Real World Intelligence

THeme and Laboratory	Outline
The Design and Implementation of an ALU Based Reconfigurable Adaptive Device Adaptive Device NEC Laboratory http://www.rwcp.or.jp/activities/achievements/AD/nec/eng/home-e.html	We have developed a general-purpose accelerator prototype system that can process a large amount of data such as images, speech, etc. at high speed in place of a CPU. We introduce the adaptive device developed for the general-purpose accelerator, the programming environment and the application of the prototype to image processing system. The developed adaptive device integrates 6 x 63 9-bit ALUs (Arithmetic and Logic Unit) and the functions and connections of the ALUs can be freely reconfigured in accordance with the algorithm. By reconfiguring the hardware that is optimal to the algorithm, high-speed processing one order of magnitude faster than software processing by a CPU can be achieved. The general-purpose accelerator prototype system contains two of these newly-developed adaptive devices and local memories. When this is inserted into the PCI slot of the host, it works in cooperation with the host CPU. The programming environment is composed of a mapping tools that synthesizes configuration data for the adaptive device from the application program written in C.
Artificial Reality Operation Theoretical Foundation GMD LAboratory http://www.rwcp.or.jp/activities/achievements/TA/gmd/RWC-GMD/	New computational models are essential to understand today's world. Artificial reality is an answer to these needs. It reproduces complex real world systems on computers with such a detail not possible before. Artificial reality systems are used for simulation, but increasingly for planning and control of real world systems also. The key technology for implementing artificial reality systems are multi-agent systems with hundred thousands of virtual agents running on a network of computers. The GMD laboratory has investigated the theoretical foundations of such systems and has developed a special software called "Flip Tick Architecture" for easing the implementation of large multi-agent systems. With this architecture the laboratory has implemented a typical artificial reality system, the traffic in cities. The system is based on a microsimulation of individual cars, buses and trucks driving on a network of roads with intersections and traffic lights. The artificial reality system will be calibrated with the traffic of the city of Bonn. The system is intended to be used for planning (e.g. scheduling of construction sites). In a next step it will be used for intelligent traffic control. This requires a data exchange between the artificial and the real system in real time.
Reasoning with uncertainty Theoretical Foundation SNN Laboratory http://www.rwcp.or.jp/activities/achievements/TA/snn/RWCP_snn2.html	One of the most striking differences between human and computer performance is that humans are capable to process information that is noisy, often incomplete and sometimes contradictory. Examples are recognition of speech at crowded parties or decision making by managers. Our research is aimed at the design of novel methods that mimick such behaviour in computer programs. It turns out that the language of ordinary probability theory is most suited to represent uncertainty. This language is very rich and provides a hybrid framework to represent pure rules ("after rain comes sunshine") as well as relations that are learned from data in an integrated manner. A consequence of the probabilistic approach is, however, that computation is too slow for many practical applications. Our research has developed approximation methods that solve this problem to a large extend. The methods are illustrated on the problem of medical diagnosis.