Machine learning in adversarial environments

Whenever machine learning is used to prevent illegal or unsanctioned activity and there is an economic incentive, adversaries will attempt to circumvent the protection provided. Constraints on how adversaries can manipulate training and test data for classifiers used to detect suspicious behavior make problems in this area tractable and interesting. This special issue highlights papers that span many disciplines including email spam detection, computer intrusion detection, and detection of web pages deliberately designed to manipulate the priorities of pages returned by modern search engines. The four papers in this special issue provide a standard taxonomy of the types of attacks that can be expected in an adversarial framework, demonstrate how to design classifiers that are robust to deleted or corrupted features, demonstrate the ability of modern polymorphic engines to rewrite malware so it evades detection by current intrusion detection and antivirus systems, and provide approaches to detect web pages designed to manipulate web page scores returned by search engines. We hope that these papers and this special issue encourages the multidisciplinary cooperation required to address many interesting problems in this relatively new area including predicting the future of the arms races created by adversarial learning, developing effective long-term defensive strategies, and creating algorithms that can process the massive amounts of training and test data available for internet-scale problems.     

Collision detection technique for multiple access protocols on radio channels

The carrier sense multiple access with collision detection protocol (CSMA/CD) has already gained wide commercial acceptance and is used in cable and optical-fibre based local computer networks. However, its use in radio channels is inhibited by the difficulty of sensing remote carriers in the presence of local transmission. A new collision detection technique for radio channels called time-split collision detection is introduced. Collision is detected by a sequence of carrier sense, preamble transmission and carrier sense again at the beginning of a packet transmission. Throughput analysis of a new multiple access protocol using time-split collision detection is also presented. The proposed protocol has throughput performance between those of the CSMA and CSMA/CD.     

Data broadcasting for dependent information using multiple channels in wireless broadcast environments

Data broadcasting is an effective approach to disseminating information to mobile clients and has attracted much research attention in recent years. In many applications, the access pattern among the data can be represented by a weighted DAG. In this paper, we consider the problem of efficiently generating the broadcast schedules on multiple channels when the data set has a DAG access pattern. We show that it is NP-hard to find an optimal broadcast schedule which not only minimizes the latency but also satisfies the ancestor property that retains the data dependency. We further derive a condition for the input DAGs under which one can generate an optimal broadcast schedule in linear time and propose an algorithm to generate the schedule. Due to the NP-completeness, we provide three heuristics for general DAGs based on the level of a vertex in the input DAGs and each heuristic uses a different policy to place vertices into the broadcast channels. There are two categories for the policies. The first category mainly considers the probability for a process to stop at a considered vertex. The second category takes the vertices which are affected most when assigning a vertex into consideration. We analyze and discuss these heuristics. A short experimental simulation is given for supporting and validating the discussion. In particular, the experimental results indicate that roughly considering the whole posterior vertices of each vertex is not precise and may not lead to good results and considering the vertices affected most when assigning a vertex will help reducing the latency.     

Cooperative multiple access channels: Achievable rates and optimal resource allocation

We investigate the multiple access channels (MAC) where sources can cooperate via half-duplex relaying and refer to it as cooperative MAC channels (CMAC). Assuming perfect channel state information (CSI) at the transmitters and the receivers, we determine the bounds on the achievable rate region of a Gaussian CMAC channel and an inner bound on the outage capacity region of a fading CMAC channel. Based on superposition modulation, a half-duplex cooperative relay scheme with optimal resource allocation is proposed to achieve the bounds of capacity region. Analytical results and simulation results show that the achievable rate region of a Gaussian CMAC channel is larger than that of a Gaussian MAC channel with direct transmission (DT) schemes. But they have the same achievable sum rate. Moreover, the proposed scheme can provide higher outage capacity region than DT schemes in a fading MAC channel due to the fact that sources can share the resources with each other to reduce outages. Supported partially by the National Natural Science Foundation of China (Grant No. 60672079), the Natural Science Foundation of Jiangsu Province (Grant No. BK2006701), and the Natinoal High-Tech Research & Development Program of China (Grant No. 2007AA01Z267)     

