面向状态可变数据流的集群调度综述

状态可变数据流(Mutable States Data Flow,MS-DF)是机器学习系统运行时的主要特征,MS-DF可由有向图来表示,其顶点由算子构成,表示机器学习运算逻辑;边代表算子之间的输入输出依赖关系.MS-DF的集群调度是保障机器学习系统高效运行的主要工作,如何高效进行MS-DF的集群调度已经成为机器学习的研究热点.其中,机器学习系统(TensorFlow、PyTorch等)作为中间层解耦了机器学习运算逻辑和资源分配(CPU,GPU,FGPA),从而机器学习无需再“独占式”静态绑定资源,而是由机器学习系统运行时动态管理,而算子是该解耦过程的关键要素,这给MS-DF的集群调度带来了新的挑战,这些挑战主要由算子资源需求刻画的准确性、算子调度决策的适应性和算子调度调整的差异性这三方面导致的.首先介绍算子资源需求的感知、协同两个机制,以克服多种算子组合导致其自身资源需求难以准确刻画的挑战;然后,通过决策约束、决策模型和决策求解来介绍算子调度决策,以应对算子状态频繁变化带来的适应性挑战;接着,介绍迁移、伸缩、挂起恢复等算子调度调整策略,以适用于不同算子状态同步方式带来的差异性挑战.最后,基于上述三个挑战,对近年来的集群调度最新研究成果进行归纳和分析,并展望MS-DF的集群调度,指出算子异构资源需求多层次分析及协同刻画、算子复杂调度约束的灵活定义和发现、学习驱动的算子低成本调度调整技术是其主要发展方向.

A Learning Evasive Email-Based P2P-Like Botnet

Nowadays, machine learning is widely used in malware detection system as a core component. The machine learning algorithm is designed under the assumption that all datasets follow the same underlying data distribution. But the real-world malware data distribution is not stable and changes with time. By exploiting the knowledge of the machine learning algorithm and malware data concept drift problem, we show a novel learning evasive botnet architecture and a stealthy and secure C&C mechanism. Based on the email communication channel, we construct a stealthy email-based P2 P-like botnet that exploit the excellent reputation of email servers and a huge amount of benign email communication in the same channel. The experiment results show horizontal correlation learning algorithm is difficult to separate malicious email traffic from normal email traffic based on the volume features and time-related features with enough confidence. We discuss the malware data concept drift and possible defense strategies.

Distributed secure quantum machine learning

Distributed secure quantum machine learning （DSQML） enables a classical client with little quantum technology to delegate a remote quantum machine learning to the quantum server with the privacy data preserved. Moreover, DSQML can be extended to a more general case that the client does not have enough data, and resorts both the remote quantum server and remote databases to perform the secure machi~ learning. Here we propose a DSQML protocol that the client can classify two-dimensional vectors to dif- ferent clusters, resorting to a remote small-scale photon quantum computation processor. The protocol is secure without leaking any relevant information to the Eve. Any eavesdropper who attempts to intercept and disturb the learning process can be noticed. In principle, this protocol can be used to classify high dimensional vectors and may provide a new viewpoint and application for future ＂big data＂.

Variational algorithms for linear algebra

Quantum algorithms have been developed for efficiently solving linear algebra tasks.However,they generally require deep circuits and hence universal fault-tolerant quantum computers.In this work,we propose variational algorithms for linear algebra tasks that are compatible with noisy intermediate-scale quantum devices.We show that the solutions of linear systems of equations and matrix–vector multiplications can be translated as the ground states of the constructed Hamiltonians.Based on the variational quantum algorithms,we introduce Hamiltonian morphing together with an adaptive ans?tz for efficiently finding the ground state,and show the solution verification.Our algorithms are especially suitable for linear algebra problems with sparse matrices,and have wide applications in machine learning and optimisation problems.The algorithm for matrix multiplications can be also used for Hamiltonian simulation and open system simulation.We evaluate the cost and effectiveness of our algorithm through numerical simulations for solving linear systems of equations.We implement the algorithm on the IBM quantum cloud device with a high solution fidelity of 99.95%.