基于知识蒸馏的YOLOv3算法研究

知识蒸馏作为一种模型压缩方法,将大网络（教师网络）学到的知识传递给小网络（学生网络）,使小网络获得接近大网络的精度。知识蒸馏在图像分类任务上获得不错的效果,但在目标检测上的研究较少,且有待提高。当前目标检测中主要基于特征提取层进行知识蒸馏,该类方法存在两个问题,第一,没有对教师网络传递知识的重要程度进行度量,第二,仅对特征提取层进行蒸馏,教师网络的知识未充分传递给学生网络。针对第一个问题,通过引入信息图作为蒸馏的监督信号,强化了学生网络对教师网络重点知识的学习;针对第二个问题,对特征提取层和特征融合层的输出同时进行蒸馏,使学生网络更充分地学习教师网络传递的知识。实验结果表明,以YOLOv3为检测模型,在不改变学生网络结构的基础上,平均类别精度（mAP）提升9.3个百分点。     

概率逻辑模型与学习研究进展

近年来，概率逻辑学习研究取得了很大进展，已经提出各种不同的形式化方法和学习方法，包括概率关系模（PRMs）、贝叶斯逻辑程序（BLPs）、逻辑贝叶斯网络（LBNs）和随机逻辑程序（SLPs）等。文章重点介绍了贝叶斯网络与一阶逻辑的结合，并以PRMs、BLPs和LBNs为例，描述了基于贝叶斯网络的概率逻辑模型（PLMs）的知识表示方法，给出了此类PLMs一般使用的参数估计方法和结构学习方法，并给出了建议的研究方向。     

基于生成树的双环网络G(N;1,s)直径求解算法

对紧优双环网络G（N；1，s）的直径求解算法做了研究，提出基于生成树的紧优双环网络G（N；1，s）求解算法，给出了双环网络的直径d（N；1，s）公式．对生成树的性质做了研究。利用C#作为编程语言来实现这一算法，并对生成树的结构模型进行了仿真实现。验证了双环网络直径的分布特点：具有最大值、最小值和中间对称性。对任意给定N而2≤s≤N-1的这样一系列双环网络中的所有的紧优双环网络都可以计算出来。该算法的时间复杂度为O（N）。     

智能教学系统中的知识树增长模型

智能教学系统（intelligent tutoring system,ITS),作为人工智能学科的重要研究应用领域,是21世纪人类社会数字化教育的必然发展方向,迄今研究逾30年。现有一些ITS系统由于知识表示以及推理方法的领域相关特性,系统构建与系统运用、系统模块之间动态有机联系不强,进而导致学生模型弱化等问题,限制了系统在进行个别化教学过程中的智能性和推广运用。在此从系统的知识表示入手,基于SC文法的知识表示体系和知识树映射方法,提出了一个动态、实时、自适应、交互式知识树增长模型（augment knowledge-tree model,AKTM)。模型包括基于SC文法的知识点表示方法、知识树结构以及知识树映射、知识树学生模型、知识点学习循环等内容,贯穿于ITS系统4大传统模块之中,并在内容和功能上与之完全集成。通过知识点层次多维属性标注和索引,实现了知识存储、处理、调用和维护动态、一体化过程。通过知识点学习循环,实现个别化、动态、自适应智能教／学过程。同时通过模型在多媒体ITS系统中的实例化设计和运用,实现了动态教／学、领域无关、人机交互、自适应、个别化等智能特点。     

基于知识蒸馏的超分辨率卷积神经网络压缩方法

针对目前用于超分辨率图像重建的深度学习网络模型结构深且计算复杂度高，以及存储网络模型所需空间大，进而导致其无法在资源受限的设备上有效运行的问题，提出一种基于知识蒸馏的超分辨率卷积神经网络的压缩方法。该方法使用一个参数多、重建效果好的教师网络和一个参数少、重建效果较差的学生网络。首先训练好教师网络，然后使用知识蒸馏的方法将知识从教师网络转移到学生网络，最后在不改变学生网络的网络结构及参数量的前提下提升学生网络的重建效果。实验使用峰值信噪比（PSNR）评估重建质量的结果，使用知识蒸馏方法的学生网络与不使用知识蒸馏方法的学生网络相比，在放大倍数为3时，在4个公开测试集上的PSNR提升量分别为0.53 dB、0.37 dB、0.24 dB和0.45 dB。在不改变学生网络结构的前提下，所提方法显著地改善了学生网络的超分辨率重建效果。     

采用贝叶斯网络建立一种智能学习环境学生代理模型的研究

贝叶斯（Bayesian）网络近年成为智能代理引人注目的研究方向，本文在介绍Bayesian网络及其构建方法、学习方法的基础上，构建了一个通用的可扩展的智能学习环境中的学生模型。     

