The first Promotion Committee meeting on the "Project for the Fundamental Information Technology of the Next Generation (RWC-RWI/PDC)" (Chairman: Professor Hidehiko Tanaka, University of Tokyo) was held on May 15, 1997. The committee, an advisory body to the director-general of the Machinery and Information Industries Bureau of the Ministry of International Trade and Industry, examined the master plan for the RWC-RWI/PDC project. The examination included the contents, promotion system and research and development schedule for the Real World Intelligence and the Parallel and Distributed Computing Technology Fields.
In response to the results of the examination, the Real World Computing Partnership decided to undertake concrete research and development activities, from FY 1997, based on the master plan.
The following is the full text of the master plan, introducing the policy of the RWC project for the second 5-year period. (Editorial Staff)
Part I - Basic Concepts
1. Background
The Real World Computing (RWC) Program started as a 10-year project in FY1992 to establish an innovative system of information processing technologies capable of handling raw information in the real world. The five-year period from FY1992 through FY1996 was allocated for explorative research, and the studies on such fields as novel functions, parallel systems, and optical technologies were conducted.
What should be done in the second half of the RWC Program was examined by the Evaluation and Promotion Committee held in March 1996 and Preparation Committees. As a result, it was decided that the research resources should be concentrated in two fields, Real World Intelligence Technology Field and Parallel and Distributed Computing Technology Field. It was also decided that the research activities should be more radical, and thus it was confirmed that in the second period starting in FY1997, important key technologies should be established in each field for developing the fundamental information technologies of the next generation while inheriting the framework of the RWC Program.
The master plan of the RWC Program dictates the narrowing of the research and development challenges through the transition from the first half to the second half. This master plan, accordingly, modifies the master plan of the RWC Program.
2. Basic Policy
The basic policy is to conduct research and development activities with suitable methods and schemes for the Real World Intelligence Technology Field and the Parallel and Distributed Computing Technology Field respectively, while inheriting the basic policy of the RWC master plan, such as supporting flexible implementation schemes and introducing competitive principles. Also included in the basic policy is an intent to coordinate and link both fields, such as the allocation of research resources, to improve the effectiveness and efficiency in establishing important fundamental information technologies of the next generation.
The Real World Intelligence Technology Field requires a new technological foundation for pioneering information technologies and for deploying them to industries. Such technological foundations cannot be accomplished unless all industrial, academic, and governmental research bodies play their respective roles in a well-coordinated fashion. To enable this coordination, the Electrotechnical Laboratory shall play a leading role under a specific framework for planning and implementing detailed research and development activities (including the allocation of research resources), and for promoting research and development activities regarding the Real World Intelligence Technology Field.
Part II - Research and Development Plan
I. Real World Intelligence Technology Field
1. Background
Current information processing technologies basically rely on the logical and procedural type information processing which is suitable for applications that perform efficient processing according to pre-defined algorithms. However, the majority of the information in the real world is pattern information which is spatially and temporally distributed and characterized by a large degree of diversity and ambiguity. To deal with pattern information, such as human recognition and dialogue, it is impossible to take all the probable situations into consideration to prepare algorithms in advance - even the algorithms themselves are unclear in many cases.
It is considered that the human brain acquires and accumulates the ability to adapt to unknown situations and changes in the environment with learning/self-organization functions. The brain independently modifies the connection between the neural network with a given experience, while processing the pattern information in a parallel and integrated manner. Therefore, it is essential to add to the current information processing technologies these abilities of information integration and learning-type information processing in order to enlarge the scope of information processing technology applications.
2. Objectives
To develop the fundamental technologies which add the ability of information integration and learning type information processing (Real World Intelligence) to conventional information processing technologies, and to enlarge the scope of application for information processing.
3. Research and Development Contents
(1) Target
To develop the following key technologies required for information integration and learning type information processing systems which can accept raw information as is in the real world, recognize or predict a specific environment or situation, and autonomously respond to it.
1) Information Integration Technologies
Technologies for processing various kinds of information in the real world, such as images and voices, which contain ambiguity and uncertainty in an integrated manner, and for recognition/understanding, overall judgment, and decision-making regarding behavior.
2) Learning/Self-Organization Technologies
Technologies that enable systems to adapt and evolve their own functions by collecting information autonomously through interactions with the real world.
