Crowd counting is an important research topic in the field of computer vision. The multi-column convolution neural network (MCNN) has been used this and achieved competitive performance. However, when crowd distribution uneven, accuracy based on MCNN still needs to be improved. In order adapt uneven distributions, global density feature taken into account paper. features are extracted added through cascaded learning method. Because some detailed during down-sampling process will lost it...

The accurate prediction of PM2.5 concentration in a agricultural park is important to understand the role plays regulating pollution and guide public close nature healthily. An artificial neural network model was established, with meteorological data, atmospheric outside structure as input factors hourly average inside output factors. Its accuracy also evaluated this study. results show that it can be concluded BP promising approach predicting park.

In order to improve the performance of traditional intrusion detection system (IDS) based on neural network, we design and implement an integrated model IDS rough set wavelet network (RWNN-IDS). This paper focuses applying RWNN for attacks recognition. We first present a conditional information entropy algorithm select smallest features set, which can ensure correct classification. then novel network(WNN), its hidden unit is multi dimensional non-product wavelet-sigmoid basis function. A...

A novel covert communication method of digital image is presented, based on generalized fuzzy c-means clustering (GFCM), human visual system (HVS) and discrete cosine transform (DCT). Therefore, the original blocks are classified into two classes according to specified characteristic parameters. So one block suited for embedding security information, but other not. Hence appropriate can be selected in an embed information by selectively modifying middle-frequency part conjunction with HVS...

Building high-accuracy and efficient Bayesian network classifiers is a hot theme of classifier in recent years. It often the case that building unrestricted with large number attributes time-consuming always gets poor result, since searching space structure huge. This paper proposes clustering based learning algorithm(CBNA), which uses mutual information to measure distances between so as divide them into groups by hierarchical clustering. Then running under these low dimensional spaces...

Large language models have been flourishing in the natural processing (NLP) domain, and their potential for recommendation has paid much attention to. Despite intelligence shown by recommendation-oriented finetuned models, LLMs struggle to fully understand user behavior patterns due innate weakness interpreting numerical features overhead long context, where temporal relations among behaviors, subtle quantitative signals different ratings, various side of items are not well explored....

Conversational recommender systems (CRSs) aim to capture user preferences and provide personalized recommendations through multi-round natural language dialogues. However, most existing CRS models mainly focus on dialogue comprehension mining from the current session, overlooking in historical sessions. The embedded user's sessions session exhibit continuity sequentiality, we refer CRSs with this characteristic as sequential CRSs. In work, leverage memory-enhanced LLMs model preference...

CTR prediction plays a vital role in recommender systems. Recently, large language models (LLMs) have been applied systems due to their emergence abilities. While leveraging semantic information from LLMs has shown some improvements the performance of systems, two notable limitations persist these studies. First, LLM-enhanced encounter challenges extracting valuable lifelong user behavior sequences within textual contexts for recommendation tasks. Second, inherent variability human behaviors...

Recently, increasing attention has been paid to LLM-based recommender systems, but their deployment is still under exploration in the industry. Most deployments utilize LLMs as feature enhancers, generating augmentation knowledge offline stage. However, recommendation scenarios, involving numerous users and items, even generation with consumes considerable time resources. This inefficiency stems from autoregressive nature of LLMs, a promising direction for acceleration speculative decoding,...

Image segmentation and image denoising are two important fundamental topics in the field of processing. Geometric active contour model based on level set method can deal with problem segmentation, but it does not consider denoising. In this paper, a new diffusion equation for noisy is proposed by incorporating some classical models into segmental process. An assumption about connection between intensity function given firstly. Some employed to describe evolution secondly. The final nonlinear...

Imbalance of data sets is a widespread problem, and unbalanced has great impact on classification results. The traditional preprocessing methods based the imbalance mainly include under sampling over sampling. Oversampling problems fitting fuzzy boundary, method will discard useful information samples. In this paper, deep learning oversampling model proposed to solve above methods. uses generation algorithm, variational auto variable code learn features few samples in set, finally combines...

This paper presents a structure first image inpainting method based on self-organizing map (SOM). SOM is employed to find the useful information of damaged image. The which includes relevant edges used simulate lost or area in described by distinct indistinct curves an this paper. obtained target separate into several parts. As soon as each part restored respectively, inpainted. efficiency proposed demonstrated simulation results.

This paper puts forward a method of bringing neural network to bear intrusion detection. When the average error can't decrease any longer, hereditary algorithm will be used continuatively train in interest acquiring optimized join parameter. The structure and joining parameter evolve at same time by algorithm. convergence effect is good adaptivity strong, suitable for real-time processing.

Since the existing supervised learning has a strong dependence on real ground labeling and ignores importance of depth differences moving objects in image, an unsupervised homography estimation algorithm is proposed. Firstly, resnet34 backbone network constructed, two feature extraction modules with shared weights are used. Then, each initial module embedded Shuffle attention mechanism (SA), which used to extract features that can provide greater help for model training. Secondly, triple...

Change is certain, but the what, where and when are unpredictable. E.g. approaching animals need to be detected promptly, also classified as friend or foe. What best way carry out detection discrimination jointly? How do humans it? We compared subjects' performance in three tasks: (1) with unpredictable stimulus onset time, (2) known (3) dual task: time. Four subjects participated a novel random-dot motion task, trial begins incoherent coherent appears after random delay. For different...