Contents - Previous - Next
This is the old United Nations University website. Visit the new site at http://unu.edu
2.1 Social Background
Supported by the development of computer and communication technologies, information technology is producing innovative change in society. This innovative change is not only affecting industrial activities such as new wealth and services and rationalized production and distribution systems but is also enhancing national living standards, regional progress, and educational and cultural growth.
The information network is growing from an intra-organizational network to an inter-enterprise network and, further, to one bringing together homes and individuals, and, likewise, information processing is growing beyond enterprises to include homes and individuals.
As a result, information to be processed will greatly increase not only in quantity but also in quality and variety, and a new technological basis that enables everyone to easily and efficiently take advantage of the various information resources of the network will be required for such an information network society.
Given the social background and the technological requirements for various application fields of information processing, computers should become more humanized and possess the capability to assist and collaborate with humans in the real-world environment.
2.2 Technological Background
Computers have evolved dramatically with the support of technological progress.
In the first stage, as is shown in figure 1, computers developed along the line of conventional von Neumann architectures in the fields of numerical computation, document processing and management of database storage and retrieval, since those application fields have clear algorithms and are suitable for processing by conventional computers.
The second stage of development was directed toward the manipulation of symbols and logic, casting light on the intelligent thought processes of humans such as logical (deductive) inference. The research fields of AI and knowledge engineering contributed by providing computers with the capability of handling symbolic representation of knowledge and inference rules. In 1981, Japan proposed the concept of Fifth Generation Computers as an approach to new generation computing based on logical programming. The goal was to provide computers with powerful logical inference capability and to open the door to the world of large-scale knowledge-information processing.
Today, computers have come to provide enormous computing power and far surpass human ability in solving well-defined problems, where the algorithms for solution exist and can be clearly stated in programming languages. Nevertheless, computers are still not very flexible when compared to human flexibility in information processing in the real world, where many problems are ill-defined and hard to describe in algorithmic terms. It might be called the "algorithm crisis," compared to the so-called software crisis.
2.3 Real-world Computing Paradigm
In order to cope with real-world problems and to open a new horizon in information-processing technology, it is essential to pursue research on the fundamental ways of human-like flexible information processing, by casting light on the intuitive or sub-symbolic level of human information processing, and to embody the results as new information-processing technologies for use with the developing hardware technologies.
In recent years, fundamental knowledge about the information-processing capability of humans has been rapidly accumulating, and hardware technologies for massively parallel computing systems are expected to reach our hands by the beginning of the twenty-first century.
Figure 1 Extension of the functions of computing systems
Given these circumstances, the concept of "Real-World Computing" is proposed as a new paradigm of information processing that aims at furnishing the "real-worldness" (or "flexibility") of human information processing to information systems.
In order to make the twenty-first century a globally prosperous information network society where humans and information systems collaborate more closely and flexibly, the Real-World Computing programme, which aims at developing novel fundamental generic technologies in the fields of information processing, is a topic of vital concern.
3. The concept of real-world computing
In the changing environment of our daily life in the real world, we humans evaluate various kinds of information, including ambiguity or uncertainty, and acquire the necessary information for making predictions, plans, or decisions in a flexible way. Such an information-processing function is characterized by the term "flexible information processing" or "real-world computing," in contrast to the conventional rigid information processing performed by computers that assumes complete information in a pre-assumed world or problem domain.
Information processing in general is a function that has been acquired by humans and creatures in the course of evolution in surviving and adapting to the changing real-world environment. Although it is a versatile function, it can be divided into the following two categories: (1) logical information processing and (2) intuitive information processing.
Logical information processing is characterized by the words conscious, analytical, deductive, serial, intensive, digital, and symbolic. Contrarily, intuitive information processing is characterized by the words unconscious, synthetic, inductive, parallel, distributed, analogue, and sub-symbolic.
Basically, information-processing technology is intended to complement or substitute for our information-processing function through mechanization of both of the two aspects classified above. Historically, however, machines that were theoretically and technologically suitable for logical information processing developed into conventional digital computers. They progressed very rapidly, supported by the tremendous growth of hardware technologies, and serial and sequential information processing have been established as today's ruling paradigm (see figure 2).
