ID 原文 译文
56348 新一轮科技革命和产业变革正在萌发,以深度学习和大数据为基础,以Alpha Go等为典型应用场景掀起了人工智能的第3次高潮. The dynamic bipartite drawing problem is a challenging NP-hard combinatorial optimization prob?lem with numerous applications. In this study, we propose a dynamic programming-based local search (DP-LS)algorithm for solving it.
56349 传统的基于统计线性化动态建模的人工智能,在处理复杂对象时遇到了可解释性、泛化性和可复现性等发展瓶颈,迫切需要建立基于复杂性与多尺度分析的新一代人工智能理论,我们称之为精准智能. Unlike previous metaheuristics reported in the literature, which move just one vertex(or two vertices) at each neighborhood iteration, the proposed DP-LS algorithm selects and performs multipleindependent neighborhood moves at the same time to enhance the search efficiency.
56350 针对复杂系统的非线性特征,精准智能构建内嵌领域知识和数学物理机理的系统学习理论,包括复杂数据科学感知、复杂系统精准构建、复杂行为智能分析3个层次. Generally, starting froma random initial solution, DP-LS iteratively explores the search space by integrating the DP-LS to locate localoptima, and then uses a perturbation procedure to escape from the local optima.
56351 具体而言,通过复杂数据科学感知建立内嵌时空特征与数理规律等具有可解释性的科学数据系统; In addition, the proposed incre?mental evaluation techniques of the insert and swap moves enhance the efficiency of the neighborhood evaluation.
56352 通过复杂系统精准构建反演具有非线性复杂逻辑关系的多层次、多尺度、可解释的人工智能动态学习模型; Extensive computational experiments on two sets of 1120 problem instances indicate that the proposed DP-LS ishighly competitive with the best-performing algorithms (including the general purpose solver Gurobi) in termsof both solution quality and computational efficiency.
56353 通过对系统复杂行为智能分析建立面向系统行为演进和全局动态分析的可解释可调控人工智能新理论和新方法. We analyzed the dynamic programming mechanism in theproposed DP-LS algorithm to determine its effectiveness (increasing the search efficiency tens of times).
56354 将上述精准智能理论应用于群体智能,提出了群体熵方法,实现了群体激发和汇聚行为复杂性度量与有效引导调控. Not onlycan the proposed DP-LS algorithm be used to solve the DBDP, it can also be used as a general methodology forsolving other combinatorial optimization problems, especially permutation optimization problems.
56355 生成式对抗网络(generative adversarial networks, GANs)训练的不稳定性问题一直是GANs研究领域最具挑战性的问题之一. Training instability in generative adversarial networks (GANs) remains one of the most challengingproblems, for which both the theoretical root and an effective solution are needed.
56356 目前,仍未从理论上找到影响GANs训练稳定性的根本原因及有效的解决办法. In this study, we theoreti?cally determined that the mutual contradiction between training the optimal discriminator and minimizing thegenerator leads to training instability in GANs.
56357 本文通过理论分析发现, GANs训练的不稳定性主要是由于训练最优判别器与最小化生成器之间相互矛盾所致. To address this problem, we propose a targeted gradient penaltytechnique.