Q-Learning Based Complex Program Dynamic Analysis Technology

Q-Learning is the representative of Reinforcement Learning. Because of prompt and efficient algorithm and self-adaptive learning, it is applied broadly and proved to be preponderant in optimizing dynamic process.In program analysis field, slice has being developed to maturity. SDG-based static slice is high efficiency and abroad in application. But, the problem must be taken into account, that how to analyze dynamic complex programs in indeterminate environments. The character is much serious that software programs are become more complex and easier affected by environments. So, the requirement of solving these problems becomes more urgently. At the same time, the capability of traditional program analysis theory and technology is badly restricted.It's a bravely attempt to explore program analysis with Q-Learning strategy. The paper introduces the advantage of Q-Learning strategy, aiming at difficulties of complex programs dynamic analysis, puts an idea of Q-Learning based complex program analysis, constructs feasible analysis algorithms, focuses and analyzes concurrent programs, object-oriented programs, and complex programs in indeterminate environments, forms Q-Learning based complex program analysis technology, resoluble idea and approach as well as reliable basis is put forward for solving the problem of complex program dynamic control. Basing above researches, the paper studies the realization of the Q-Learning based complex program analysis and control agent (QACA).The main works of the paper includes three aspects as following. Firstly, through technology study of the Q-Learning based complex program analysis, some hard questions are solved in concurrent programs, object-oriented programs and complex programs in indeterminate environments, intelligence, efficiency and accuracy enhanced. Secondly, on the basis of Q-Learning based complex program analysis, the paper studies the complex program optimization control, boosts up the capability of complex program control automatically, offers feasible idea and idiographic method for improving safety and enhancing quality of those large-scale programs or important software system. Thirdly, the paper explores and experiments the realization of the QACA system, especially to the crucial step and relevant technology.The main contributions of the paper are listed below.Q-Learning based dynamic program analysis concept and approach is put forward. Its primary feasible technology comes into being.New series concepts are introduced at first time, such as Q dependence link (QDL) and Q dependence graph (QDG). These concepts combine essentials of the two fields'knowledge, and intercompare the two fields'concepts and learn from each strong point to offset the other's weakness.Novel approach is put in, that is analyzing after distinguishing characters and discriminating management for those complex programs. Not only improves the learning and analyzing efficiency of QACA, but also enhances the pertinency and accuracy of analysis and control.For the actual rapidity and effect of the QACA learning, training and timely analyzing, the paper proposes a new program express and treatment through taking the advantage of traditional and developed program analysis theory and technology, information quantity restricted and process periods shortened.Proposes and realizes an idea and a method that real time analysis and control over the complex program under optimization strategy, thinks through and designs a new work style that embeds the QACA system into the large scale important software system dynamic course, basic study to the realization is fulfilled.