On the other hand, intuitive information processing has been studied in the research fields of pattern recognition and learning. The original model was Perceptron, and recently neural network computing has been the case. However, intuitive information processing is still immature in current information technology. Conventional computers are far behind humans in this aspect, and their lack of flexibility is due to the unbalanced development of those two aspects.
Figure 2 Separation and unification of the two aspects of information processing
As has been seen in the previous section, development of information systems that have human-like flexible information-processing functions and can cope with real-world problems is now one of the most important demands common to various fields, such as pattern-information processing, knowledge-information processing, intelligent robots, and friendly man-machine interface, which aim at further advanced information processing.
To develop such systems, it is, first of all, important to explore the intuitive aspect of information-processing functions of humans and embody it as a new technology. By coordinating the two aspects of intuitive and logical information processing and integrating them, Real-world Computing (hereafter denoted by RWC), which has the following novel functions, will be established as a new paradigm of information technology:
- the function to integrate a variety of complex and intricately related in formation containing ambiguity or uncertainty and to reach an appropriate (approximate) decision or solution within a reasonable time,
- the function to actively acquire necessary information and knowledge and to learn general knowledge inductively from examples,
- the function to adapt the system itself to users and a changing environment. Applications of RWC systems are expected to include a wide range of real-world problems, such as ill-defined (incomplete) problems like the understanding of situations in a noisy environment, large-scale problems such as the simulation of social and economic phenomena, real-time problems such as man-machine interface with virtual reality, and autonomous control of intelligent robots.
The key technical requirements for the flexibility of RWC systems are: openness, robustness, and real-time, where by "openness" is meant that the system can adaptively and autonomously change or extend itself to cope with the unexpected situations it encounters in the real world; by "robustness" is meant tough and stable behaviour of the system for distorted or fluctuating information input; and "real-time" means that the system can respond within a reasonably short time.
The reason why we humans can maintain the flexible information-processing ability characterized by the above real-worldness is that our brains incorporate distributed representation of information, massively parallel processing, learning, and self-organizing ability, and information-integration ability. Therefore, the following will be key concepts for realizing the above characteristics of RWC systems:
(1) Flexible information processing: the functional aspect of RWC, which is characterized by the admissibility and integration of ambiguous and uncertain information and the capability of adaptation and learning.
(2) Massively parallel and distributed processing: the computational aspect of RWC, characterized by processing of multi-modal, multi-variate, and strongly correlated information in a massively parallel and distributed manner.
4. Outline of RWC programme
4.1 Purpose of Research and Development
Information systems of the twenty-first century will be based on not a single but various key technologies, such as those for massively parallel computing, optical computing, neural computing, and logic-based computing. These technologies need to be flexibly integrated into information systems in order to cope with real-world problems.
The main purpose of the RWC programme is to lay the technological foundation for the advanced information society of the twenty-first century. This programme is aimed at establishing the basis for flexible and advanced information technologies that are closely allied to humans and are capable of processing a variety of information in the real world. Such technologies seem essential for creating a cooperative relationship between humans and computers and for producing innovative and generic technologies toward the advanced information society of the twenty-first century.
The primary goal of this programme is not to develop a single computer but to contribute toward the realization of these significant but not yet established technologies. Herein, we will try to establish the theoretical foundation for these technologies, to explore their potentials, and to create some specific style of their integration. Some important real-world problems will be tackled for confirming the possibility and usefulness of the new technologies. By disclosing these experiences to the public, it is also intended that Japan will contribute to the development of the common knowledge and wealth of humankind.
In order to accomplish these fundamental and ambitious goals, it is very important and imperative to promote international and interdisciplinary cooperation in the research fields and to support collaboration among industries, national institutes, and universities.
4.2 Subjects of Research and Development
The research and developments in the RWC programme are divided into the following mutually related subjects:
- theoretical foundation,
- novel functions for application,
- computational bases.
