Various PR China science activities, related to computers & policy

[David Farber <farber () central cis upenn edu>]

From: 
 Dr. David K. Kahaner
 US Office of Naval Research Asia
 (From outside US):  23-17, 7-chome, Roppongi, Minato-ku, Tokyo 106 Japan
 (From within  US):  Unit 45002, APO AP 96337-0007
  Tel: +81 3 3401-8924, Fax: +81 3 3403-9670
  Email: kahaner () cs titech ac jp

ABSTRACT. Various PR China science activities, related to computers & policy.


Most of the following material is from newspapers, journals, etc.

Parallel Computer Architectures for Machine Vision

Scientists from Qinghua University have published details of two
competing three-level hybrid parallel computer architectures for machine
vision: one based on two systems, the other on three systems.  Using
these architectures, the authors have implemented a number of
algorithms, including stereo vision matching and real-time obstacle
detection for a mobile robot.

Scientists in Qinghua University's computer department have published
details of their research on two parallel computer architectures for
machine vision.  The first, described in the Shenyang Journal Xiaoxing
Weixing Jisuanji Xitong (Nov 92), consists of Aspex Inc's
pipelined image processing engine (PIPE) for low-level tasks such as
sensor input, preprocessing, and image segmentation, and for a portion
of the mid-level tasks (segmentation to description).  A transputer
network was chosen for the remaining mid-level tasks as well as
high-level tasks such as recognition and interpretation.  The pipe is a
single-instruction-stream multiple-data-stream (SIMD) parallel system,
while the transputer network is a multiple-instruction-stream
multiple-data-stream (MIMD) parallel system.  The overall parallel
architecture is therefore a mixed-mode one that is dynamically
reconfigurable (every 1/60 second) in the pipe portion and is
reconfigurable in the transputer portion via program revisions.

This architecture was developed for research on an integrated
intelligent mobile robot vision/navigation system.  The pipe is a
real-time pipelined image processor that carries out one pipeline
operation per video field period (1/60 second).  It consists of a video
interface, an input module, three to eight modular processing stages
(MPSS), an output module, an inconic-to-symbolic mapping (ISMAP) module,
and six video buses.  Each MPS contains three look-up tables (LUTS) and
two arithmetic logic units that form  a multifunctional input device;
two local memories, each capable of storing four 256x256-x8-bit images;
one preoperation LUT; two parallel convolvers with four operating modes
(3x3 arithmetic, 9x1 arithmetic,  3x3 boolean convolution, and 9x1
boolean convolution); and one two-value-function LUT.  The pipe has
three local data channels or paths (forward, recursive, and backward)
and six global data channels (the video buses).  A transputer network
allows several topological structures.  For the host, one can use either
a PC-bus-based microcomputer such as a 386 or a VME-bus-based Sun
workstation.  The Qinghua University scientists studied both the
five-processing-element TMBO4 transputer acceleration board (PC bus) and
the 16-processing- element VME-bus-based acceleration board.  On the
TMBO4, one of the processing elements is the master node, with 4
megabytes of memory, for communications with the host and for
management; the other four processing elements each have 1 megabyte of
local memory.  Using this pipe + transputer parallel architecture, the
scientists have implemented a number of algorithms, including stereo
vision matching, image compression coding, real-time obstacle detection,
and color channel segmentation.

The second of the two competing architectures, described in the Beijing
Journal Jisuanji Xuebao (Nov 92), consists of a digital signal processor
(DSP)-based low-level Vision Processing Module (VPM), with Iconic I/O;
the pipe as a mid-level VPM, with Iconic input but symbolic output; and
the transputer-based Parallel Graph Reduction (PGR) machine as a
high-level VPM, with symbolic I/O.  Working on the same platform--an IBM
PC/AT or other 386 running the TDS 700D transputer development
system--these three modules make up a hybrid vision computer, seen as an
efficient solution for processing images with large volumes of data and
critical real-time requirements, such as in mobile robot vision.

Perception and preprocessing is done in the DSP-based low-level VPM,
some segmentation is also done there, but most is done in the mid-level
VMP. Description, recongition and comprehension are done (mostly) at the
highest level. 

