Search

Watermarking neural networks is a quite important means to protect the intellectual property (IP) of neural networks. In this paper, we introduce a novel digital watermarking framework suitable for deep neural networks that output images as the results, in which any image outputted from a watermarked neural network must contain a certain watermark. Here, the host neural network to be protected and a watermark-extraction network are trained together, so that, by optimizing a combined loss function, the trained neural network can accomplish the original task while embedding a watermark into the outputted images. This work is totally different from previous schemes carrying a watermark by network weights or classification labels of the trigger set. By detecting watermarks in the outputted images, this technique can be adopted to identify the ownership of the host network and find whether an image is generated from a certain neural network or not. We demonstrate that this technique is effective and robust on a variety of image processing tasks, including image colorization, super-resolution, image editing, semantic segmentation and so on.

The computer database that handles big data efficiently has created convenience for modern society, but it has gradually exposed more potential safety hazards in the application process. Computer databases storing massive data are prone to information loss, content tampering and other problems in the face of various attacks. With the development of the Internet and the continuous breakthrough of network technology, there are more and more hidden dangers that undermine network security, and the level and scope of damage are also increasing. This paper mainly describes the related concepts of intrusion detection technology, and puts forward the effective implementation of computer database intrusion detection technology based on virtualization technology, as well as the related strategies of fully optimizing computer database intrusion detection technology, so that the computer database can be effectively protected. Diversified intrusion detection system technology is widely used, which can not only ensure the safe and normal operation of database system, but also prevent the loss of important information and the destruction of structural integrity.

Due to the severe propagation loss nature of millimeter wave (mmWave), multi-hop relay transmission is a promising mmWave backhaul solution in ultra dense network. In this paper, we propose a multi-path routing scheme taking all small cells' backhaul into account to increase the system throughput. Under multi-hop constraint, we aim to optimize the downlink resource allocation of mmWave band. A two-objective link scheduling problem is formulated, where one objective is to ensure fairness among user equipments (UEs), and the other is to increase the system throughput. Since the two objectives are contradictory, we cannot obtain their optimal solutions simultaneously. To solve this problem, we propose a concurrent transmission based heuristic algorithm, which is called QoS (i.e., quality of service) -aware fairness scheduling (QFS). The proposed QFS scheme aims at transmitting as much data as possible in each time slot and scheduling appropriate links to guarantee the fairness. Simulation results show that QFS scheme can provide better fairness than conventional schemes. The percentage of UEs capable of achieving the minimum data rate requirements by the proposed QFS scheme is about 92.4%, which is greatly higher than those of the conventional schemes. In addition, the throughput of QFS scheme is about 3.18 times higher than Time Division Multiple Access (TDMA) scheme.

Due to the low cost and flexible deployment, wireless self-backhaul can be a promising backhaul solution for ultradense heterogeneous networks (HetNets). In this paper, we investigate the scheduling issues in wireless self-backhaul HetNets. To coordinate the inter-cell interference among small base stations (SBSs), a graph coloring algorithm is used to divide the SBSs into multiple SBS sets. SBSs belonging to the same set can receive or transmit data at the same time. We formulate a slot and sub-channel allocation optimization problem to maximize the downlink (DL) throughput of all SBSs subject to access-backhaul link matching constraints and user data rate requirements. Since the formulated problem is Integer Non Linear Program (INLP) whose solution is intractable, the solution is obtained by dividing the problem into two sub-problems. Firstly, a Linear Problem (LP) is formulated for solving the slot allocation problem, which roughly guarantee user data rate and access-backhaul link matching requirements. Secondly, two low-complexity heuristic algorithms for sub-channel allocation are proposed. Each base station (BS) will use corresponding sub-channel allocation algorithm in a distributed manner at working slot.

Due to the flexible deployment, in-band wireless backhaul solutions are preferred to the small cells in the ultradense networks (UDN). This paper manages the scheduling issues of in-band wireless backhaul in UDN by jointly considering the effect of access-backhaul link constraint, the interference and the minimum user data rate requirement. The scheduling problem that aims at maximizing the system throughput is formulated, whose solution is NP-hard. To reduce the calculation complexity and obtain a practical scheduling solution, A scheme named as coloring based access-backhaul scheduling (CBABS) is proposed. In CBABS, the minimum data rate requirements are satisfied by a QoS guarantee algorithm, the interference is managed by coloring algorithm (CA), the matching is done by ensuring the same data rate on access and backhaul link. Simulation results show that 98% of users' minimum downlink data rate requirements and 99% of users' minimum uplink data rate requirements are satisfied. The RB utility ratio is improved by 30% in CBABS compared to traditional scheduling scheme.