A three-storied structure as shown in figure 3 is the fundamental framework for the organization of research and development in this programme.
4.2.1 Theoretical Foundation
The research objective is to establish a new theoretical foundation for flexible information processing. For this purpose, it is necessary to expand and generalize the conventional framework of information processing and to clarify the principle, or "soft logic," commonly underlying flexible information processing.
The research topics are:
- Flexible representation of information
- Evaluation of information and processing models
- Flexible storage and recall of information
- Integration of information and of processing modules
- Learning and self-organization
- Optimization methods
In particular, integration of multi-modal information (and of heterarchical processing modules) and learning and self-organization (optimization and adaptation) are the most fundamental issues, and how to implement these in a framework of massively parallel and distributed information processing will be a key point.
Figure 3 Organization of research and development
4.2.2 Novel Functions for Application
Research and development on novel functions for application should be directed toward investigating elemental novel functions, which are on the whole important for realizing flexible information systems to solve a wide range of real-world problems.
Research topics are classified into the following categories.
- Flexible recognition and understanding of multi-modal information
- Flexible inference and problem solving based on flexible information base
- Flexible interactive environment for man-machine communication
- Flexible and autonomous control
Some typical application problems will be tackled for exploring the integration of these elemental technologies and for demonstrating their effectiveness.
These novel functions will be implemented on the following computational bases.
4.2.3 Computational Bases
RWC involves processing large volumes of spatio-temporally distributed information at a high speed, while taking into account their mutual interactions. As a new computational basis to support it, computing systems that can exploit parallel and distributed processing at several processing levels should be developed. For this purpose, research and development from the following perspectives is important; how to integrate these technologies will also be investigated.
- General-purpose massively parallel computing systems
- Neural systems as a kind of special-purpose system
- Optical computing systems
In the following section, more detailed contents for each research subject will be described.
5. Theoretical foundation
The objective of research is to lay the theoretical foundation for the technological realization of human-like flexible information processing as a new paradigm of information processing.
So far, much theoretical research has been done in research fields related to flexible and parallel distributed information processing. Such fields are, for example, pattern recognition, multi-variate data analysis, probabilistic and statistical inference, fuzzy logic, neuro-computing, machine learning, regularization, and various optimization methods, and so on.
In order to provide a theoretical foundation for flexible information pro ceasing, it is important not only to continue in-depth study in these research areas but also to clarify the theoretical framework of "soft logic" commonly underlying these fields and to aim at constructing a new unified theoretical base for dynamic RWC. Here, probabilistic and statistical formulation of problems and non-linear dynamics in conjunction with learning and self-organization will be a key approach.
It will be necessary to expand and generalize the conventional framework of information processing in all aspects of information representation, processing, and evaluation, and to systematize basic theories and elementary novel functions for application on the flexible framework.
The following are conceived of as the most fundamental issues of common concern:
- integration of multi-modal information (and of heterarchical processing modules);
- learning and self-organization (optimization and adaptation).
Implementing these in a generalized and flexible framework of massively parallel and distributed information processing will be a key point in developing novel functions for application such as flexible recognition, inference, and control.
It will also be important to learn and get inspiration from nature, namely to take into account new findings in scientific research into brains, the evolution process of creatures, and ecological systems.
In what follows we will investigate the topics of theoretical research that will serve as the theoretical foundation for the novel functions in various applications and possibly for new computing architectures.
5.1 Flexible Representation of Information
In order to treat various kinds of information in the real world (such as images, speech sounds, and languages) and to construct heterarchical flexible systems, it is first of all necessary to establish a flexible framework for information representation. The framework should be flexible enough not only to represent various kinds of information in a unified manner but also to represent certainty of information. It should also be suitable and efficient for implementing associative memory and learning/self-organization procedures.
Important research topics include distributed representation (patterns) vs. concentrated representation (symbols), spatial representation vs. temporal representation, probabilistic representation, topological representation, hierarchical representation, and the representation of information and knowledge as constraints, etc.
