Abstracts of IEEEPapers of
Univ.Doz. Dipl.Ing. Dr. Michael Heiss
 Heiss, M.: Lernen niedrigdimensionaler Kennfelder (German,
English). Habilitationsschrift an der Technischen Universität Wien, Mai 1995.
 Heiss, M.: Online learning or tracking of disrete
inputoutput maps. IEEE Transactions on Systems, Man, and Cybernetics, 27A(5): 657668,
Sept. 1997.
Online learning or tracking of discrete inputoutput maps
 Heiss, M.
Siemens AG, Wien, Austria
This paper appears in: Systems, Man and Cybernetics, Part A, IEEE Transactions
on
On page(s): 657  668
Sept. 1997
Volume: 27 Issue: 5
ISSN: 10834427
References Cited: 38
CODEN: ITSHFX
INSPEC Accession Number: 5686460
Abstract:
This paper shows how a slowly timevarying nonlinear mapping can be learned, if, for every
possible input value, the corresponding estimated output value is stored in memory. This
representation form can be called "flash map", or pointwise representation, or
lookup table. Thus, very fast access to the mapping is provided. The learning process is
performed online during regular operation of the system and must avoid "adaptation
holes" which could occur when some of the points are more frequently updated than
other points. After analyzing the problems of previous approaches we show how radial basis
function networks can be modified for flash maps and present the tent roof tensioning
algorithm which is exclusively designed for learning flash maps. The convergence of the
tent roof tensioning algorithm is proved. Finally, we compare the two approaches
concluding that under the flash map restriction the tent roof tensioning algorithm is the
better choice for learning lowdimensional mappings, if a polygonal approximation of the
desired mapping is sufficiently smooth.
Index Terms:
feedforward neural nets; online learning; discrete inputoutput maps; timevarying
nonlinear mapping; flash map; radial basis function networks; tent roof tensioning
algorithm; convergence; nonlinear mappings; polygonal approximation; associative memory;
generalisation; interpolation
PDF FULLTEXT
 Heiss, M. : Errorminimizing deadzone for basis function
networks. IEEE Transactions on Neural Networks, 7(6):15031505, 1996.
Errorminimizing dead zone for basis function networks
 Heiss, M.
Inst. fur Allgemeine Elektrotechnik Automobilelektronik, Tech. Univ. of Vienna, Austria
This paper appears in: Neural Networks, IEEE Transactions on
On page(s): 1503  1506
Nov. 1996
Volume: 7 Issue: 6
ISSN: 10459227
References Cited: 14
CODEN: ITNNEP
INSPEC Accession Number: 5441376
Abstract:
The incorporation of dead zones in the error signal of basis function networks avoids the
networks' overtraining and guarantees the convergence of the normalized least mean square
(LMS) algorithm and related algorithms. A new socalled errorminimizing dead zone is
presented providing the least a posteriori error out of the set of all convergence
assuring dead zones. A general convergence proof is developed for LMS algorithms with dead
zones, and the errorminimizing dead zone is derived from the resulting convergence
condition. The performance is compared with the performance of classical dead zones.
Index Terms:
feedforward neural nets; errorminimizing dead zone; basis function networks; error
signal; least mean square algorithm; convergence
PDF FULLTEXT
 Hofbauer, A. and Heiss, M.: Divergence effects for online
adaptation of membership functions. Intelligent Automation and Soft Computing, 4(1):
3952, 1998.
 Heiss, M. and Kampl, S.: DoubleExponential Sigmoidal
Functions for Neural Networks. Elektrotechnik und Informationstechnik e&i 114(7/8):
360363, 1997.
 Heiss, M. and Kampl, S.: Multiplicationfree radial basis
function network. IEEE Transactions on Neural Networks, 7(6):14611464, November 1996.
Multiplicationfree radial basis function network
 Heiss, M.; Kampl, S.
Inst. fur Allgemeine ElekrotechnikAutomobilelektronik, Vienna Univ. of Technol., Vienna,
Austria
This paper appears in: Neural Networks, IEEE Transactions on
On page(s): 1461  1464
Nov. 1996
Volume: 7 Issue: 6
ISSN: 10459227
References Cited: 22
CODEN: ITNNEP
INSPEC Accession Number: 5441371
Abstract:
For the purpose of adaptive function approximation, a new radial basis function network is
proposed which is nonlinear in its parameters. The goal is to reduce significantly the
computational effort for a serial processor, by avoiding multiplication in both the
evaluation of the function model and the computation of the parameter adaptation. The
approximation scheme makes use of a gridbased Gaussian basis function network. Due to the
local support of digitally implemented Gaussian functions the function representation is
parametric local and therefore well suited for an online implementation on a
microcomputer. A gradient descent based nonlinear learning algorithm is presented and the
convergence of the algorithm is proved.