A tree-structured index allocation method with replication over multiple broadcast channels in wireless environments

Broadcast has often been used to disseminate frequently requested data efficiently to a large volume of mobile units over single or multiple channels. Since mobile units have limited battery power, the minimization of the access and tuning times for the broadcast data is an important problem. There have been many research efforts that focus on minimizing access and tuning times by providing indexes on the broadcast data. We have studied an efficient index allocation method for broadcast data with skewed access frequencies over multiple physical channels which cannot be coalesced into a single high bandwidth channel. Previously proposed index allocation techniques have one of two problems. The first problem is that they require equal size for both index and data. The second problem is that their performance degrades when the number of given physical channels is not enough. These two problems result in an increased average access time for the broadcast data. To cope with these problems, we propose a tree-structured index allocation method. Our method minimizes the average access time by broadcasting the hot data and their indices more frequently than the less hot data and their indexes over the dedicated index and data channels. We present an in-depth experimental and theoretical analysis of our method by comparing it with other similar techniques. Our performance analysis shows that it significantly decreases the average access and tuning times for the broadcast data over existing methods.     

A robust analysis of adversarial attacks on federated learning environments

Federated Learning is a growing branch of Artificial Intelligence with the wide usage of mobile computing and IoT technologies. Since this technology uses distributed computing paradigm to do the learning part, most of the participating components are mobile devices and come outside the range of protection offered by a centralized system. As a result, several security issues such as data leakage, communication issues, poisoning, system manipulation via the backdoor, and so on arise with the usage of such a methodology. These sorts of attacks are categorized into various categories concerning their modus operandi. In this study, we review such attacks, namely poisoning attacks, inferencing attacks, their types, and working in a Federated Learning environment in detail. This study will give a precise idea of security issues faced in Federated Machine Learning and possible solutions.     

Designing robust network topologies for wireless sensor networks in adversarial environments

In this paper, we address the problem of deploying sink nodes in a wireless sensor network such that the resulting network topology be robust. In order to measure network robustness, we propose a new metric, called persistence, which better captures the notion of robustness than the widely known connectivity based metrics. We study two variants of the sink deployment problem: sink selection and sink placement. We prove that both problems are NP-hard, and show how the problem of sink placement can be traced back to the problem of sink selection using an optimal search space reduction technique, which may be of independent interest. To solve the problem of sink selection, we propose efficient heuristic algorithms. Finally, we provide experimental results on the performance of our proposed algorithms.     

On the weakest resource for coordination in arbitrarily varying multiple access channels with conferencing encoders

If senders and a receiver of an arbitrarily varying multiple-access channel (AV-MAC) have access to outputs of discrete correlated memoryless sources, the same rate region is achievable as if common randomness were available, no matter how small the correlation is. This reduces the necessary amount of cooperation in an AV-MAC considerably. Moreover, to transmit blocklength-n words, no more than order log n source outputs are required.     

Adaptive data access in broadcast-based wireless environments

Power conservation and client waiting time reduction are two important aspects of data access efficiency in broadcast-based wireless communication systems. The intention of data access methods is to optimize client power consumption with the least possible overhead on client waiting time. We propose an adaptive data access method which builds on the strengths of indexing and hashing techniques. We show that this method exhibits better average performance over the well-known index tree-based access methods. A new performance model is also proposed. This model uses more realistic assessment criteria, based on the combination of access and tuning times, for evaluating wireless access methods. This new model provides a dynamic framework to express the degree of importance of access and tuning times in an application. Under this new model, the adaptive method performance also outperforms the other access methods in the majority of cases.     

A coding theorem for multiple access channels using a maximum likelihood list decoding scheme

Bounds on the probability of error for list decoding, in which the receiver lists L messages rather than one after receiving a message, have been obtained by Elias, Ebert, and Shannon, Gallager and Berlekemp. In the present paper the problem of list decoding has been considered for a multiple access channel. The paper contains an upper bound on the probability of error for list decoding in this case.     

