Finally, since the previous encoding layer of the transformer contains more detailed features of the pairwise, we present a multiperspective residual strategy to supplement the detailed features to the following encoding layer from the channel perspective and the feature one, respectively. The extensive experiments confirmed that SFPred outperforms eight state-of-the-art methods for the prediction of miRNA-disease associations, and the ablation experiments validate the effectiveness of the proposed innovations. The recall rates for the top-ranked candidate miRNAs related to the diseases and the case studies on three diseases indicate SFPred's ability in screening the reliable candidates for subsequent biological experiments.

This paper addresses the limitations of the Contrastive Language-Image Pre-training (CLIP) model’s image encoder and proposes a segmentation model WSSS-ECFE with enhanced CLIP feature extraction, aiming to improve the performance of the Weakly Supervised Semantic Segmentation (WSSS) task. WSSS-ECFE employs the Enhanced Bottleneck module proposed in this paper and adds dynamic residual connection to improve the model’s processing effect on complex scenes. In terms of implementation, the Enhanced Bottleneck module employs the Swish activation function and the Depthwise Separable Convolution to enhance the feature extraction and segmentation capability of the model, and uses multiple attention mechanisms to further optimize the feature representation and segmentation accuracy. The WSSS task on the public datasets PASCAL VOC 2012 and MS COCO 2014 achieves 82.6% and 56.3% mean intersection over union (mIoU), achieving state-of-the-art performance in models with low resource requirements.

Frequent itemset mining has always been one of the most classic tasks in data mining. It provides effective decision-making and judgment for many problems. A novel MPL (multi-partition list) structure is proposed in this paper combining bit combination and linear table structure. The MPL is composed of arrays where each unit stores a combination of items rather than a single item, which addresses the limitations of maintaining many pointers in the traditional tree structure. In addition, the MPL stores the least valid information required in the mining process. This paper further proposes a bit splicing frequent itemset mining algorithm based on dynamic grouping (BSFIM-DG) for the MPL. The algorithm dynamically calculates the number of grouping by using coverage according to the dataset’s characteristics. The candidate itemset is obtained by the bit-splicing method. The length of the MPL to be traversed is determined by the low-bit feature of the candidate itemset. The search space is reduced with the corresponding pruning strategy. Experiments on various open datasets demonstrate that the algorithm has excellent running speed, especially since the support is low. The proposed algorithm has a similar running speed to the BCLT-O and the FP-growth on some datasets. In terms of memory usage, the algorithm is better than the FP-growth and comparable to the BCLT-O, but there is still a particular gap with the Bit-combination algorithm. Nevertheless, as the pace of technology updates and iteration is getting faster and faster, it is very feasible to exchange space for speed.

This paper proposes a fault-tolerant frequent itemset mining algorithm (FT_HTlist) based on the linear table when the fault-tolerance is 1. The algorithm uses the method of concatenating 1 in the highest bit of the binary number of the known fault-tolerant frequent patterns to generate the candidate fault_tolerant patterns, called FT_Candidate. The algorithm is based on the data structure of the linear table for fault-tolerant frequent itemset mining. This method does not need recursion, so it reduces the consumption of mining space. At the same time, the paper proposed a deduplication algorithm to remove the support for repeat calculations. So the algorithm has a strong advantage in spatial performance. In addition, the algorithm only needs to mine two horizontal chains of the FT_Candidate, thus reducing the consumption of mining time. Finally, the paper shows the time performance and space performance of the proposed algorithm under sparse datasets and dense datasets. The results show that our algorithm has better mining time than other algorithms, and the horizontal chain reduces the memory occupation of the algorithm.