(2) Overview
More specifically, the research and development is addressed to the following items:
To establish information integration and learning/self-organization technologies by constructing three types of systems for confirming novel functions including elementary functions for recognition/understanding, interaction, problem solving, and control with concurrent research methods, linked with the research on the theoretical and algorithmic foundation;
To develop devices capable of real-time and adaptive processing useful for realizing these technologies; and
To develop the real-world information databases, benchmark challenges, and software libraries which are necessary for research and development.
1) Systems for Confirming Novel Functions
Multimodal Functions An agent-driven human interface for using information systems with voices and images combined. Autonomous Learning Functions An agent system which can autonomously move around in the real environment, collect and learn information in the environment and its surroundings through sensing and dialogue. Self-Organizing Information Base Functions An agent system which can compile, summarize, retrieve, and present the various information in the real world or on an information network in a self-organizing manner.
2) Theoretical and Algorithmic Foundation
Basic theories and common techniques for information integration, learning/self-organization, and optimization.
3) Real World Adaptive Devices
Development of the Next-Generation Field Programmable Gate Arrays (adaptive devices) which can perform real-time processing and be adaptable to various applications.
4. Remarks
Regarding systems that confirm novel functions employing the concurrent research method, the concentration of research resources into the most effective research and development for establishing information integration technologies and learning/self-organization technologies will be examined through an evaluation process. Also, specific methods for promoting technology transfer to various industry fields will be investigated.
II. Parallel and Distributed Computing Technology Field
1. Background
It is expected that in the beginning of the 21st century, speed improvements of processor chips will reach the physical limit of device technology, and improvements of computer system performance will rely on parallel and distributed technologies at various levels.
The expected technical approach will be to develop technologies to achieve high-speed processing on a multiprocessor system which is integrated on a chip, and to develop parallel and distributed processing technologies which provide the optimal parallel processing environment by dynamically configuring heterogeneous computing resources which reside on distributed systems, in response to diversified computing demands. Another challenge is to develop a demonstrative parallel application which runs efficiently in a parallel environment.
2. Objectives
To develop fundamental technologies required for realizing the next-generation parallel and distributed environment (seamless parallel and distributed computing environment) to provide optimal parallel computing in response to the ever-changing variety of computing demands, and to improve the ability of developing the next-to-next-generation computing architecture.
3. Contents of Research and Development
(1) Target
To establish parallel software technologies for effective application of hardware performance of a multiprocessor system, and to establish fundamental technologies required for the parallel computing environment where computing resources in various architectures can be used without needing to acknowledge the differences in architectures. In addition, our target is to develop demonstrative state-of-the-art parallel applications.
(2) Overview
1) Multiprocessor Computing Technologies
Research and development of the elementary parallelization techniques, typified by the parallelism exploitation technique and scheduling as the key software technologies in our approach to develop next-generation high-performance multiprocessor systems and a single-chip multiprocessor (SCM). Software approaches will focus on the following two areas:
Exploiting and Using Parallelism Techniques
This technique consists of the fundamental program analysis technique and the performance evaluation technique for using parallelism across multiple granularities, including instruction level, loop level, and procedure level intrinsic to programs.
Scheduling Techniques
This scheduling allows effective mapping of resulting parallelism onto hardware, and enables the overlapping of processing and data transfers.
Also, studies will be conducted on basic architectures such as processor architecture and memory architecture which assist in the effective adaptation of the developed elementary parallelization techniques.
2) Seamless Parallel and Distributed Computing Technologies
The research and development of technologies required for systems which can provide optimal parallel computing ability in response to the ever-changing variety of computing demands of distributed systems in a typical platform will primarily focus on the following two areas.
System Architecture Technologies
These technologies consist of communication/ memory architecture, programming, libraries, and environment application technologies for improving parallel execution performance and realizing performance scalability in a heterogeneous distributed environment.
Optical Interconnection Technologies
These technologies consist of optical transmission modules, optical/electric interfaces, and the constituent devices which enable a super-high-speed parallel transmission of data involved in computing.
3) Parallel Application Technologies
The target for these technologies is to examine a framework for research and development as a basis for developing state-of-the-art parallel applications, and to conduct research and development with demonstrative approaches focused on the following research fields most expected to affect industries not directly related to information technology.
Computational Chemistry and Computational Biology Fields
This category consists of molecular dynamics, protein structure prediction, gene finding, genome sequence analysis, etc.
Large-Scale Systems Field
This category refers to system analysis and large-scale information retrieval for large-scale systems (such as social infrastructures).