数据驱动的应用自适应技术综述

应用自适应是软件工程和服务计算这一交叉领域的研究热点之一,应用通过感知自身和环境的变化,动态调整自己的行为与流程,以便在环境与需求发生非确定性变化的情况下继续高效地达成服务目标.近年来随着大数据和人工智能的发展,传统的基于软件模型控制的方法已经不再适用于当今动态和复杂的应用环境.相比而言,数据驱动的方法不依赖于数学模型和专家知识,而是以概率和数理统计为基础,通过应用服务运行的反馈数据，逐步学习和理解复杂多变的环境,继而学习出自适应系统的模型.因此，数据驱动的应用自适应具有感知性、适应性、自治性和协作性等特点,适用于流程复杂的应用服务场景,如物联网、智能交通、分布式计算等.从自适应框架出发,参考了认知计算的相关特点，总结出数据驱动的智能自适应框架，并分别综述了其中的表征学习、模式识别、决策规划和规则演化这4种技术在近几年数据驱动的自适应技术中的应用,重点探索了如机器学习、深度学习和强化学习等新技术在其中所起的作用,并总结和展望了自适应技术在服务计算领域的发展趋势.

Survey of Data-Driven Application Self-Adaptive Technology

Self-adaptation has always been a hot topic in the interdisciplinary field of software engineering and service computing. By perceiving the changes of themselves and the environment, applications dynamically adjust their behaviors and processes to continue achieving service goals efficiently under the circumstances of the non-deterministic changes of environment and requirements. With the recent development of big data and artificial intelligence(AI), traditional model-based control methods in software engineering are no longer suitable for dynamic and complex service computing environments nowadays. In contrast, the data-driven approach does not rely on mathematical models and expert knowledge but is based on probability and mathematical statistics. By applying the feedback data of service operation, the approach gradually learns and understands the complex and changeable environmental feedback, and then learns the model of the adaptive system. Therefore, the data-driven self-adaptive service computing has the characteristics of perceptibility, adaptability, autonomy and collaboration, etc. It is suitable for more complex application scenarios, such as the Internet of things, intelligent transportation and distributed computing. Based on the self-adaptive framework and the related characteristics of cognitive computing, a data-driven intelligent adaptive framework is proposed. And then, we have reviewed the application of representation learning, pattern recognition, decision planning and rule evolution in data-driven adaptive technology in recent years, respectively. It mainly explores the application of machine learning, deep learning and reinforcement learning in these technologies. And finally, it concludes the development of self-adaption and looks forward to the future trends.