基于多教师知识蒸馏网络的钢铁表面缺陷诊断

随着人工智能技术的发展，基于深度学习的产品表面缺陷检测逐渐成为工业品质检验的重要手段。然而，传统的单一模型在复杂数据分布下的效果不尽如人意，因此采用集成学习的方法提高模型的诊断精度。文章提出了一种基于多教师知识蒸馏网络的工业产品表面缺陷诊断方法。该方法将3个结构差异较大的模型（ResNet50、Inception-v4和EfficientNet）作为教师网络，再将知识蒸馏到一个轻量化的学生网络（MobileNetv3）中，从而诊断钢铁表面缺陷。通过实验验证可知，多教师知识蒸馏网络比单一模型具有更高的准确率和鲁棒性。实验结果表明，该方法不仅可以提高钢铁表面缺陷诊断的精度，而且可以有效应用于其他复杂数据分类问题。     

一种基于贝叶斯网络的作战重心评估模型

作战重心（Center of Gravity）是指战役体系中敌我双方的关键环节。作战重心评估是一个经验性、模糊性的过程。贝叶斯网络作为一种不确定知识表示模型,具有概率论及图论基础,对于解决复杂系统决策问题具有较强的优势,适合用于作战重心评估。文中提出并实现了一种基于贝叶斯网络推理的作战重心评估模型。通过该模型,可以定量地评估各个环节对于证据的重要程度,从而确定该作战过程中的作战重心。文中使用联合树（Clique Tree）算法进行贝叶斯网络精确推理,并详细阐述了推理过程中联合树建立,消息传递的过程。最后通过实例验证,基于贝叶斯网络推理的模型能够有效地对作战重心进行定量的评估。     

B+树阶数m的最优选取

本文指出对普通的B^ 树作了一定的限制后，当选择B^ 树适当的阶数m时，对B^ 树索引文件的时间和空间利用率有一个最优的设计过程，分别给出了应用在两种数据库系统模型中的B^ 树索引文件结构：（1）具有有限内存的数据库系统（Database Systems with Limited Amount of Main Memory,简称为DBSLAM），在DBSLAM模型中，我们只是缓存B^ 树中的部分结点；（2）具有极大内存的数据库系统（Database Systems with Very Large Amount of Main Memory，简称为DBSVLAM）,在DBSVLAM模型中，我们将缓存B^ 树中所有的结点，基于以上两种模型，我们分别建立了两个不同的时空开销模型，并以此进行了优化设计。     

A3C深度强化学习模型压缩及知识抽取

异步优势演员评论家(asynchronous advantage actor-critic,A3C)构建一主多从异步并行深度强化学习框架，其在最优策略探索中存在求解高方差问题，使主智能体难以保证全局最优参数更新及最佳策略学习.同时，利用百万计算资源构建的大规模并行网络，难以部署低功耗近端平台.针对上述问题，提出紧凑异步优势演员评论家(Compact＿A3C)模型，实现模型压缩及知识抽取.该模型冻结并评价A3C框架中所有子智能体学习效果，将评价结果转化为主智能体更新概率，保证全局最优策略获取，提升大规模网络资源利用率.进一步，模型将优化主智能体作为“教师网络”，监督小规模“学生网络”前期探索与策略引导，并构建线性衰减损失函数鼓励“学生网络”对复杂环境自由探索，强化自主学习能力，实现大规模A3C模型知识抽取及网络压缩.建立不同压缩比“学生网络”，在流行Gym Classic Control与Atari 2600环境中达到了与大规模“教师网络”一致的学习效果.模型代码公布在https://github.com/meadewaking/Compact＿A3C.     

基于Web的智能教学系统的研究与实现

基于Web的智能教学系统综合传统的ITS系统优势,同时又结合Web的特点,系统具有智能性,能够智能地引导学生学习。它是以认知科学为理论基础,综合利用人工智能技术、教育心理学、计算机科学等多门学科的成果而形成的一种对学生实施有效教学的技术。提出了一种基于Web的智能教学系统模型,给出了智能教学系统的知识库的设计、学生模型的构建方法及教学策略的设计,并探讨了学生模型的学习评价,最后对实现系统的关键技术进行了研究。     

智能辅助教学写作环境CourseTalk的设计与实现

本文介绍了一个智能辅助教学写作环境CourseTalk的设计与实现.在这个环境中使用了综合媒体结构(Hypermedia)、Prolog和黑板结构等技术,以及推理与查询相结合的搜索策略,并提出一种通过编辑直接定义关系数据以及可以用Prolog语言方便地描述教学策略的方法.同时还讨论了有关知识获取、教学策略选择和学生模型等问题.     

