Study on Parallel Processing of Multimedia in AI Systems

为了在人工智能（AI）系统的框架内给出多媒体信息的并行处理技术，我们研究了该技术的构成要素。我们还增强了这些元素，解释如下。智能绘图算法和感性表达方法的研究我们开发了人工智能系统理解、解释和表达多媒体信息的算法。然后，我们将这些算法与识别方法和归纳学习方法相结合，分别进行了研究。因此，组合算法实现了高层次的处理。我们还考察了它在艺术领域的“感性”表现。音乐信息处理的研究我们开发了一个基于分形技术的音乐编曲系统，并将其应用于图像编码。然后，我们对该系统进行了增强，以获得音乐的整体配置。通过对音乐结构的考虑，增强后的系统不仅可以生成和表达主要的、更多的旋律，还可以生成和表达音乐的对位部分和其他部分。基于战略解决方法的问题解决方法研究我们开发了一种对一般问题解决框架进行归纳、概括和使用战略知识的方法。在一个应用领域检验了其效果。我们还研究了进化方法中的另一种方法。将这两种方法结合起来还有待于今后的工作。利用并行归纳学习算法获取知识我们研究了归纳学习方法在多媒体信息分类和解释中的应用。我们开发并比较了三种不同的方法来并行化学习方法。这项研究得到了公认的结果。我们需要进一步开发增强功能，以用于更大规模的应用。基于并行技术的高速溯因推理方法研究在溯因推理的框架下，我们开发了基于并行处理的加速技术。这是在由许多计算机通过网络连接的实际环境中考虑的。在实际的网络系统中，一台计算机必须由许多用户使用，并执行许多任务。我们开发了一种在这种动态环境下工作的并行处理方法。少

In order to give parallel processing technology for multimedia information within the framework of artificial intelligence (AI) systems, we have investigated elements for the technology. We also enhanced these elements which are explained as follows.1. Research on an intellectual drawing algorithm and a 'kansei' expression methodWe developed algorithms for AI systems to understand, to interpret and to express multimedia information. Then, we combined these algorithms with a recognition method and an inductive learning method, which are investigated separately. Thus, the combined algorithm achieves a high-level processing. We also inspected it in an artistic domain for 'kansei' expression.2. Research on music information processingWe developed a music composition and arrangement system based on a fractal technology, which has been used for image coding. Then, we enhanced this system in order to give total configuration of music. The enhanced system can generate and express not only main … More melodies but also counter parts of music and other parts of music, by considering structure of music.3. Research on problem solving methods using a strategic solution methodWe developed a method to induce, to generalize and to use strategic knowledge for the general problem solving framework. Its effect is inspected in an application area. We also investigated another approach within evolutionary methods. Combining these two approaches are remain for future works.4. Knowledge acquisition using parallel inductive learning algorithmsWe studied application of inductive learning methods to classify and to interpret multimedia information. We developed and compared three different approaches to parallelize the learning methods. This research archived a recognized result. We need to develop further enhancement to be used for larger scale applications.5. Research on a high-speed abductive reasoning method using a parallel techniquesIn the framework of abductive reasoning, we developed speed-up techniques using parallel processing. This is considered in a practical environment which consists of many computers connected by a network. In the real network system a computer must be used by many users and works for many tasks. We developed a parallel processing method that works in such a dynamic environment. Less