Index Terms:
feedforward neural nets; multiplicationfree radial basis function network; adaptive
function approximation; computational effort; serial processor; gridbased Gaussian basis
function network; digitally implemented Gaussian functions; microcomputer; gradient
descent based nonlinear learning algorithm; convergence
PDF FULLTEXT
 Halper, Ch., Heiss, M., and Brasseur, G.: Digitaltoanalog
conversion by pulsecount modulation methods. IEEE Transactions on Instrumentation and
Measurements, 45(4):805814, August 1996.
Digitaltoanalog conversion by pulsecount modulation methods
 Halper, C.; Heiss, M.;
Brasseur, G.
Inst. fur Allgemeine Elektrotech. Automobilelektronik, Univ. of Technol., Vienna, Austria
This paper appears in: Instrumentation and Measurement, IEEE Transactions on
On page(s): 805  814
Aug. 1996
Volume: 45 Issue: 4
ISSN: 00189456
References Cited: 22
CODEN: IEIMAO
INSPEC Accession Number: 5338056
Abstract:
Three lowcost digitaltoanalog converters (DACs) are described and compared. These
designs can easily be implemented in an integrated circuit: the conventional pulsewidth
modulation (PWM) DAC, the new pulsecount modulation (PCM) DAC and the firstorder noise
shaping (FONS) DAC. All three methods control the ratio of the sum of all pulse durations
to the constant total period. As the pulse durations are integral multiples of a unit
pulse, all three can be classified as pulsecount modulation methods. Block diagrams of
all three DACs consisting of a simple digital circuit and a lowpass filter are presented.
For a constant digital input value the worst case ripple of the filter output is used to
calculate the cutoff frequency of the lowpass filter. Approximations for the 3 dB cutoff
frequency of firstorder, secondorder and fourthorder Butterworth lowpass filters are
given. The dynamic properties are analyzed in the time domain (settling time) and in the
frequency domain (unfiltered output spectrum of a fullscale sine wave input). The main
influences on the static accuracy are analyzed. A case study demonstrates the abilities of
PCM and FONS.
Index Terms:
digitalanalogue conversion; pulse width modulation; pulse circuits; lowpass filters;
Butterworth filters; frequencydomain analysis; timedomain analysis; pulsecount
modulation; lowcost DAC; pulsewidth modulation; PWM; firstorder noise shaping; pulse
durations; digital circuit; lowpass filter; worst case ripple; cutoff frequency;
fourthorder Butterworth lowpass filters; secondorder Butterworth lowpass filters;
dynamic properties; time domain; settling time; frequency domain; unfiltered output
spectrum; fullscale sine wave input; static accuracy; PCM
PDF FULLTEXT
 Heiss, M.: Reinforcement learning or tracking of
inputoutput maps. in Applied Artificial Intelligence, 8(4):483496, 1994.
 Heiss, M.: Inverse passive learning of an inputoutput map
through updatesplinesmoothing. IEEE Transactions on Automatic Control, 39(2):259268,
February 1994.
Inverse passive learning of an inputoutputmap through updatesplinesmoothing
 Heiss, M.
Lab. for Inf. & Decision Syst., MIT, Cambridge, MA, USA
This paper appears in: Automatic Control, IEEE Transactions on
On page(s): 259  268
Feb. 1994
Volume: 39 Issue: 2
ISSN: 00189286
References Cited: 70
CODEN: IETAA9
INSPEC Accession Number: 4660567
Abstract:
This paper presents a robust method of learning passively a onedimensional
inputoutputmap when receiving only indirect information about the correct
inputoutputmap (e.g., only the sign of the deviation between the actual estimated output
value and the correct output value is obtained). This information is obtained for only one
inputoutput combination per updating cycle. The approach is to increment or decrement
step by step the output values of the actually stored map and then to apply global or
local cubic spline smoothing in order to avoid "adaptation holes" at points
which are never updated or less frequently updated than other points. This method works
with noisy measurements as well as slowly timevarying systems. Even discontinuous changes
of the desired inputoutputrelation do not result in instability. Problems of convergence
and stability are treated and design rules are given.
Index Terms:
learning by example; splines (mathematics); inverse passive learning; inputoutputmap;
updatesplinesmoothing; robust method; incrementation; decrementation; cubic spline
smoothing; adaptation holes; discontinuous changes; convergence; stability; I/O map
PDF FULLTEXT
 Heiss, M.: Errordetecting unitdistance code. IEEE
Transactions on Instrumentation and Measurements, 39(5):730734, 1990.
Errordetecting unitdistance code
 Heiss, M.
