一种深度强化学习空间关系与记忆融合方法研究

深度强化学习结合了深度学习在视觉上强大的感知能力来解决复杂环境的序列决策问题，但是由于采样效率低，对于复杂高维数据输入，学习其重要特征较为困难.为了从序列样本中更有效地提取信息，本文提出在深度强化学习中融合空间关系推理和记忆推理(Spatial Relationship Reasoning and Memory Reasoning,SRRMR)的模型结构.模型分为空间关系推理和记忆推理两部分，空间关系推理使用注意力机制作为空间关系学习方法隐式地推理任意两个实体间的关系，注意力机制中的查询向量融合了记忆推理的内容;记忆推理将输入图像的特征和关系作为记忆的输入，利用自注意力与记忆组成部分进行推理和交互，并将交互的结果存储在记忆单元中，使得记忆存储单元融合了空间信息与记忆信息.SRRMR模型在不同种类的Atari游戏中进行了训练和验证，结果表明，空间关系推理与记忆推理的融合在7/15个游戏环境中以更少的交互次数收敛到更好的结果，记忆推理网络在12/15个游戏中获得提升，提升智能体学习效率，更高效地利用序列中的样本，提高了强化学习的样本利用率.     

基于序演算的推理系统OCRS及其判定算法

序演算可以保留很多语言学的信息,可以更自然的刻画日常交流中的推理,因此序演算成为很多自然逻辑推理系统的核心方法.在Zamansky等学者工作的基础上,以序演算为核心,结合Lambek演算,增加了性质类序关系,考虑了模糊量词的语义特征,构造了基于序演算的自然逻辑推理系统OCRS.该系统所描述的推理范围较之以前的自然逻辑系统进一步扩大.另外,也给出了OCRS系统中序演算的判定算法,证明了其判定算法的对应定理.     

基于序关系的非完备信息案例推理集成方法研究

针对案例推理系统中数据集存在数据缺失的非完备信息问题,利用序关系基本原理,设计了案例推理集成方法(ORCBR)。通过对非完备信息下确定符号属性、确定数值属性、区间数值属性以及模糊语言属性等属性间相似性度量的研究,计算出目标案例与历史案例的相似性矩阵。在此基础上,利用序关系构建了相似性矩阵中不同属性的集成排序算法,从而得到最相似历史案例。通过对UCI数据库中非完备信息数据集的测试表明,OR-CBR方法比经典案例推理方法准确率高、效率高,很好地解决了非完备信息数据集的案例推理问题。     

基于推理网络的文本检索模型

在研究现有文本信息检索技术的基础上,设计了基于推理网络的文本检索模型.提出一种改进的推理算法,以实现从文档观察事件到索引词出现事件的推理,使新模型可以更全面地利用文本数据信息.最后通过一个推理网络实例来说明实现推理的数学过程.     

对抗样本训练图分类器进行模型推理质量评估

重叠社区发现是社交网络分析与挖掘中的一个重要研究问题,现有的大部分方法都要求采用人工方法预先设定社区个数[K],这样做存在很多问题。将无限潜特征模型推广应用到关系型数据,以非参数贝叶斯层次模型为框架建立带重叠社区结构的网络生成模型,就可以避免预先设定[K]的值。然而,关系型无限潜特征模型的后验参数推理结果是一个[N×K]列的0、1矩阵上的概率分布,如何对这种多变量结构参数进行后验推理结果总结和后验推理质量评估是一个挑战,因此提出了一种利用基于对抗样本训练图卷积神经网络的图分类器来帮助总结推理结果和估计推理质量的方法。     

不确定时态信息表示的统一模型

时态信息表示和推理是人工智能研究中的一个重要课题,现有的模型大多只能表示确定时态信息,然而现实生活中很多事件的发生结束等时态信息都是不确定的。故提出了一个表示不确定时态信息的统一模型,可用于描述各种具有确定或不确定时态信息的事件。该模型首先定义各类时态对象(如时间点、时间区间)以及它们之间的关系,并给出时态对象间的传递关系表,利用该表能进行时态一致性约束满足问题的求解。最后,给出了两个不确定时态推理的例子,表明了该模型的实际应用意义。     

基于MBR的主方向关系一致性检验

定性的空间推理在地理信息系统、人工智能、数据库及多媒体等领域中的应用越来越引起人们的注意.空间推理的基础理论以及相应算法也在不断地创新和发展.方向关系推理是空间推理研究领域的重要分支,利用区间代数及矩形代数理论,以物体的极小边界盒(minimum bounding rectangle,简称MBR)为模型,提出了一种基于MBR的主方向关系与矩形代数关系相结合的推理方法.利用该方法,可以将矩形代数良好的计算性质应用于空间方向关系推理中,实现了矩形代数与基于MBR主方向关系的相互转换方法、主方向关系合成及求反方法、主方向关系中凸(convex)关系判定方法及方向关系一致性检验算法.     