5.2 Evaluation of Information and Processing Models
For actively interacting with the real world and learning or self-organizing from experience, RWC systems should have a flexible and systematic framework (criterion) for evaluating the importance of various kinds of input information, the output information that is a result of processing, and the processing models themselves.
One important issue will be to extend the evaluation framework from the conventional hard type, which mainly deals with "true/false," to more flexible and quantitative types, such as information criteria (e.g. AIC and MDL) and energy, which can serve as objective functions for optimization and regularization, including information-integration processes and processing models themselves.
5.3 Flexible Storage and Recall of Information
The highly sophisticated function of the human brain's memory is a key to the flexible information processing of humans. RWC systems should have such flexible memory functions to store and associatively recall various kinds of information. Thus the research should aim at establishing the theoretical analysis of associative memory and developing new efficient mechanisms for flexible associative memory.
Important research topics are association using probabilistic reasoning, association using structural similarity, associative memory for storing time series, and associative memory using non-linear dynamical systems, etc.
5.4 Integration of Information and of Processing Modules
Information processing such as inference, prediction, and planning can be considered as an integration process of various kinds of information and knowledge. It is important to carry out the theoretical research to analyse these information-integration processes and develop a new flexible way of controlling the processes.
Multi-variate data analysis methods will provide the methods of information integration, although most of those are limited to the linear transformations. Nonlinear extension of those methods will be important. Neural network models are considered as giving a kind of non-linear extension. Regularization theory will provide a method for incorporating various kinds of constraints.
Cooperative processing by a huge number of processing modules is also an important issue of information integration. Research on the integration of processing modules will be important. Investigating the information-integration process of human cognitive systems will also be important as a source of inspiration.
5.5 Learning and Self-organization
In addition to information integration, learning or self-organization is one of the most important topics of flexible information processing. It will play an important role in constructing adaptive autonomous systems, complex heterarchical systems or flexible databases. Thus, it is necessary to construct computational theory of learning/self-organization for exploring the "learnability" of various concepts or structures and to develop novel. efficient algorithms.
Important research topics are learning from uncertain information, learning of probabilistic knowledge, learning of heterarchically structured knowledge, learning algorithms using active information acquisition, methods for incorporating existing knowledge with learning process, and selection of learning models, etc.
5.6 Optimization Methods
Information-integration processes can be formalized as optimization processes. Learning and self-organization can also be considered as optimization procedure. Solving these optimization problems usually demands a huge amount of computation. In order to surmount this difficulty, developing approximately correct optimization methods that can be executed effectively on massively parallel systems will be an important area of research.
Important research topics include probabilistic optimization methods such as simulated annealing, genetic algorithms, ecological algorithms, evolutionary algorithms, optimization using non-linear dynamical systems such as neural network models, and other non-linear optimization techniques, etc.
6. Novel functions for application
The objective of research and development is to investigate elemental novel functions in the various application fields, closely cooperating with the theoretical foundations, and to embody those in respective application fields or integrate them to demonstrate new flexible information systems.
RWC systems support should approximate various human activities by acquiring and processing various kinds of information in the real world such as images, speech sounds, tactile sensations, and so on, which are massive and modal and moreover subjected to incompleteness and uncertainty by nature. Thus, the RWC systems require novel functions with flexibility of various kinds. Terms such as robustness, openness, and real-time reflect different attributes of flexibility. Therefore, the key issue for novel functions is how to realize such flexibility.
The novelty of the functions should emerge from new concepts of theory or algorithm suitable to RWC. The bottlenecks of conventional information processing should be alleviated through new kinds of flexible functions such as integration of symbol and pattern and learning/self-organization. Merely combining conventional technologies or making ad-hoc systems for specified tasks is not what is desired.
Since flexible information processing intends to expand the abilities of in formation processing beyond the limitations of the conventional one, the range of expected application fields is quite wide. These fields can be divided into the following categories:
- Flexible recognition and understanding of various kinds of information, such as pattern information - images, speech sounds, and symbolic information like natural languages;
- Flexible inference and problem solving based on flexible information bases that admit direct treatment of information and have capabilities of learning and self-organization;
- Flexible human interface and simulation for realizing mutual interaction between humans and the real world;
- Flexible control and autonomous systems interacting with the real-world environment.