High throughput MAC layer multicasting over time-varying channels

Efficient, scalable and robust multicasting support from the MAC layer is needed for meeting the demands of multicast based applications over WiFi and mesh networks. However, the IEEE 802.11 protocol has no specific mechanism for multicasting. It implements multicasting using broadcasting at the base transmission rate. We identify two fundamental reasons for performance limitations of this approach in presence of interference and realistic time-varying channels: (a) Channel-state Indifference: irrespective of the current quality of the channel to the receivers, the transmission always uses the base transmission rate; (b) Demand Ignorance: packets are transmitted by a node even if children in the multicast tree have received those packets by virtue of overhearing. We propose a solution for MAC layer multicasting called HIMAC that uses the following two mechanisms: Unary Channel Feedback (UCF) and Unary Negative Feedback (UNF) to respectively address the shortcomings of 802.11. Our study is supported by measurements in simulations. We observe that the end-to-end throughput of multicast sessions using MAODV can be increased by up to 74% while reducing the end-to-end latency by up to a factor of 56.     

Cooperative target pursuit by multiple UAVs in an adversarial environment

This paper presents the development of a cooperation strategy for multiple UAVs to pursue a target moving in an adversarial environment where threat exposure should be minimized, and obstacles and restricted areas should be avoided. A probabilistic approach is used to model the adversarial environment. A cost function is defined to quantify placement of UAVs around the target in formation in terms of threat exposure level and distance to the target. The cost function is used to develop a cooperation strategy for a team of UAVs to follow the target such that the total threat exposure of the team and the average distance to the target throughout the pursuit are minimized according to the weighting coefficients specified. The cooperation strategy has the feature of collision avoidance as well as data-fusion-based estimation of the target trajectory based on noisy measurements. Simulation results have demonstrated that the cooperation reduces the risk of losing the target during the pursuit while avoiding obstacles and restricted areas. Further, the UAVs guided by the cooperation strategy can follow the target closer without increasing the total threat exposure level as compared to cases where the UAVs pursue the target without cooperation.     

A hybrid access method in a multiple access channel

This paper considers a multiple-access communication channel with an infinite number of users. We show that if a controlled slotted Aloha protocol is used, then messages with variable length will have a negative impact on the average message delay. In order to alleviate this problem, a mixed mode (Hybrid) access method is suggested under which the channel bandwidth is split into two sub-channels, managed under different policies. Messages whose length is less than or equal to a critical value are transmitted in one sub-channel under a slotted Aloha policy. The rest of the messages are sent through a separate portion of the channel bandwidth, using a reservation protocol. We show that under this hybrid access method the average delay of a message is greatly improved.     

Simulating multiple character interactions with collaborative and adversarial goals

This paper proposes a new methodology for synthesizing animations of multiple characters, allowing them to intelligently compete with one another in dense environments, while still satisfying requirements set by an animator. To achieve these two conflicting objectives simultaneously, our method separately evaluates the competition and collaboration of the interactions, integrating the scores to select an action that maximizes both criteria. We extend the idea of min-max search, normally used for strategic games such as chess. Using our method, animators can efficiently produce scenes of dense character interactions such as those in collective sports or martial arts. The method is especially effective for producing animations along story lines, where the characters must follow multiple objectives, while still accommodating geometric and kinematic constraints from the environment.     

A trust degree based access control in grid environments

The purpose of grid computing is to enable coordinated resource sharing and support cooperative work between different domains in dynamic grid environments. In order to protect each participant’s privilege and security, a secure and efficient access control is essential. This paper presents a new approach of access mechanism based on trust relationships across domains. A new calculation method of trust in grid is proposed and the difference between intro-domain trust and inter-domain trust is analyzed. In addition, a novel access control framework combined with trust degree is given from this proposal. It is shown to be adaptive for both intro-domain and inter-domain conditions. Hence, a prototype system based on the proposed model is introduced; furthermore, it has been shown as a dynamic and fine-granularity access control method through performance analyses and has also been demonstrated as a suitable system for grid environments.     