一种基于正向云变换的混合计算神经网络及其应用

针对数值信息与定性领域知识相互融合的计算问题,提出了一种基于云变换的混合计算神经网络模型。利用正向正态云发生器可实现定性概念到量化数值描述之间不确定关系的转换机制,建立基于云变换的混合信息计算逻辑和神经网络模型。将定性概念谓词通过云变换在概率意义下转换为数值变量,把计算规则表示为神经元,利用神经网络的学习性质来实现对定量与定性混合信息的自适应计算和推理。在算法设计中,将网络性质参数整合为一个粒子,利用粒子群算法对混合计算神经网络进行整体优化求解。以石油地质研究中的沉积微相自动识别为例,实验结果验证了模型和算法的有效性。     

扩展时段时序逻辑的推理机制

该文在扩展时段时序逻辑的基础上提出了一种推理机制，这种推理机制基于时间Petri网模型及基本不等式规则，可由一组已知的扩展时段时序关系推出一些未知的扩展时段时序关系，对不确定时间段内发生的事件及其相互关系具有较好的描述能力，这种推理机制的优势在于定性地对扩展时段之间的时序关系进行推理分析，利用时间Petri网模型，可以对复杂时序逻辑关系进行化简，比单纯利用不等式规则的推理更直观，也更简单，是一种行之有效的方法。     

Credal网络推理的一种不完全枚举法

Credal网络是研究不确定环境下知识表示和因果推理的一种图模型,其条件概率值可以用不精确的区间或不等式定性地表示,使得表达方式更加灵活有效。Credal网络的推理是计算一定证据下的后验概率最大值和最小值,给出了一种Credal网络推理的新方法,该方法是在桶消元框架下通过枚举计算部分因子函数值,使计算量大大减小,并且可以得到精确的结果。最后用一个实例说明了该方法的可行性。     

A Fuzzy Inference Network Model for Search Strategy Using Neural Logic Network

In this paper, a fuzzy inference network model for search strategy using neural logic network is presented. The model describes search strategy, and neural logic network is used to search. Fuzzy logic can bring about appropriate inference results by ignoring some information in the reasoning process. Neural logic networks are powerful tools for the reasoning process but not appropriate for the logical reasoning. To model human knowledge, besides the reasoning process capability, the logical reasoning capability is equally important. Another new neural network called neural logic network is able to do the logical reasoning. Because the fuzzy inference is a fuzzy logical reasoning, we construct a fuzzy inference network model based on the neural logic network, extending the existing rule inference network. And the traditional propagation rule is modified.     

Probabilistic Argumentation Systems and Abduction

Probabilistic argumentation systems are based on assumption-based reasoning for obtaining arguments supporting hypotheses and on probability theory to compute probabilities of supports. Assumption-based reasoning is closely related to hypothetical reasoning or inference through theory formation. The latter approach has well known relations to abduction and default reasoning. In this paper assumption-based reasoning, as an alternative to theory formation aiming at a different goal, will be presented and its use for abduction and model-based diagnostics will be explained. Assumption-based reasoning is well suited for defining a probability structure on top of it. On the base of the relationships between assumption-based reasoning on the one hand and abduction on the other hand, the added value introduced by probability into model based diagnostics will be discussed. Furthermore, the concepts of complete and partial models are introduced with the goal to study the quality of inference procedures. In particular this will be used to compare abductive to possible explanations.     

Logical radial basis function networks a hybrid intelligent model for function approximation

The aim of this article is to introduce a new approach for fuzzy neural network models which can be used effectively in function approximation problems. The proposed model is introduced as an adaptive two-level fuzzy inference system. The architecture of the model is basically a two-layer network of new types of fuzzy-neurons which act as fuzzy IF–THEN rules. The model can be considered as a logical version of the Radial Basis Function networks (RBF). Genetic Algorithms have been adopted as the learning mechanism of the proposed model. Simulations show both highly nonlinear mapping and reasoning capabilities together with simpler structure and better performance when compared with classical neural networks.     

基于知识表示学习的实时语义数据流推理

传统的语义数据流推理使用前向或后向链式推理产生确定性的答案,但是在复杂的传递规则推理中效率不高,无法满足实时数据流处理场景对答案的及时性要求。因此,提出一种基于联合嵌入模型的知识表示方法,并应用于语义数据流处理中。将规则与事实三元组联合嵌入并利用深度学习模型进行训练,在推理阶段,根据查询中涉及的规则建立推理模板,利用深度学习模型对推理模板产生的三元组进行预测和分类,将结果作为查询和推理答案输出。实验表明,对于复杂规则推理,基于知识表示学习的实时语义数据流推理能够在保障较好推理准确性和命中率的前提下有效地降低延迟。     