Traditional Convolutional Neural Networks (CNNs) can efficiently acquire local features, while ViT uses a Transformer structure that captures global contextual information. In this paper, a new high-performance lightweight target detector is designed by combining the respective advantages of CNN and ViT: ShuffleViTNet. In this paper, we improve the VIT Block in MobileVIT, firstly, we use channel-by-channel convolution to replace the normal convolution to reduce the memory access cost (MAC), number of parameters (Parmas) and floating-point operations (FLOPs) in the model inference. The 1x1 convolution is then used to fuse the local features learned by the convolution with the global features learned by the Transofmer, which further lightens the model and allows for easier modification of the number of channels and migration to other lightweight neural networks. Finally a lightweight network friendly to mobile devices is constructed using the improved VIT Block. On the ImageNet1k dataset, ShuffleViTNet achieves 75.2% Top-1 accuracy, the model's floating-point operation is 536.5 MFLOPs, which is 44.0% less compared to MobileViT-XS, and the cost of memory accesses for model inference is saved by about 25.6%, and the required memory is saved by about 66.37%. For the object detection task on the MS-COCO dataset, ShuffleViTNet obtains higher mAP accuracy. The experimental results show that ShuffleViTNet is a high-performance lightweight network and can be well applied to downstream tasks.

The proposed method in this paper proposes an end-to-end unsupervised semantic segmentation architecture DMSA based on four loss functions. The framework uses Atrous Spatial Pyramid Pooling (ASPP) module to enhance feature extraction. At the same time, a dynamic dilation strategy is designed to better capture multi-scale context information. Secondly, a Pixel-Adaptive Refinement (PAR) module is introduced, which can adaptively refine the initial pseudo labels after feature fusion to obtain high quality pseudo labels. Experiments show that the proposed DSMA framework is superior to the existing methods on the saliency dataset. On the COCO 80 dataset, the MIoU is improved by 2.0, and the accuracy is improved by 5.39. On the Pascal VOC 2012 Augmented dataset, the MIoU is improved by 4.9, and the accuracy is improved by 3.4. In addition, the convergence speed of the model is also greatly improved after the introduction of the PAR module.

The model based on object detection algorithm shows great advantages in accuracy and precision rate. However, due to the different sizes of the targets to be detected in the images, coupled with the interference of factors such as occlusion and scene complexity, at the same time, the object detection has the problem of too small a percentage, which is prone to miss detection and false detection. Therefore, the performance of the model needs to be further improved. In this paper, based on the shortcomings of existing models, we propose the Invo-YOLOv5s model, which can well improve the accuracy of detection and anti-interference ability. We conducted experiments on the model on the helmet dataset, and the experiments showed that the Invo-YOLOv5s model was selected for training, and the final obtained model detection accuracy reached 94.9%, which is 2.3% higher than the accuracy of the original YOLO model.

Aiming at the speed of frequent itemset mining, a new frequent itemset mining algorithm based on a linear table is proposed. The linear table can store more shared information and reduce the number of scans to the original dataset. Furthermore, operations such as pruning and grouping are also used to optimize the algorithm. For different datasets, the algorithm shows different mining speeds. (1) In sparse datasets, the algorithm achieves an average 45% improvement in mining speed over the bit combination algorithm, and a 2-3 times improvement for the classic FP-growth algorithm. (2) In dense datasets, the average improvement over the classic FP-growth algorithm is 50-70%. For the bit combination algorithm, there are dozens of times of improvement. In fact, the algorithm that integrates bit combinations with bitwise AND operation can effectively avoid recursive operations and it is beneficial to the parallelization. Further analysis shows that the linear table is easy to split to facilitate the data batch mining processing.

With the prevalence of deep learning, people use multi-modality information for interpretation and reasoning. In this paper, a cross-modality encoder CMEEA (cross-modality encoder representation based on external attention mechanism) is adopted to improve the accuracy of the model. The two external storage units of CMEEA can be regarded as a dictionary of the whole dataset to improve the performance of the network, while learning more representative features of the input and reducing the computational cost. This paper uses five pretraining tasks that help the model learn internal modal and cross-modal relationships. It also verifies the generalizability of the model by applying the pretraining cross-modal model to the visual inference task NLVR 2 , and improves the previous results by 0.1%. By fine-tuning the pretrain parameters, the model improves by 1.3% on visual question answering (VQA).

