Clustering is a fundamental problem in many data-driven application domains, and clustering performance highly depends on the quality of data representation. Hence, linear or non-linear feature transformations have been extensively used to learn better representation for clustering. In recent years, lot works focused using deep neural networks clustering-friendly representation, resulting significant increase performance. this paper, we give systematic survey with learning views...

Machine learning is one of the most prevailing techniques in computer science, and it has been widely applied image processing, natural language pattern recognition, cybersecurity, other fields. Regardless successful applications machine algorithms many scenarios, e.g., facial malware detection, automatic driving, intrusion these corresponding training data are vulnerable to a variety security threats, inducing significant performance decrease. Hence, vital call for further attention...

We propose a general purpose variational inference algorithm that forms natural counterpart of gradient descent for optimization. Our method iteratively transports set particles to match the target distribution, by applying form functional minimizes KL divergence. Empirical studies are performed on various real world models and datasets, which our is competitive with existing state-of-the-art methods. The derivation based new theoretical result connects derivative divergence under smooth...

A big convergence of language, vision, and multimodal pretraining is emerging. In this work, we introduce a general-purpose foundation model BEIT-3, which achieves excellent transfer performance on both vision vision-language tasks. Specifically, advance the from three aspects: backbone architecture, task, scaling up. We use Multiway Transformers for modeling, where modular architecture enables deep fusion modality-specific encoding. Based shared backbone, perform masked "language" modeling...

Session-based recommendation nowadays plays a vital role in many websites, which aims to predict users' actions based on anonymous sessions. There have emerged studies that model session as sequence or graph via investigating temporal transitions of items session. However, these methods compress into one fixed representation vector without considering the target be predicted. The will restrict ability recommender model, diversity and interests. In this paper, we propose novel attentive...

Modeling user preference from his historical sequences is one of the core problems sequential recommendation. Existing methods in this field are widely distributed conventional to deep learning methods. However, most them only model users' interests within their own and ignore dynamic collaborative signals among different sequences, making it insufficient explore preferences. We take inspiration graph neural networks cope with challenge, modeling sequence into framework. propose a new method...

Since sequential information plays an important role in modeling user behaviors, various recommendation methods have been proposed. Methods based on Markov assumption are widely-used, but independently combine several most recent components. Recently, Recurrent Neural Networks (RNN) successfully applied tasks. However, for real-world applications, these difficulty the contextual information, which has proved to be very behavior modeling. In this paper, we propose a novel model, named...

As we head towards the IoT (Internet of Things) era, protecting network infrastructures and information security has become increasingly crucial. In recent years, Anomaly-Based Network Intrusion Detection Systems (ANIDSs) have gained extensive attention for their capability detecting novel attacks. However, most ANIDSs focus on packet header omit valuable in payloads, despite fact that payload-based attacks ubiquitous. this paper, propose a intrusion detection system named TR-IDS, which...

We propose a simple algorithm to train stochastic neural networks draw samples from given target distributions for probabilistic inference. Our method is based on iteratively adjusting the network parameters so that output changes along Stein variational gradient maximumly decreases KL divergence with distribution. works any distribution specified by their unnormalized density function, and can black-box architectures are differentiable in terms of we want adapt. As an application our...

In practice, deep neural networks have been found to be vulnerable various types of noise, such as adversarial examples and corruption. Various defense methods accordingly developed improve robustness for models. However, simply training on data mixed with examples, most these models still fail defend against the generalized noise. Motivated by fact that hidden layers play a highly important role in maintaining robust model, this paper proposes simple yet powerful algorithm, named <italic...

Software-defined networking (SDN) is one of the most prevailing paradigms in current and next-generation networks. Basically, highly featured separation control data planes makes SDN a proper solution towards many practical problems that challenge legacy networks, for example, energy efficiency, dynamic network configuration, agile measurement, flexible deployment. Although its applications have been extensively studied several years, research security still infancy. Typically, suffers from...

We propose MaxUp, a simple and effective technique for improving the generalization performance of machine learning models, especially deep neural networks. The idea is to generate set augmented data with some random perturbations or transforms, minimize maximum, worst case loss over data. By doing so, we implicitly introduce smoothness robustness regu-larization against perturbations, hence improve generation performance. For example, in Gaussian perturbation, MaxUp asymptotically...

The hypersonic shock-shock interaction flow field at double-wedge geometries controlled by plasma synthetic jet actuator is experimentally studied in a Ma=8 high-enthalpy shock tunnel with the purpose of exploring novel technique for reducing surface heat flux real flight environment. results demonstrate that increasing discharge energy advantageous eliminating wave, shifting wave point, and shortening control response time. oblique can be completely removed when actuator's grows from 0.4 J...

In contemporary society, fatigue driving is a major cause of traffic accidents, making accurate and timely detection critical for improving safety. this study, we propose novel model, CSA-YOLO, designed to enhance the accuracy efficiency facial detection. The model based on YOLOv9s network introduces several key improvements address limitations traditional methods, which often lose edge information. First, Cross-Stage Partial Network (C3 module) replaces RepNCSPELAN4 module model’s ability...