The independently designed DSP-based low-level VPM is composed of two
basic components.  A camera video signal acquisition, storage, and
display system converts the video input into three 512x512x8 frames
stored in frame memory devices (one frame buffer stores the original
input image while the other two store processed results, such as
convolution of most significant bit and least significant bit).  The
digital signals are then converted to analog and sent through a color-
matching board, then to a color display.  The output portion of this
entire system incorporates Inmos Limited's IMSG 175 color LUT,
permitting 256K false-color varieties.  The other basic component of the
low-level VPM is the image preprocessor, which comprises a DSP unit and
an I/O LUT.  The LUT incorporates 256x8x4 SRAMs and has programmable
transform functions.  The heart of the image preprocessor is the INMOS
A110 image/signal processing subsystem, which has 21 multiplier/adders
providing a performance of 420 million operations per second.

The mid-level VPM consists of the PIPE, providing image segmentation,
description, and recognition.  With the PIPE, the authors have
implemented KG-transform-based real-time (0.5-second) obstacle detection
in a mobile robot vision system and have developed a stereo vision
parallel algorithm based on a  loose matching method.  Experiments show
that for a 256x256 image with a parallax of eight pixels, the algorithm
can be run in 10 seconds.

The high-level VPM consists of the parallel graph reduction (PGR)
simulation system developed by the Qinghua University scientists.  This
system consists of two B003 boards, each with four processing element
(PE) modules (each PE module contains one 20-megahertz T414 transputer
and 256K Dram), and one B004 board containing one T414, 2 megabytes of
RAM, and IBM PC BUS interface logic.

--------------------------------------------------------------------------

Galaxy-II Supercomputer

The 1 GOPS Galaxy-II, the first domestic Chinese general-purpose 1 GOPS
parallel processing supercomputer, was developed in 53 months by more
than 260 specialists in the computer institute at the University of
Science & Technology for National Defense (USTND) in Changsha.  The
system was formally certified in November 1992.  The computer, housed in
a 3-meter-high hexagonal case, has a shared main memory tightly coupled
with four central processors linked by high-speed synchronous
communications channels.  Master clock speed is 50 megahertz (20ns),
main memory capacity is 256 megabytes, and there are two independent
10-megabit-per-second I/O subsystems.  For some computations, the
average speed-up ratio is 3.537, maximum speed-up radio is 3.8 (out of a
theoretical maximum of 4.0 for a four-processor system).  Basic word
length is 64 bits, and 64-bit peak performance is 400 MFLOPS.

Also passing appraisal were an independently designed operating system
occupying over 100,000 lines of code, plus development and applications
systems.  A medium-range (5- to 7-day) weather forecasting software
system jointly developed by USTND and the State Meteorology Center was
one of the applications systems that passed appraisal.  It was run on
the Galaxy-II as part of the rigorous tests conducted for certification.

Note: Cray Research introduced its XMP line of supercomputers in
1982 and delivered the four-processor version a year later.  The first
public customer to fully utilize the XMP-4 system was the European
center for medium range weather forecasting (ECMWF), whose mission is
identical to the meteorological program that China has stated was the
impetus for the Galaxy-II's development.  While the floating-point
computational speed quoted for the Galaxy-II is much less than that of
the XMP, the Galaxy-II's 1-GOPS operational speed might make it
comparable for some specialized, nonscientific applications such as
signal processing.  The 3.8 maximum speed-up ratio cited for the
Galaxy-II is similar to that obtained by ECMWF after installing its
weather forecasting models on the XMP-4, as documented in published Cray
Research technical information, although I cannot tell if the Galaxy's
published speed-up was for the weather model.

--------------------------------------------------------------------------

China Implements Technology Research Center Construction Plan

Beijing ZHONGGUO KEXUE BAO (CHINESE SCIENCE NEWS) in Chinese 8 Dec 92 p1

"China Implements Technology Research Centers Plan to Promote
Transferring of Science and Technology Achievements, and To Improve
Industrial Competitiveness")

Based on China's experiences in planning for science and technology
(S&T) development in the last decade, and the need of instituting
China's socialized market economy, the State Planning Commission (SPC)
has officially formulated a plan to implement National Technology
Research Centers (NTRC) to speed up the transfer of S&T achievements to
production; to promote the integration of economy and S&T, as well as
the integration of research institutes and enterprises; and eventually
to strengthen the industrial competitiveness of China.