Android隐式信息流检测的本体模型

针对传统的信息泄漏检测技术无法有效检测Android应用中存在的隐式信息泄露的问题,提出了一种将控制结构本体模型与语义网规则语言（SWRL）推理规则相结合的Android隐式信息流（ⅡF）推理方法。首先,对控制结构中能够产生隐式信息流的关键要素进行分析和建模,建立控制结构本体模型;其次,通过分析隐式信息泄露的主要原因,给出基于严格控制依赖（SCD）隐式信息流的判定规则并将其转换为SWRL推理规则;最后,将添加的控制结构本体实例与推理规则共同导入到推理引擎Jess中进行推理。实验结果表明：所提方法能够推理出多种不同性质的SCD隐式流,公开样本集的测试准确率达到83.3%,且推理耗时在分支数有限时处于合理区间。所提模型方法可有效辅助传统信息泄露检测提升其准确率。     

一种应用于故障诊断中的高效推理算法

针对经典联合树推理算法的信息传播共享和推理时间等问题, 提出一种高效联合树推理算法. 该算法基于获得的证据信息和查询节点对原始的网络结构化简, 然后在化简后的网络结构上进行联合树推理. 在信息传递过程中, 该算法可以实现不同证据下的信息共享. 经仿真验证, 高效联合树算法能够在保证准确率的同时, 以更短的时间作出诊断推理. 基于现场收集的数据, 建立水泥回转窑故障诊断系统模型并应用改进的算法实现了精准且快的故障诊断.

     

Meta-path reasoning of knowledge graph for commonsense question answering

Commonsense question answering (CQA) requires understanding and reasoning over QA context and related commonsense knowledge, such as a structured Knowledge Graph (KG). Existing studies combine language models and graph neural networks to model inference. However, traditional knowledge graph are mostly concept-based, ignoring direct path evidence necessary for accurate reasoning. In this paper, we propose MRGNN (Meta-path Reasoning Graph Neural Network), a novel model that comprehensively captures sequential semantic information from concepts and paths. In MRGNN, meta-paths are introduced as direct inference evidence and an original graph neural network is adopted to aggregate features from both concepts and paths simultaneously. We conduct sufficient experiments on the CommonsenceQA and OpenBookQA datasets, showing the effectiveness of MRGNN. Also, we conduct further ablation experiments and explain the reasoning behavior through the case study.     

Formal Nonmonotonic Theories and Properties of Human Defeasible Reasoning

The knowledge representation and reasoning of both humans and artificial systems often involves conditionals. A conditional connects a consequence which holds given a precondition. It can be easily recognized in natural languages with certain key words, like “if” in English. A vast amount of literature in both fields, both artificial intelligence and psychology, deals with the questions of how such conditionals can be best represented and how these conditionals can model human reasoning. On the other hand, findings in the psychology of reasoning, such as those in the Suppression Task, have led to a paradigm shift from the monotonicity assumptions in human inferences towards nonmonotonic reasoning. Nonmonotonic reasoning is sensitive for information change, that is, inferences are drawn cautiously such that retraction of previous information is not required with the addition of new information. While many formalisms of nonmonotonic reasoning have been proposed in the field of Artificial Intelligence, their capability to model properties of human reasoning has not yet been extensively investigated. In this paper, we analyzed systematically from both a formal and an empirical perspective the power of formal nonmonotonic systems to model (i) possible explicit defeaters, as in the Suppression Task, and (ii) more implicit conditional rules that trigger nonmonotonic reasoning by the keywords in such rules. The results indicated that the classical evaluation for the correctness of inferences has to be extended in the three major aspects (i) regarding the inference system, (ii) the knowledge base, and (iii) possible assumed exceptions for the rule.     

信息检索的概率模型

The study of mathematical models on information retrieval is an important area in the Information Retrieval community. Because of the uncertainty characteristic of IR,the probability model based on statistical probability is apromising model from recent to future. Those models can be classified into classical models and probability network models. Several famous models are introduced and their shortcomings are pointed out in this paper. We also clarifythe relationship of these models and introduce a new models based on statistical language model curtly.     

Information networks fusion based on multi-task coordination

Information networks provide a powerful representation of entities and the relationships between them. Information networks fusion is a technique for information fusion that jointly reasons about entities, links and relations in the presence of various sources. However, existing methods for information networks fusion tend to rely on a single task which might not get enough evidence for reasoning. In order to solve this issue, in this paper, we present a novel model called MC-INFM (information networks fusion model based on multi-task coordination). Different from traditional models, MC-INFM casts the fusion problem as a probabilistic inference problem, and collectively performs multiple tasks (including entity resolution, link prediction and relation matching) to infer the final result of fusion. First, we define the intra-features and the inter-features respectively and model them as factor graphs, which can provide abundant evidence to infer. Then, we use conditional random field (CRF) to learn the weight of each feature and infer the results of these tasks simultaneously by performing the maximum probabilistic inference. Experiments demonstrate the effectiveness of our proposed model.