VoestAlpineAutomotive GmbH, Vienna, Austria
This paper appears in: Instrumentation and Measurement, IEEE Transactions on
On page(s): 730  734
Oct. 1990
Volume: 39 Issue: 5
ISSN: 00189456
References Cited: 15
CODEN: IEIMAO
INSPEC Accession Number: 3781063
Abstract:
A particular kind of unitdistance code is proposed. Unitdistance codes are applied
especially to absolute encoders (linear or rotary). The various positions of the encoder
are represented by the various code words of the code. In contrast to conventional
unitdistance codes, this code offers the possibility of detecting all singlebit errors
except the two singlebit errors that cause a code word which represents the adjacent
position. Unitdistance codes are characterized by a Hamming distance of 1. Thus, Hamming
distance is not a unit of measurement for the code's errordetection ability. Therefore, a
Hamming distance is defined especially for unitdistance codes excluding the adjacent code
words from the calculation of distances. An application shows the utility of the code in
the field of instrumentation and measurement.
Index Terms:
error detecting code; Gray code; path distance; flash A/D convertor; unitdistance code;
absolute encoders; singlebit errors; Hamming distance; instrumentation; measurement;
analoguedigital conversion; error detection codes
PDF FULLTEXT
 Heiss, M.: Kennfelder in der Regelungstechnik.
Automatisierungstechnik at, 43(8):363367, 1995.
 Leichtfried, J. and Heiss, M.: Ein kennfeldorientiertes
Konzept f�r FuzzyRegler. Automatisierungstechnik at, 43(1):3140, 1995.
 Heiss, M., Heiss, D., and Kampl, S.: Lernen linear
interpolierter Kennlinien. Automatisierungstechnik at, 42(11):497506, 1994.
 Heiss, M.: Pulsanzahlmodulator statt Pulsbreitenmodulator
zur Verbesserung der Reglerdynamik. Automatisierungstechnik at, 41(11):428432, 1993.
 Heiss, M.: Regressionsparabelfilter und Differenzierer.
Automatisierungstechnik at, 37(12):468470, 1989.
 Heiss, M.: Schnelle Berechnung der
"closedloop"�bertragungsfunktion. Automatisierungstechnik at, 36(12):487488,
1988.
 Heiss, M.: Symmetrische Komponenten bei elastischen
Begrenzungen. Automatisierungstechnik at, 35(8):334335, 1987.
 Hofbauer, A. and Heiss, M.: The origin of spikes during
online adaptation of membership functions. International Journal Automation Austria,
4(1):114, 1996.
 Leichtfried, J. and Heiss, M.: FuzzyRegler als gl�ttender
RegelInterpolator. International Journal Automation Austria, 3(2):4761, 1995.
 Heiss, M.: Basisfunktionsnetzwerke als Grundlage zur
Kennfeldinterpretation von neuronalen Netzen und FuzzyReglern. Elektrotechnik und
Informationstechnik e & i, 112(7/8):345353, 1995.
 Heiss, M.: Monotonieerhaltende Gl�ttungsverfahren und
Monotonisierung. Elektrotechnik und Informationstechnik e & i, 110(5):233238, 1993.
 Heiss, M.: L2optimale St�tzstellen f�r linear
interpolierte Kennlinien. Elektrotechnik und Informationstechnik e & i,
108(12):555557, 1991.
 Heiss, M.: Pseudologarithmisches Vergessen von
Abtastwerten. Elektrotechnik und Informationstechnik e & i, 108(4):149151, 1991.
 Heiss, M.: Adaption von Kennlinien in Echtzeit.
Elektrotechnik und Informationstechnik e & i, 106(10):398402, 1989.
 Heiss, M.: Optimal deadzone characteristic for minimizing
the aposteriori error in basis function networks (invited). In IEEE Conference on
Decision on Control (CDC 96), pp. 476477, Kobe, Japan, 1996.
Optimal deadzone characteristic for minimizing the aposterior error in basis function
networks
 Heiss, M.
Tech. Univ. Wien, Austria
This paper appears in: Decision and Control, 1996., Proceedings of the 35th IEEE
Conference on
On page(s): 476  477 vol.1
1113 Dec. 1996
1996
Volume: 1
ISBN: 0780335902
IEEE Catalog Number: 96CH35989
Number of Pages: 4 vol. 4858
References Cited: 5
INSPEC Accession Number: 5520093
Abstract:
The incorporation of deadzones in the error signal of basis function networks avoids the
networks' overtraining and guarantees the convergence of the normalized LMSalgorithm and
related algorithms. A new socalled errorminimizing deadzone is presented providing the
least aposteriori error out of the set of all convergence assuring deadzones.