Beginning in 1988, on the basis of instituting the plan to tackle
China's key S&T problems and the plan for China's key industrial
experimentations, SPC, relying on the nation's major institutes and
colleges, has selected seven meritorious items to launch the
establishment of NTRC's test projects to strengthen the weak link of
technologizing the research achievements.  These centers are:

  Chinese Academy of Science Shanghai Microelectronics Engineering 
       Research Center, 
  MMEI's Beijing Large-Scale Integrated Circuit CAD Research Center, 
  Semiconductor Basic Materials Technology Research Center of the China
       Non-Ferrous Metals Institute,
  Industrial Automation Technology Research Center of Zhejiang University, 
  Optical Communications Technology Research Center of Wuhan Posts and 
       Telecommunications Research Institute, 
  Dyestuff Technology Research Center of Shenyang Chemical Engineering
       Institute, 
  Optoelectronic Device Technology Research Center of CAS.  

The total investment in these seven items will amount to around 300
million yuan and will be borne by the SPC, other concerned departments
and the organizations undertaking the projects.

Through preliminary practice and repeated adjustments and studies, SPC
will implement the NTRC construction plan during the Eighth 5-Year Plan.
NTRC will conduct research on the systematic integration of key
technologies and general technologies, on technology industrializing for
China's industrial development, as well as on the realization of
transferring the S&T achievements to markets or enterprises.  NTRC's
missions are: 

  First, to carry out the systematic integration of key technologies and
general technologies and their industrialization during industrial
development.

  Second, the introduction, digestion, absorption and then improvement
of advanced technologies and applied technologies.

  Third, the cultivation of high-quality technological personnel to
provide information and consultation services for industrial
development.


The plan's preliminary goals during the Eighth 5-Year Plan and the early
Ninth 5-Year Plan as decided by SPC are as follows: By unifying a few
hundred large- and medium-sized enterprises or a few dozen large
conglomerates, a group of technology research centers will be
established among China's concerned research organizations, enterprises,
and universities.  The centers, supported by enterprises and government,
will become a strong S&T achievement conversion force.  Based on market
demands, a few hundred important research achievements will be
transferred to industries every year. To ensure the implementation of
this plan, SPC, through policy and investment guidance, will make both
the enterprises and the concerned organizations participate in the NTR
construction. SPC will mobilize the necessary capital to carry on the
work. Overall, the ratio of government investment to the combined
capital of concerned departments, enterprises, and construction units
will be about 1:1.  These technology research centers will be completed
in 1997. Their functions mainly fall in the fields of

   electronic information technology and its application,
   chemical engineering, 
   new materials, and 
   high-efficiency energy utilization, as well as environmental protection.

In the formulation and the organization of the NTRC plan, SPC adopted
the principle of "expert evaluation, selection of the best through
competition, and the SPC summation for balance."  The selection and
evaluation of the NTRC projects will be entrusted to experts, and
research will proceed according to pertinent guidelines and scope.

--------------------------------------------------------------------------

Basic Research Significant to National Defense Policymaking

Beijing KEJI RIBAO (SCIENCE AND TECHNOLOGY DAILY) in Chinese 4 Dec 92 p6

"Importance of Basic Research as Revealed From Policy Decisions of
'Missile and Bomb' Development")

The experience of China's defense build-up has proven that "winning the
basic research battle" is the prerequisite for building up national
defense, because theories often affect policy decisions, especially
crucial policy decisions.

For example, as early as 1955, the important decision to be made was
which air-defense course China should adopt to strengthen national
defense.  Naturally the idea was to develop defense aircraft, but from a
deeper theoretical point of view, China needed guided missiles first.
Missiles are superior to airplanes for both offense and defense, as they
have much higher Mach numbers than airplanes.  However, at that time the
first satellite of the former USSR had not been launched, the experiment
of intercontinental ballistic missile had not succeeded yet, and there
was no consensus as to whether missile technology could become the
reality of defense technology.  China was then backward in science and
technology (S&T) and unable to develop missile technology.  Whether
China might possibly make a strategic policy mistake was the great
problem.

At that time, Professor Qian Xuesen, fresh from abroad, prevailed over
all dissenting views and pointed out that China should develop missiles,
for the significant reason that to master and develop missile or rocket
technology was not necessarily more difficult than developing aircraft.
The missile materials would be used only once while the airplane
materials would require  repeated uses, hence, the aircraft engine and
other structures demanded specific materials. This materials development
that required a long period of cumulated experiences could be ignored if
China had decided to develop missiles. The only problem of missile
development was guidance. A breakthrough on guidance could be achieved
within a short period of time. In addition, the Chinese people are
intelligent. The important policy decision that China should develop
guided missiles was based on the theoretical analysis of China's
internal conditions and the prospect of international technological
development.

