Computer Organization Laboratory

Robert H. Fujii Associate Professor	Satoshi Nishimura Associate Professor	Wanming Chu Research Associate

Robert Fujii: Projects carried out during the academic year 2004-2005 Spiking Neural Networks Spiking Neural Networks (SNN) use inter-spike time coding to process data. In conventional neural networks, the rate code is instead used. Anew learning algorithm for an SNN that uses spike trains (spike sequences) as inputs was proposed. This learning algorithm first decodes the input spike train into a spatio-temporal pattern and then uses simple criteria for learning. The algorithm was applied successfully on classification problems. An analog/digital implementation (simulation) of a spiking neural network decoder was carried out. Future work will involve mathematical modeling of basic parts of the brain with the aim of realizing practical artificial neural networks which in the future can be implanted in the brain. Autonomous Robotics Participated in the AIBO robot national competitions: 1)Basic Aibo movement competitions; 2)AIBO movement and image processing under changing lighting conditions; and 3) AIBO soccer game. Multi-agent cooperative behavior algorithm for a robot soccer game was researched; in this algorithm, robot agents learned to use an appropriate soccer team formation based on experience gained while playing against an opponent team. In addition, a cooperative robot behavior algorithm based on reinforcement learning for a single prey and multiple hunter environment was also developed. In the future work on cooperative robot behavior, close to real-time image processing for robots, and application of biologically inspired artificial neural/fuzzy approaches for robots will be explored. Computer-Aided Music A computer system capable of recognizing the main melody line of amusical piece was developed. Oncethe piece has been identified, the computer system is capable of playing the remainder of the musical piece. The system was developed using the software package MAX/MSP. A computer system which separates music into phrases and thematics subjects using the basic rules of the Generative Theory of Tonal Music (GTTM) was developed using the software package MAX. The purpose of the system is to analyze musical data by enumerating, grouping, and extracting features. Future work will explore computer-aided music arrangement and further research on musical analysis. Satoshi Nishimura: Hardware for realistic image synthesis, Computer architecture for computer graphics, Reconfigurable architecture for multimedia applications, and Realtime systems for computer music. Wanming Chu: Parallel Multithreaded Architectures This research focused on a Multiple-Threaded Multiple-Pipelined processor architecture by using the horizontal multithreading technique which tries to fetch, issue and execute multiple instructions from multiple threads in parallel in every clock cycle. Since there are multiple threads in our horizontal multithreading processor, the branch predictions may interfere each other. As a result,it will make the correct prediction rate low. We have implemented the several branch predictors schemes on proposed processor model.The simulation results will help us to make the design decisions. Architectures of interconnection networks Adual-cubeDC(m) has m+1 links per node where m is the degree of a cluster (m-cube), and one extra link is used for connection between clusters. The dual-cube mitigates the problem of increasing number of links in the large-scale hypercube network while keeps most of the topological properties of the hypercube network. The efficient algorithms for finding a nonfaulty routing path between any two nonfaulty nodes in the dual-cube with a large number of faulty nodes were proposed.

Refereed Journal Papers

[fujii-01:2004]	Hesham H. Amin and Robert H. Fujii. Spiking Neural Network Inter-Spike Time Based Decoding Scheme. IEICE Transactions on Circuits and Systems, to appear(1):to appear, 2005.
	A new scheme which utilizes the input inter-spike intervals for decoding an input spike train is proposed.
[fujii-02:2004]	Hesham H. Amin and Robert H. Fujii. Spike Train Classification Based on Spiking Neural Networks. WSEAS Transactions on Circuits and Systems, to appear(1):to appear, 2005.
	A new learning algorithm for a spiking neural network that uses spike sequence (spike train) times as inputs is introduced.
[nisim-01:2004]	Shin ichi Kinuwaki, Satoshi Nishimura, and Roman Durikovic. Single-Pass Volume Rendering on Programmable Graphics Hardware. Journal of 3D Images, 18(4):74-79, 2004.
	We present a novel method for interactive visualization of 3D volumetric data based on graphics processing units (GPUs). Volume rendering is a method of chioce for displaying 3D data fields obtained from CT or MRI. Recent commodity graphics cards for PCs haveGPUs which enable real-time volume rendering at very low cost. Previous methods for GPU-based volume rendering generate images through multiple passes with storing intermediate results on texture memory. The method proposed in this paper requires fewer or even a single pass using the direct calculation of ray direction vectors on pixel shaders and thus achieves higher efficiently in both speed and memory requirements. Our volume renderer implemented on GeForce 6800 reduces 30% of the setup processing time in comparison with the previous techniques.

