End-to-end person search aims to jointly detect and re-identify a target person in raw scene images with a unified model. The detection task unifies all persons while the re-id task discriminates different identities,...
End-to-end person search aims to jointly detect and re-identify a target person in raw scene images with a unified model. The detection task unifies all persons while the re-id task discriminates different identities, resulting in conflict optimal objectives. Existing works proposed to decouple end-to-end person search to alleviate such conflict. Yet these methods are still sub-optimal on one or two of the sub-tasks due to their partially decoupled models, which limits the overall person search performance. In this paper, we propose to fully decouple person search towards optimal person search. A task-incremental person search network is proposed to incrementally construct an end-to-end model for the detection and re-id sub-task, which decouples the model architecture for the two sub-tasks. The proposed task-incremental network allows task-incremental training for the two conflicting tasks. This enables independent learning for different objectives thus fully decoupled the model for personsearch. Comprehensive experimental evaluations demonstrate the effectiveness of the proposed fully decoupled models for end-to-end person search.
The rapid development of Mobile Internet has spawned various mobile applications (apps). A large number of apps make it difficult for users to choose apps conveniently, causing the app overload problem. As the most ef...
详细信息
Blockchain has received great attention in recent years and motivated innovations in different scenarios. However, many vital issues which affect its performance are still open. For example, it is widely convinced tha...
详细信息
Web-based logs contain potentially useful data with which designers can assess the usability and effectiveness of their choices. Clustering techniques have been used to automatically discover typical user profiles fro...
详细信息
ISBN:
(纸本)9780769519326
Web-based logs contain potentially useful data with which designers can assess the usability and effectiveness of their choices. Clustering techniques have been used to automatically discover typical user profiles from Web access logs recently. But it is a challenging problem to design effective similarity measure between the session vectors, which are usually high dimensional and sparse. Nonnegative matrix factorisation approaches are applied to dimensionality reduction of the session-URL matrix, and the spherical k-means algorithm is used to partition the projecting vectors of the user session vectors into several clusters. Two methods for discovering typical user session profiles from the clusters are presented last. The results of experiment show that our algorithms can mine interesting user profiles effectively.
Evolutionary Multitasking (EMT) paradigm, an emerging research topic in evolutionary computation, has been successfully applied in solving high-dimensional feature selection (FS) problems recently. However, existing E...
详细信息
Existing persistent memory file systems usually store inodes in fixed locations, which ignores the external and internal imbalanced wears of inodes on the persistent memory (PM). Therefore, the PM for storing inodes c...
详细信息
ISBN:
(数字)9781728110851
ISBN:
(纸本)9781728158020
Existing persistent memory file systems usually store inodes in fixed locations, which ignores the external and internal imbalanced wears of inodes on the persistent memory (PM). Therefore, the PM for storing inodes can be easily damaged. Existing solutions achieve low accuracy of wear-leveling with high-overhead data migrations. In this paper, we propose a Lightweight and Multi-grained Wear-leveling Mechanism, called LMWM, to solve these problems. We implement the proposed LMWM in Linux kernel based on NOVA, a typical persistent memory file system. Compared with MARCH, the state-of-theart wear-leveling mechanism for inode table, experimental results show that LMWM can improve 2.5× lifetime of PM and 1.12× performance, respectively.
There are more than 10 million new stroke cases worldwide every year, and stroke has become one of the main causes of death and disability. In recent years, with the rapid development of computer science and technolog...
详细信息
ISBN:
(纸本)9781665421751
There are more than 10 million new stroke cases worldwide every year, and stroke has become one of the main causes of death and disability. In recent years, with the rapid development of computer science and technology, through the combination of Internet of things, deep learning, big data and other emerging technologies with traditional medicine, a new field of intelligent medicine has been developed. The scene of this paper is for stroke patients to use functional electrical stimulation equipment for rehabilitation training. By preprocessing the collected training data of MEMS patients, combined with the fully connected neural network (FCNN) model, the patient's upper limb posture can be intelligently recognized, which can make the intervention control of the rehabilitation system more efficient and intelligent. However, due to the damage of the stroke patients' action function, the existing sample data scale is small. In order to solve the problem of over fitting of network model caused by limited sample data in intelligent posture recognition, This paper proposes to expand the sample through data windowing operation to obtain a better performance recognition model.
In response to the challenges posed by nonindependent and identically distributed (non-IID) data and the escalating threat of privacy attacks in Federated Learning (FL), we introduce HyperFedNet (HFN), a novel archite...
详细信息
In an adaptive security-critical system, security mechanisms change according to the type of threat posed by the environment. Specifying the behavior of these systems is difficult because conditions of the environment...
详细信息
In an adaptive security-critical system, security mechanisms change according to the type of threat posed by the environment. Specifying the behavior of these systems is difficult because conditions of the environment are difficult to describe until the system has been deployed and used for a length of time. This paper defines the problem of adaptation in security-critical systems, and outlines the RELAIS approach for expressing requirements and specifying the behavior in a way that helps identify the need for adaptation, and the appropriate adaptation behavior at runtime. The paper introduces the notion of adaptation via input approximation and proposes statistical machine learning techniques for realizing it. The approach is illustrated with a running example and is applied to a realistic security example from a cloud-based file-sharing application. Bayesian classification and logistic regression methods are used to implement adaptive specifications and these methods offer different levels of adaptive security and usability in the file-sharing application.
暂无评论