In August 1992, China launched a 7-ton Australian satellite into orbit.
The news elated the whole nation. It was exactly the outcome of the
policy made during 1955 and 1956. In retrospect, the success of this
important policy owes to the prediction that a breakthrough of the
guidance problem could be resolved in a short period of time.

There is another example involving policymaking on the most advanced
defense technology. In 1955, China decided to develop atomic energy,
i.e., to build atomic bombs.  Which technological route should China
follow?  Should it be developing Pu239? or U235?  or U233?  It was easy
to rule out U233 which would involve the thorium industry, hence, both
the uranium system and the thorium system would be involved.  At that
time China could only work on one system.  In the uranium system, there
were also two technologies: producing Pu239 from a reactor, or
developing the method of producing U235 through isotope separation.  At
first glance, the method of separating Pu235 from the mixed system of
U235 and U238, because the former used chemical separation and the
latter, physical separation.  Meticulous research indicated that as far
as separation was concerned, chemical separation should be easier than
physical separation.  However, technologically, separation of plutonium
and uranium was not very easy due to the high radioactivity of the Pu-U
system, and the severe toxicity of plutonium.  To ensure workers'
safety, the cleanup technology would be difficult, moreover, it would
take a long time to cool down the radioactivity.  Therefore, the best
choice should be the separation of U235 and U238 isotopes to speed up
the atomic industry.

Furthermore, technical publications at the time indicated that there
were two methods to detonate  an atomic bomb: one was called the
"gun-barrel" method which detonated the bomb by closing up quickly two
U235 hemispheres together; the other was called the "implosion" method
which detonated the bomb by using dynamite  explosion to quickly shrink
the size of a Pu239 sphere.  The implosion method was considered more
technically advanced, and it would detonate the bomb with higher
destructive power than the gun-barrel method.  The problem was whether
China was capable of producing a uranium bomb detonated by the implosion
method.  It was an unknown factor to the Chinese people at that time!
However, the theoretical physicists in the Ministry of Nuclear Industry
worked together and concluded that theoretically it was entirely
possible to detonate a U235 bomb through implosion.  The participating
theoreticians recalled that should be theoretical research have wavered
then, the complete industry structure would have to be redeployed.
Therefore, the course of atomic energy development has further proven:
win the theoretical war first!

The "missile and bomb" problems were resolved through the cooperation of
many scientists and technologists, workers, officers and men of the
People's Liberation Army, political workers, logistical workers, as well
as leading cadres.  Nevertheless, tracing back to the route of success,
the foremost breakthrough was the winning of the theoretical battle.
"Winning the theoretical battle" is  not only the conclusion from
experiences in developing the most advanced defense technology but also
the important principle that must be considered in building up China's
economy.


-------------------------------------------------------------------------

Accelerating Reform of R&D Institutes for National Defense Urged

Beijing KEJI RIBAO (Science and Technology Daily) in Chinese 30 Nov 92 p3

Article by Armaments Industry Corporation Deputy Office Chief, Wu Shanggui

The 14th National Party Congress clearly promotes the establishment of a
socialist market economy system, which has interested the great majority
of China's scientific organizations into converting from utility
organizations to management organizations in keeping with the new
dynamic mechanism of the market economy system.  Most of China's
military industry R&D institutes are still of the 1960s, locked into the
former Soviet model, and still hanging on to the framework of the
planned economic system, subordinate to the compartmented ministerial
scientific research system, which obviously is not compatible with the
needs of today's national defense scientific research structure and the
market economy.  There many drawbacks to that kind of scientific
research system:

1. Secular specialized organizations, and S&T and economics are "cuts of
different hides".  The scientific research academies and institutes of
China's military industries are singular military entities whose main
duty is toward the "military battlefield", and the research and
manufacture of new armaments for military units.  Now, the mission of
the military industry's scientific research academies and institutes has
changed from being mainly for the "military battlefield" to being
simultaneously directed at the "economic construction battlefield".  The
thought processes of the specialized secular organizations of the
military industry scientific research institutes cannot change on the
spot, and in a rather short period military industry S&T and economic
construction have become cuts of different hides.