The following are conceived of as important directions to pursue for realizing integrated systems of novel functions:
- Real-world, adaptable, autonomous systems;
- Information-integrating interactive systems.
The former means cooperation with the real world. These systems will be flexible systems that can autonomously understand and control the environment through active and adaptive interaction with the real world and work for the purpose of partial replacement of human activities in the real world. Here it will be necessary to cope with the uncertain, incomplete, and changeable characteristics of the real world. Necessary novel functions are understanding of environment, modelling of the real world, planning for action sequences, and optimal control for adaptation, etc.
The latter refers to cooperation with humans. These systems will be flexible systems that support and enhance human intelligent capabilities such as problem solving and information creation through enlarged communication channels between humans and systems. Here it will be necessary to both understand and integrate varied information flexibly to assist humans in solving problems and creating new information. Necessary novel functions are question and answer by spoken natural language, understanding of intentions from various types of information produced by humans, realizing intelligent and interactive assistance for retrieving and presenting valuable information from a large amount of data in database, intelligent simulation to create new information findings and forecast transient states in the real world, integration methods for combining human factors, and a computational model of the real world, etc.
The fundamental novel functions should be evaluated from the viewpoint of how they contribute to realizing these two systems and from the viewpoint of how they make breakthroughs in their own technological fields, related to the attributes of RWC, namely, robustness, openness, and realtime.
In the following, we will investigate research topics that seem necessary for the realization of both systems.
6.1 Flexible Recognition and Understanding
We can easily and flexibly recognize and understand various kinds of information in the real world, such as images and speech sounds. This ability is a typical example of intuitive information processing and will be indispensable for enabling RWC systems to possess autonomy and to undertake smooth communication with humans. It will also form the base for inference and problem solving at higher levels.
Many efforts have been made in research to mechanize these functions, but the state of the art is still far behind the level of human ability. The difficulty of realizing recognition/understanding in the real-world environment lies in the incompleteness, uncertainty, and ambiguity inherent to natural patterns and further in the ill-defined and ill-posed nature of the problems themselves.
Therefore, a new framework for solution is needed that admits these difficulties and can integratively include various constraints (knowledge) and also subjective value judgements, for example flexible frameworks such as optimization, constraint satisfaction, Bayesian inference, hidden Markov models, learning/self-organization from examples, or a new paradigm of parallel computation unifying these.
Important research topics are as follows:
(1) For image understanding, the formulation of processing modules for early visual information (such as colour, shape, movement) and integration mechanisms for these, interactive segmentation and understanding of scene and motion images on the basis of flexible modelling of objects and the world environment, understanding of facial expression and gestures to infer human intentions, etc.
(2) For speech understanding, noise suppression and robust recognition under noisy environments, speaker-independent understanding of conversational speech, formulation of heterarchical processing modules from the signal level to the linguistic level and their integration mechanisms for these, flexible mechanisms for syntactic and semantic analyses, etc.
(3) For natural language understanding, robust parsers applicable to large volumes of real-world coded sentences containing incompleteness, flexible methods to manage such sentence data as databases or electronic dictionaries and to utilize those for understanding, algorithms for extracting conceptual information, computational models for understanding situations and for integrating knowledge units to treat dialogue, etc.
6.2 Flexible Inference and Problem Solving
Conventional AI realizes intelligence by reducing human intelligence, such as inference and problem solving, to problems of logical operations and retrieval in knowledge representation by symbols. Such methods are efficient for restricted or rigorously abstracted problem domains, but for the variety of problem solving required in the real world, they face many difficulties, such as the limitation of symbolic representation and manipulation, combinatorial explosion and knowledge acquisition.
In the real world, much information and knowledge contains inconsistency, incompleteness, and uncertainty. There are many cases where the problems to be solved themselves cannot be described completely, and problem formulation is required based on a user's incomplete statements. In addition, real-time problem solving is also an important factor. Therefore, a new scheme of flexible inference and problem solving is necessary to cope with these problems. This will be attained by expanding the conventional framework to a more flexible one.