The spectroscopic Single-Molecule Localization Microscopy (sSMLM) is an emerging tool offering exciting new capabilities for single-molecule localization and tracking. By simultaneously capturing the spatial location and full spectra of fluorescent emission from single molecules, sSMLM can significantly extend the number of distinct target species for multiplexed super-resolution imaging and offer desirable resolving power for functional super-resolution imaging. However, extracting accurate spectral information in sSMLM remains challenging due to the poor signal-to-noise ratio (SNR) of spectral images set by limited photon budget from single-molecule fluorescent emission and inherent electronic noise during the image acquisition using digital cameras. Here, we developed an unsupervised learning-based method consisting of a convolutional autoencoder and an unsupervised clustering algorithm for high-fidelity spectral denoising and high-accurate molecular discrimination. Compared to existing learning-based spectral classification methods used in sSMLM, highly accurate molecular discrimination can be achieved even at low photon budget while high-SNR spectra can be simultaneously recovered without undermining the fidelity of the spectral details. Since the developed data processing flow can be directly integrated into any current sSMLM hardware, we hope the unsupervised spectral denoising and classification method can facilitate highly accurate molecular discrimination for multiplexed super-resolution imaging and single-molecule tracking.

In this era of rapid development of information science and technology, data coding technology has been widely used in multimedia, computer, communication and other fields. Non-deterministic combined coding is further explored in deterministic combined coding. It is a kind of lossless coding. In combinatorial mathematics, the operation of the combination number calculates the ordinal number of the sequence to be coded. On the basis of the deterministic combination coding, the adjacent identical elements are eliminated and processed separately, which reduces the length of the ordinal number. At the same time, the reference sequence is encoded to further save space. However, the ordinal calculation of the non-deterministic combination coding is still a complicated process, and the coding speed is still slow. This article uses openMP technology to parallelize the ordinal calculation part of the non-deterministic combination coding. The running time of coded ordinal on the eight-core CPU is 23.38% of the original, which is 4.28 times optimized, and the overall coding time efficiency of the 17 segmented file instances has been improved about 2.94 times. It indicates the effectiveness of the parallel strategy in this article. Finally, this article analyzes the experimental results.

提出了一种针对竹笛技巧分类的数据集Breath和两个用于竹笛技巧分类的神经网络参考模型Breath1d和Breath2d,并针对此数据集的不同分类任务给出了最佳方法.将Breath数据集划分成子集,以多层感知机为性能评价基准方法,先用Breath1d和Breath2d模型对子集进行训练和预测,再用长短期记忆网络模型进行辅助测试,最后得出了最适合子任务的分类参考模型.对全数据集进行分类时,将Breath2d与Breath1d模型进行融合,并采用数据增强方法使全集分类准确率达到0.913.与传统音频分类任务相比,该工作扩展了音乐分类的研究领域,对民族音乐现代化发展有着良好的推动作用.

In the inference process of existing deep learning models, it is usually necessary to process the input data level-wise, and impose a corresponding relational inductive bias on each level. This kind of relational inductive bias determines the theoretical performance upper limit of the deep learning method. In the field of sleep stage classification, only a single relational inductive bias is adopted at the same level in the mainstream methods based on deep learning. This will make the feature extraction method of deep learning incomplete and limit the performance of the method. In view of the above problems, a novel deep learning model based on hybrid relational inductive biases is proposed in this paper. It is called CCRRSleepNet. The model divides the single channel Electroencephalogram (EEG) data into three levels: frame, epoch, and sequence. It applies hybrid relational inductive biases from many aspects based on three levels. Meanwhile, multiscale atrous convolution block (MSACB) is adopted in CCRRSleepNet to learn the features of different attributes. However, in practice, the actual performance of the deep learning model depends on the nonrelational inductive biases, so a variety of matching nonrelational inductive biases are adopted in this paper to optimize CCRRSleepNet. The CCRRSleepNet is tested on the Fpz-Cz and Pz-Oz channel data of the Sleep-EDF dataset. The experimental results show that the method proposed in this paper is superior to many existing methods.