2.  Huge organizations and units, "Large temple, lots of saints" and
"Many monks, little porridge", reduces the military industrial R&D to
self subsistence and tough going.  In the past, this enormous military
industrial S&T outfit was raised and nourished from "The emperor's
party".  In form, its administrative structure was multilayered,
irrespective of the size or authorized strength of institutes.  The
staff and workers reclined and fed on the corpus of the institute, and
the institutes reclined and fed on the corpus of the country.  Their
assignments were passed down from above, funds and materials were handed
down from the state; the state took care of everything.  And so things
have gone on for some time, and a mentality of dependency -- "wait,
lean, and  want" -- evolved.  When state scientific research dwindled it
was tough going for the military industry academies and institutes to
subsist and develop.

3.  The talent drain is serious, the talent structure is not rational,
and the experience structure is deformed; small at the top and bottom,
and fat in the middle.  Generally speaking, the best shape for a stable
and rational experience structure is a "pyramid".  A study of 29
specialized institutes in the armaments industry showed that high-grade
professionals make up 14.6 percent, middle level professionals 43.7
percent, and beginning professionals 41.7 percent.  The reason for this
distorted structure, besides the specialized technical positions
appointment system itself being less than perfect, and selection
controls for appointments to high-grade technical positions index now
being defunct, is that a fair amount of defense industry institutes are
located in third line areas where conditions are backward, and the loss
rate for university graduates assigned to those areas is over 30
percent.

To deal with these problems, and to seize the present opportunity to
further bring to fruition the spirit of the 14th National Party
Congress, and step up the pace of reform in China's military industry
S&T system, the following reform measures should be taken:

1.  Implement a new dynamic mechanism, of "integrating the military and
the people" and "one institute, two systems".  In the leadership system,
the command administration, plans & programs, and even in the S&T
elements of the military industry academy and institutes, the principle
of "One divided into two" should be put into effect.  For military
products R&D the directive planning and administrative mechanism should
remain in effect.  The development of civilian products should be
directed toward the market.  A principle of "Hold on to what is critical
(military R&D), let the other loose" (civilian product development)
should be final.  The arms industry, for example, at present has a total
military products output value of only 30 percent of the total value of
production output.  In order to accelerate the pace of the reform for
"protecting what is critical, and letting the other go the limit", the
practice of "storing water to raise fish" should be pursued.  Keep on
with the 30 percent core of military products R&D geared toward the
"military battlefield" and the research and manufacture of advanced
weapons and facilities  for military units; and let the other 70 percent
of the S&T talent "open the flood gates and let the fish go to the sea",
and engage in the economic construction battlefield.

2.  "Take down the temple and disperse the saints:, erect a "small
booth, high level" military industrial R&D organization.  First,
dismantle and compress the unnecessary multilayered administrative
technical office organizations, and reduce the administrative and
odd-jobs ratio.  Then, to resolve the "many monks, little porridge"
conflict, the government should greatly increase investment in military
industry R&D, and not allow the longstanding "half starved" armaments
research institutes situation to continue.  And third, since it is
evident that the administrative expenses are already very limited, they
will surely be unable to support a large military industrial academy and
large institutes.  The principles of "everything for the people: and
"operating expenses according to number of people" as applied to
military industrial scientists and technicians should be redirected
toward subsidizing the maintenance of military product S&T, raising
investments in military products R&D, and guaranteeing a sustained,
stable, and coordinated development of the national defense R&D.

3.  Make a well managed "development of talent" one of the long range
strategies for reform of the military industry R&D system.  Development
and adjustment, gradual improvement of the structure, and rational
deployment, will be helpful for "maintaining the military and
transforming the people", and developing the military industry S&T
system of the commercial economy.  Especially with respect to the
military industrial scientists and technicians in the backward
environments of the third line, in order to ensure the completion of
major military R&D projects for national security, there is a need to
raise the wage level on a large scale and to implement key protective
policies.  And, preferential policies should be formulated and perfected
to attract S&T talent to the remote third line areas, and to prevent and
negate this sort of disregard for national interests and "contrary"
interest in seeking only personal gain in rejecting third line area
assignments.  And again, there must be a strengthening of spirited
cultural construction, and a stronger and more stable military industry
S&T force in remote areas.  Scientists and technicians must accept their
selection and assignment in the interest of the Party and the country,
and as the country needs to send them to remote and backward areas to
work they should blend their intelligence and wisdom into socialist
modernization and construction.