Full-duplex routing with low-complexity sequential-decision for throughput enhancement in dynamic access networks

Integrating the Full-Duplex (FD) communication technology in multi-hop Cognitive Radio Networks (CRNs) can significantly enhance routing decisions such that overall network performance is improved. Unfortunately, when employing the FD technology in multi-hop routing, the channel assignment becomes very challenging due to the possible simultaneous transmissions/reception. Hence, careful low-complexity channel assignment over the various hops along a path is needed to avoid hop-to-hop interference and enhance network performance. In this paper, we propose a low-complexity sequential-decision routing protocol, referred to as Sequential FD (SFD)-aware protocol. The SFD protocol identifies a set of possible routes between any source–destination communicating pair. For each discovered route, SFD successively assigns channels for each hop along the router such that the maximum route capacity is achieved. Then, SFD selects the routes with the highest capacity. Unlike most of existing FD-CRN channel assignment/routing protocols, which require optimization procedures with NP-hard time complexity with the number of channels and hops, the time-complexity of the proposed SFD protocol is polynomial. Compared to the optimal FD-aware routing (obtained using the brute-force search method) and maximum per-hop rate FD-unaware routing protocols, the simulation results confirm the effectiveness of our proposed protocol and its superiority over the existing FD-unaware baseline routing protocol in terms of the end-to-end network throughput. The results also confirm that the SFD protocol provides solutions within 5% of the optimal ones.     

基于复合生成对抗网络的对抗样本生成算法研究

对抗样本能够作为训练数据辅助提高模型的表达能力,还能够评估深度学习模型的稳健性.然而,通过在一个小的矩阵范数内扰乱原始数据点的生成方式,使得对抗样本的规模受限于原始数据.为了更高效地获得任意数量的对抗样本,探索一种不受原始数据限制的对抗样本生成方式具有重要意义.鉴于此,提出一种基于生成对抗网络的对抗样本生成模型(multiple attack generative adversarial networks, M-AttGAN).首先,将模型设计为同时训练2组生成对抗网络,分别对原始数据样本分布和模型潜在空间下的扰动分布进行建模;然后,训练完成的M-AttGAN能够不受限制地高效生成带有扰动的对抗样本,为对抗训练和提高深度神经网络的稳健性提供更多可能性;最后,通过MNIST和CIFAT-10数据集上的多组实验,验证利用生成对抗网络对数据分布良好的学习能力进行对抗样本生成是可行的.实验结果表明,相较于常规攻击方法,M-AttGAN不仅能够脱离原始数据的限制生成高质量的对抗样本,而且样本具备良好的攻击性和攻击迁移能力.     

移动自组网媒体接入控制协议吞吐量与公平性均衡设计

针对移动自组网媒体接入控制(MAC)协议高吞吐量、低公平性的失衡问题,提出一种基于最优接入概率的简化协议MAC-FT.首先推导公平条件下最优吞吐量与节点数、节点数与空闲接入概率的定量关系,设计空闲接入概率评估模型,利用李雅普诺夫漂移函数证明模型的可行性和稳定性;其次利用自回归滑动平均(ARMA) 模型滤波实现空闲时隙接入概率的计算,并通过比例积分控制器(PIC)完成动态控制;最后综合分析吞吐量和公平性等性能.实验结果表明MAC-FT公平索引值为0.98,吞吐量为6.15 Mb/s,接近最优值1和5.85 Mb/s,比渐进最佳回退(AOB)、空闲感知(IS)、协议分布式协调(DCF)、改进协议启发式缓变协议(GDCF)性能更优,实现吞吐量和公平性的均衡.     

Adaptive rate coding using convolutional codes for asynchronous code division multiple access communications over slowly fading channels

This paper presents a method of code rate adaptation using punctured convolutional codes for direct sequence spread spectrum communication systems over slowly fading channels. A blind channel estimation technique is used to estimate the nature of the multi-user channel at the detector (before the decoder). The path gains obtained from the channel estimation technique are used to adapt the code rates. Punctured codes derived from a specific rate 1/2 (M = 4) mother code are used to provide error protection corresponding to the actual channel state. The upper and lower bounds on the bit error probability and the upper bound on the error event probability are derived for hard-decision and soft-decision decoding over Rayleigh and Rician fading channels. The throughput gains obtained using the adaptive scheme and the performances of the punctured codes are studied.