Non-deterministic combinatorial coding is a further study on the basis of deterministic combinatorial coding, so as to obtain a new lossless codingmethod. The non-deterministic combinatorial encoding is mainly to encode adjacent and different character sequences. The benchmark sequence of the encoding sequence is obtained in descending order according to the character frequency value. In the calculation process of the ordinal number, a proportional operation is used to replace the permutation and combination operation to reduce the huge amount of calculation, but with the length of the sequence increasing, the amount of calculation required to calculate the ordinal number is still large. The decoding process needs to test character one by one according to the benchmark sequence, and the amount of calculation is greater. In view of this situation, a new optimization scheme is proposed in this article, which makes the large number operations involved in the encoding/decoding process only take the higher weighted part of the nodes for operation, thus the amount of calculation is greatly reduced and the encoding/ decoding speed is improved. At the same time, it will not affect the correctness of the decoding. Experiments show that the optimization algorithm significantly improves the coding efficiency of non-deterministic combinatorial coding.

A new target detection network-Parameter Derivative Network is proposed for the first time in this paper, which is based on an anchor box-based detector. The parameter derivation module is the core of the network, and Its essence is to make use of image features to generate specific convolution kernels to extract the features of target objects. It solves the problem of reduced accuracy of small target objects in image detection. Compared with traditional convolution, this method can effectively obtain the context information of the target object and improve the detection accuracy. After the image features are extracted by the backbone network, they are sent to the parameter derivation module to perform a feature screening, which can effectively reduce the ratio between the foreground and the background in the feature map. It ensures the foreground context information and suppresses the background information at the same times. It is different from the traditional network feature extraction method. Firstly, the convolution kernel of the parameter derivation module is generated after the image is convoluted, and the convolution kernel is formed by processing the same channel, and then the feature map is convoluted. It can effectively suppress background information and reduce the imbalance of front background information. Secondly, in the parameter derivation module, the influence of average pooling and maximum pooling on image classification is verified in this paper. Compared with the maximum pooling, the average pooling has more local context information, which is more helpful for the detection and recognition accuracy of small target objects. Finally, the experimental results show that the parameter derivation module proposed in this paper is particularly effective in the detection and positioning of targets of different sizes without increasing any overhead, especially for small targets. At the same time, for targets of other sizes, this structure can also improve well and the maximum increase in detection performance is 0.36%.

Sleep disorders seriously affect human health and life. Accurate classification of sleep stages is the key to detecting and treating sleep disorders. In recent years, methods based on deep learning have surpassed traditional machine learning methods and human experts. However, the internal structure of deep learning is complex and requires to be designed by expert who is familiar with the computer and medical fields. This paper aims to analyze the relational induction biases in the existing sleep staging model based on deep learning, and explores the basic principles of sleep staging model design. This paper analyzes relational induction biases such as translation invariance, time invariance and hierarchical processing. Firstly, the model is divided into three categories according to whether it contains convolution layer with translation invariance and recurrent layer with time invariance: convolutional neural network framework, recurrent neural network framework and hybrid neural network framework. Secondly, according to the hierarchical processing method of the frame, segment and sequence in the model, a more detailed classification is carried out. Thirdly, it analyzes the impact of different relational induction biases in the model on the performance of sleep staging. It is proposed that a relational inductive bias matched the task should be introduced when the automatic sleep staging model is designed. Finally, the advantages and limitations of sleep staging based on deep learning are discussed, and it may be necessary to use more advanced relationship induction bias to express knowledge more abstractly and combine it with other artificial intelligence technologies in the future.