首页>
外国专利>
SOFTWARE ENGINEERING METHOD AND SYSTEM FOR ENABLING NEUROPLASTICITY, SELF-ADAPTIVE AND PROGRESSIVE MACHINE LEARNING, IN ARTIFICIAL INTELLIGENCE
SOFTWARE ENGINEERING METHOD AND SYSTEM FOR ENABLING NEUROPLASTICITY, SELF-ADAPTIVE AND PROGRESSIVE MACHINE LEARNING, IN ARTIFICIAL INTELLIGENCE
展开▼
机译:在人工智能中实现神经可塑性,自适应和渐进式机器学习的软件工程方法和系统
展开▼
页面导航
摘要
著录项
相似文献
摘要
The primary goal of this present invention is to realize the concept ofneuroplasticity in the artificial intelligence by enabling a computer softwaretoself-adapt to new discoveries of unanticipated or unpredictable events(learning through study) and to self-adjust its behaviour based on theoutcomesof earlier event processing for self-adaptive and progressive machine learning(learning from experience).This present invention introduces a novel software engineering approach,method, model, architecture and system based on a Dynamic MultidimensionalArray and associated programming model, method and system to allow acomputer software to self-adjust its behaviour with reflection of newdiscoveriesand based on the outcomes of earlier processing of events.The decision making trees and associated branch instructions in thetraditionalsoftware programs are replaced with indexed direct retrievals of a set ofactionsto be performed for processing of events from the Dynamic MultidimensionalArray. The Dynamic Multidimensional Array enables software programmes tomake decisions autonomously for handling business events, scientific triggersorenvironmental phenomena, without requiring any changes to the softwareprogrammes after deployment as they autonomously adapt to new discoveriesand earlier outcomes.This novel software engineering approach, method, model, architecture andsystem is used to enable self-adaptive and progressive machine learningthrough realization of neuroplasticity in artificial intelligence. Thispresentinvention obviates the need for the complex, error-prone and time-consumingbranch instructions that have been extensively used in the traditionalsoftwareprogrammes and have been attributed negatively to software safety, resilience,performance and scalability.The Information Technology has advanced significantly especially in the lasttwodecades of the 20th century. However, the fundamental principle, model andarchitecture of the computers remained exactly the same: a calculator or"number cruncher" based on the Von Neumann architecture. The computersoftware which is the brain of computers remained the same as well from theperspective of the fundamental principle, model and architecture. All thebusiness scenarios and associated processes need to be analyzed in advanceand predetermined for a series of calculation instructions.In this Information Age, we need a learning machine that can discoverunknown circumstances and self-adapt to them and continuously rebuild itsknowledge base with reflection of earlier outcomes, or "experiences" likehumanbrains rewire themselves with reflection of new information and experiences,i.e.,neuroplasticity.This topic of self-adaptive and self-managed systems has been widely studiedover the past five decades in a wide variety of application areas. The commongoal of these studies is to enable the software capabilities for autonomouslyadapting to various events and circumstances by automatically adjusting thebehaviours of software accordingly with the unanticipated and unpredictablesituations.The most critical issues in designing a self-adaptive system are that majorityofadaptations cannot be determined or predicted in advance and,consequently, that the requirements cannot be completely specified due tohigh degree of uncertainties.With the shared common goals and objectives, several nonprofit organizationsand academic and industrial consortia have been established globally toaddress these issues and challenges through an international collaboration.Among others, the "European Alliance for Innovation" has publicly solicitedpapers or proposals on Self-Adaptive Systems (SAS). In addition to academiaand research communities around the world, several consortia such as SoftwareEngineering for Self-Adaptive Systems (self-adaptive.org) and SoftwareEngineering for Adaptive and Self-Managing Systems (SEAMS) are calling forparticipation in self-adaptive, self-managing, self-healing, self-optimizing,self-configuring, and self-organizing systems.A preliminary search of prior art has yielded numerous academic and industrialarticles and papers related to self-adaptive systems or software, autonomoussystems, dynamic systems, cognitive systems, artificial intelligence, machinelearning, deep learning, etc. However, those articles and papers mainlydiscussthe issues and challenges associated with realizing the concept of theself-adaptive systems.There are some papers that propose models or methods related toself-adaptation of computer software. For example, the method for Self-AdaptiveSystem (SAS) and the model called the Policy-based Self-Adaptive Model(PobSAM) discuss different approaches for developing and specifyingself-adaptive systems by employing policies to govern and adapt the systembehaviour. However, the adaptation logic as well as configurations are fixedwithin the software programs and do not change dynamically after deploymentof the software.In addition, quite a few prior art related to the Supporting Vector Machine(SVM)are discovered through a preliminary search of prior art. They use theSupportingVector Machine (SVM) for Big Data analytics based on linear supervisedlearningby managing key words, phrases and information units in a common "vector".These technologies, therefore, are unrelated to this present invention at all.In summary, the methods and systems described in those prior art do notprovide the software capabilities for systematically adapting to unforeseensituations. The following two fundamental issues and challenges are yet to beresolved because:(1) Various events and their associated states are defined at a low-level ofabstraction, and the algorithms for adaptation and transition of states arehard-coded inside the software programs; and(2) Steady-state logics and configurations are fixed in the initialimplementationof software and cannot change dynamically during execution.The primary goal of this present invention is to address the fundamentalissuesand challenges mentioned above, in a completely different approach which isinspired by neuroplasticity of human brain. This present invention introducesamultidimensional array to dynamically manage the event-state associations andcorrelations. It also introduces a method and system for autonomously adaptingto various events and circumstances by adjusting the internal behaviours ofsoftware accordingly with the unanticipated and unpredictable situations. Moreimportantly, software autonomously adapts to various events andcircumstances without requiring any changes to the program logic afterdevelopment and deployment of the software.4
展开▼