Referred Proceeding Papers

[fujii-03:2004]	Hesham H. Amin and Robert H. Fujii. Input Arrival-Time-Dependent Decoding Scheme for a Spiking Neural Network. In Michel Verleysen, editor, The 12th European Symposium of Arti??cial Neural Networks, pages 355-360, Bruges, Belgium, Apr. 2004. IEEE, d-side.
	A new method for decoding input spikes according to their absolute arrival times is proposed.
[fujii-04:2004]	Hesham H. Amin and Robert H. Fujii. Spike Train Decoding Scheme for a Spiking Neural Network. In Michel Verleysen, editor, The 2004 International Joint Conference on Neural Networks, pages 477-482, Bruges, Belgium, July 2004. IEEE, d-side.
	A decoding scheme for a spike train which can uniquely distinguish them is propsed.
[fujii-05:2004]	Yoshio Konishi and Robert H. Fujii. Incremental Learning of Temporal Sequences Using State Memory and a Resource Allocating Network. In Michel Verleysen, editor, The 12th European Symposium of Artificial Neural Networks, pages 2889-2893, Bruges, Belgium, July 2004. IEEE, d-side.
	A network which can learn temporal patterns incrementally is proposed. Gaussian functions are used to represent various output values. Chunking is used to group similar patterns.
[w-chu-01:2004]	Yamin Li, Shietung Peng, and Wanming Chu. Binomial-Tree Fault Tolerant Routing in Dual-Cubes with Large Number of Faulty Nodes. In Ji-Huan He Jun Zhang and Yuxi Fu, editors, Proceedings of the International Symposium on Computational and Information Sciences, pages 51-56, Shanghai, China, December 2004. Donghua University, Lecture Notes in Computer Science (LNCS 3314), Springer.
	A dual-cube DC(m) has m+1 links per node where m is the degree of a cluster (m-cube), and one extra link is used for connection between clusters. The dual-cube mitigates the problem of increasing number of links in the large-scale hypercube network while keeps most of the topological properties of the hypercube network. In this paper, we propose efficient algorithms for finding a nonfaulty routing path between any two nonfaulty nodes in the dual-cube with a large number of faulty nodes. A node v 2 DC(m) is called k-safe if v has at least k nonfaulty neighbors. The DC(m) is called k-safe if every node in DC(m) is k-safe. The first algorithm presented in this paper is an off-line algorithm that uses global information of faulty status. It finds a nonfaulty path of length at 2 most d(s;t)+O(k ) in O(jFj +m) time for any two nonfaulty nodes s and t k in the k-safe DC(m) with number of faulty nodes jFj < 2 (m+1 ?? k), where 0 ?? k ?? m=2. The second algorithm is an online algorithm that uses local information only. It can find a fault-free path with high probability in an arbitrarily faulty dual-cube with unbounded number of faulty nodes.

Academic Activities

[nisim-02:2004]

Satoshi Nishimura, November 2004. Organization Chair, 57th IPSJ SIGMUS conference

Ph.D. and Other Theses

[fujii-06:2004]	Ogura Hideki. Graduation Thesis: Development of AIBO's Moving and Image Processing Library and Application for Goal Keeper, University of Aizu, 2005. Thesis Advisor: Fujii, R.
[fujii-07:2004]	Sugawara Yosuke. Graduation Thesis: A Study of Music Player System Following Inputting Song Intro, University of Aizu, 2005. Thesis Advisor: Fujii, R.
[fujii-08:2004]	Yoshida Kouji. Graduation Thesis: Automated Color Detection Table Generation for RoboCup Soccer Field, University of Aizu, 2005. Thesis Advisor: Fujii, R.
[fujii-09:2004]	Yamakami Norihisa. Graduation Thesis: System Control Program Organization Dependent Robot Control EAEciency, University of Aizu, 2005. Thesis Advisor: Fujii, R.
[fujii-10:2004]	Sato Takashi. Graduation Thesis: Enhancement of Cooperative Behavior through Soccer Formation Learning, University of Aizu, 2005. Thesis Advisor: Fujii, R.
[fujii-11:2004]	Maeda Tomoka. Graduation Thesis: Self-Localizing Robots, University of Aizu, 2005. Thesis Advisor: Fujii, R.
[fujii-12:2004]	Suzuki Emi. Graduation Thesis: A Max-Based Implementation of Grouping Rules of GTTM, University of Aizu, 2005. Thesis Advisor: Fujii, R.
[nisim-03:2004]	Yutaka Matsuzaki. Graduation Thesis: Development of the Samuel-to-LilyPond translator - Music score publishing using the Samuel language, University of Aizu, 2004. Thesis Advisor: Satoshi Nishimura
[nisim-04:2004]	Takeru Ishikawa. Graduation Thesis: Evaluation of Music Programming Systems based on the Delay TimeComparison of MIDI Signals, University of Aizu, 2004. Thesis Advisor: Satoshi Nishimura
[nisim-05:2004]	Koji Itagaki. Graduation Thesis: Hardware Design for the Karplus-Strong Algorithm with Vibrato Effect, University of Aizu, 2004. Thesis Advisor: Satoshi Nishimura
[nisim-06:2004]	Kozue Sawada. Parallel Processing of the Two-Demensional Partially Solving Method, University of Aizu, 2004. Thesis Advisor: Satoshi Nishimura
[nisim-07:2004]	Asuka Takayasu. Graduation Thesis: Cache Evaluation of Multi resolution Volume Rendering using Wavelet Transformation, University of Aizu, 2004. Thesis Advisor: Satoshi Nishimura
[nisim-08:2004]	Yusuke Mizuguchi. Hardware Implementation of the Ray-Flow System for Real-Time Ray Tracing, University of Aizu, 2004. Thesis Advisor: Satoshi Nishimura

2006 THE UNIVERSITY OF AIZU ALL RIGHTS RESERVED