Important research topics will include flexible information representation integrating symbols and patterns, flexible knowledge acquisition from real-world information (learning of probabilistic structures or causal relationships), stochastic inference and analogy based on soft logic that integrates probability theory and fuzzy logic, constraint satisfaction or network dynamics to solve ill-posed problems or problems with many tacit assumptions as optimization problems, massively parallel processing for the fast solution, and modelling of the processes of inference and problem solving in humans, etc.
6.3 Flexible Information Bases
Flexible information bases to handle various kinds of information and knowledge in the real world are important in the sense that they will support intellectual activities of humans in the forthcoming era of high utilization of information networks and also in the sense that they will form a basis for novel functions such as flexible understanding and recognition, inference, problem solving, and action control.
So far, several types of databases have been proposed and developed, such as relational, object-oriented, deductive, and knowledge bases. However, there are still many problems with the conventional databases and knowledge bases, which rely mainly on symbolic and logical representation and retrieval.
Two new technologies should be focused on here. One is the flexible representation and memorization of the variety of real-world information within a unified framework for an inner information model ensuring tractability. The other is the flexible and semantically correct retrieval of the required items corresponding to the intentions of users.
Related important research topics include flexible information representation and data structures able to reflect the topological characteristics of objects and suitable for treating hierarchies of information; self-organization functions to cope with large-scale real-world information; learning of correspondences and cooperative relationships between various kinds of information; evaluation of information value and feature extraction for information abstraction or automatic indexing; detection of lacking information and active information acquisition; inference of the user's intention from incomplete or ambiguous requirements and completion through dialogue; and high-speed retrieval by massively parallel processing, etc.
6.4 Flexible Human Interface and Simulation
Novel ways of using RWC systems must be developed in parallel with the development of the individual novel functions mentioned above. This new information processing will provide a new environment that enables broad, cooperative relationships between humans and computers and enlarges human intelligence activities. Users will be able to interact with the computing systems by such natural means as spoken language, gestures, and facial expressions, and will be able to receive information by means of real-time 3-D images, etc. This means that users can concentrate their intelligence efforts on more creative activities, freed from learning the exhausting skills required for communication with conventional computers through narrow channels.
Related research topics to realize such an information processing environment will include broad-band multi-modal interface (linguistic and visual communication, including recognition of human body actions, facial expressions, etc.), information display system with virtual reality, cognitive and behaviour models including sensor fusion to understand human intentions.
RWC will also realize novel simulation technologies to solve very complicated and difficult problems. This will assist human thinking, creative activity, and decision-making by means of visualizing phenomena that have no inherent visual appearance. Instead of performing expensive and time-consuming physical experiments that may not always be precise, the computing system will provide users with powerful tools to simulate very large-scale complex systems and to predict their behaviour in real-time. Prediction of untapped phenomena and future events, such as the global environment and weather forecasting, will be an important application of RWC.
Related research topics are learning/adaptation-type simulation, prediction and control of complex/chaotic time series, large-scale simulation and decision-making support for solving environmental, economic, and traffic problems, etc.
6.5 Flexible and Autonomous Control
Research needs to focus on the development of the technologies required for the realization of flexible, autonomous, coordinated systems operating in the read world in real time. Robots are typical examples of such systems, and applications include aids for the elderly or physically handicapped.
Such systems will be composed of various functional modules that interact with each other in perception, decision-making, and action control. On the other hand, the real world where the system operates is subject to incompleteness and ambiguity in the available information, to dynamic change in the physical environment and limitations on available time and space.
Therefore, the important problems to be solved are how to integrate these various functions and how to control the interactions between the functional modules in order to achieve desired goal states under such real-world constraints.
Related research topics are flexible modelling of the environment, task and control, active and distributed sensing and sensory integration, on-site planning and distributed cooperative searching, structurization and coordination of multi-agents (for sensing, planning, and action) for real-time skilful manipulation of objects, and maintenance of consistency between the internal world model and the dynamic real world, etc.
Contents - Previous - Next