Applying a student modeling with non-monotonic diagnosis to Intelligent Virtual Environment for Training/Instruction

We present a student modeling approach that has been designed to be part of an Intelligent Virtual Environment for Training and/or Instruction (IVET). In order to provide the proper tutoring to a student, an IVET needs to keep and update dynamically a student model taking into account the student’s behaviour in the Virtual Environment. For that purpose, the proposed student model employs a student ontology, a pedagogic diagnosis module and a Conflict Solver module. The goal of the pedagogic diagnosis module is to infer which learning objectives have been acquired or not by the student. Nevertheless, the diagnosis process can be complicated by the fact that while learning the student will not only acquire new knowledge, but he/she may also forget some previously acquired knowledge, or he/she may have some oversights that could mislead the tutor about the true state of the student’s knowledge. All of these situations will lead to contradictions in the student model that must be solved so that the diagnosis can continue. Thus, our approach consists in applying diagnosis rules until a contradiction arises. At that moment, a conflict solver module is responsible of classifying and solving the contradiction. Next, the student ontology is updated according to the resolution adopted by the Conflict Solver and the diagnosis can continue. This paper mainly focuses on the design of the proper mechanisms of the student model to deal with the non monotonic nature of the pedagogic diagnosis.     

Bayesian networks for student model engineering

Bayesian networks are graphical modeling tools that have been proven very powerful in a variety of application contexts. The purpose of this paper is to provide education practitioners with the background and examples needed to understand Bayesian networks and use them to design and implement student models. The student model is the key component of any adaptive tutoring system, as it stores all the information about the student (for example, knowledge, interest, learning styles, etc.) so the tutoring system can use this information to provide personalized instruction. Basic and advanced concepts and techniques are introduced and applied in the context of typical student modeling problems. A repertoire of models of varying complexity is discussed. To illustrate the proposed methodology a Bayesian Student Model for the Simplex algorithm is developed.     

基于云模型的智能教学系统中学习质量的评价

提出将云模型理论应用于智能教学系统的学生模型构建中,建立一种基于云模型的学生学习质量评价方法,利用云对概念的贡献程度进行数据离散化,并引入云变换计算隶属云,最后结合极大判定算法求出更加符合实际的学习质量等级划分。实验结果表明,这种新的评价方法得出的隶属概念不仅能够反映出学生对知识点的掌握程度,还能够反映出学生在学习中的发挥稳定性、心理素质等情况,有利于提高智能教学系统的应用效率。     

智能辅导系统

六十年代初期,计算机技术开始应用于教学领域,这时的计算机辅助教学(CAI)系统只能于简单的课程安排和进阶测验。七十年代,随着人工智能(AI)的不断发展和成熟,人们开始考虑智能性引     

Student diagnosis in practice; bridging a gap

This paper presents a novel framework for looking at the problem of diagnosing a student's knowledge in an Intelligent Tutoring System. It is indicated that the input and the conceptualisation of the student model are significant for the choice of modeling technique. The framework regards student diagnosis as the process of bridging the gap between the student's input to the tutoring system, and the system's conception and representation of correct knowledge. The process of bridging the gap can be subdivided into three phases, data acquisition, transformation and evaluation, which are studied further. A number of published student modeling techniques are studied with respect to how they bridge the gap.     

A Framework for the Initialization of Student Models in Web-based Intelligent Tutoring Systems

Initializing a student model for individualized tutoring in educational applications is a difficult task, since very little is known about a new student. On the other hand, fast and efficient initialization of the student model is necessary. Otherwise the tutoring system may lose its credibility in the first interactions with the student. In this paper we describe a framework for the initialization of student models in Web-based educational applications. The framework is called ISM. The basic idea of ISM is to set initial values for all aspects of student models using an innovative combination of stereotypes and the distance weighted k-nearest neighbor algorithm. In particular, a student is first assigned to a stereotype category concerning her/his knowledge level of the domain being taught. Then, the model of the new student is initialized by applying the distance weighted k-nearest neighbor algorithm among the students that belong to the same stereotype category with the new student. ISM has been applied in a language learning system, which has been used as a test-bed. The quality of the student models created using ISM has been evaluated in an experiment involving classroom students and their teachers. The results from this experiment showed that the initialization of student models was improved using the ISM framework.     

Student modeling and assessment in intelligent tutoring of software patterns

This paper presents the design, implementation, and evaluation of a student model in DEPTHS (Design Pattern Teaching Help System), an intelligent tutoring system for learning software design patterns. There are many approaches and technologies for student modeling, but choosing the right one depends on intended functionalities of an intelligent system that the student model is going to be used in. Those functionalities often determine the kinds of information that the student model should contain. The student model used in DEPTHS is a result of combining two widely known modeling approaches, namely, stereotype and overlay modeling. The model is domain independent and can be easily applied in other learning domains as well. To keep student model update during the learning process, DEPTHS makes use of a knowledge assessment method based on fuzzy rules (i.e., a combination of production rules and fuzzy logics). The evaluation of DEPTHS performed with the aim of assessing the system’s overall effectiveness and the accuracy of its student model, indicated several advantages of the DEPTHS system over the traditional approach to learning design patterns, and encouraged us to move on further with this research.