Index Terms:
feedforward neural nets; optimal deadzone characteristic; aposteriori error
minimization; basis function networks; normalized LMSalgorithm convergence;
errorminimizing deadzone; least aposteriori error; convergenceassuring deadzones
PDF FULLTEXT
 Heiss, M.: DeadZone Adaptation vs. Overtraining Phenomenon
for Basis Function Networks (invited). In Proc. of IMACS Symposium on Mathematical
Modelling (MATHMOD'97), pp. 757761, Vienna, 1997.
 Hofbauer, A. and Heiss, M.: The origin of spikes during
online adaptation of membership functions. In Proc. Int. Symp. on Fuzzy Logic (ISFL'95),
pp. A1017, ETH Zurich, May 1995. Academic Press.
 Kampl, S. and Heiss, M.: Multiplicationfree radial basis
function network. In American Control Conference (ACC 95), pp. 37823785, Seattle, 1995.
IEEE.
 Heiss, M. and Leichtfried, J.: Selflearning fuzzy controller
with smooth transfer characteristic and guaranteed convergence. In IEEE Conference on
Control Applications (CCA 94), pp. 12511256, Glasgow, 1994.
Selflearning fuzzy controller with smooth transfer characteristic and guaranteed
convergence
 Heiss, M.; Leichtfried, J.
Inst. fur Allgemeine Elektrotech./Automobilelektronik, Wien Univ. of Technol., Austria
This paper appears in: Control Applications, 1994., Proceedings of the Third
IEEE Conference on
On page(s): 1251  1256 vol.2
2426 Aug. 1994
1994
ISBN: 0780318722
IEEE Catalog Number: 94CH34207
Number of Pages: 3 vol. xlii+1952
References Cited: 15
INSPEC Accession Number: 4886203
Abstract:
The paper presents a userfriendly way to design a smooth nonlinear control surface. The
method can be seen as a fuzzy control design tool, but it can also be seen in the context
of neural networks, Bspline basis functions, or simply as a tool for setting up an
inputoutput map. The design process is composed of two steps. First, an expert knowledge
is used in a rule based manner to set up the main structure of the control surface.
Second, an automatic learning algorithm is used to improve the control surface and to
compensate for slowly time varying effects, like the aging of the system. Applications are
mentioned and the convergence of the learning algorithm is proved under real world
conditions.
Index Terms:
fuzzy control; selfadjusting systems; nonlinear control systems; intelligent control;
neural nets; convergence; knowledge based systems; control system synthesis; learning
systems; self learning fuzzy controller; smooth transfer characteristic; convergence;
smooth nonlinear control; neural networks; Bspline basis functions; rule based system;
automatic learning algorithm
PDF FULLTEXT
 Heiss, M.: Methods of learning or tracking passively an
inputoutput map without neural networks. In Workshop notes: RealWorld Applications of
Machine Learning at the European Conference on Machine Learning, pp. 114, Wien, 1993.
 Heiss, M.: Inverse passive learning of an inputoutput map
through updatesplinesmoothing. In American Control Conference (ACC 92), pp. 23202326,
Chicago, 1992. IEEE.
 Heiss, M.: Sind Regeln eines FuzzyReglers nur St�tzstellen
im EinAusgangskennfeld? In Informationstagung Mikroelektronik ME'95, pp. 38, Wien, 1995.
�VE Schriftenreihe Nr. 8.
 Heiss, M.: Lernen von Kennfeldern als Maßnahme zur
Verringerung der Umweltbelastung. In Informationstagung Mikroelektronik ME'93, pp.
195201, Wien, 1993. �VE Schriftenreihe Nr. 5.
 Heiss, M.: Berechnung von LeiterbahnInduktivit�ten.
Elektronik, 37(18):103104, 1988.
 Heiss, M.: Instabile Betriebsarten der
Pulsbreitenmodulation. Elektronik, 38(20):6465, 1989.
 Heiss, M. and Dittrich, W. Pulsanzahlmodulator als
D/AUmsetzer. Elektronik, 38(19):9698, 1989.
 Augesky, Ch. and Heiss, M.: Elektronische
Gleichf�rmigkeitsregelung bei Dieselfahrzeugen. Elektronikschau, pp. 4243, Juli 1990.

Collaboration
maturity and the offshoring cost barrier: the tradeoff
between flexibility in team composition and crosssite
communication effort in geographically distributed
development projects
Lasser, S.
Heiss,
M.
Siemens AG, Austria
This paper appears in:
Professional Communication Conference, 2005.
IPCC 2005. Proceedings. International
Publication Date: 1013 July 2005
On page(s): 718  728
EISBN: 0780390288
Number of Pages: x+839
ISBN: 078039027X
INSPEC Accession Number:8680629
Digital Object Identifier: 10.1109/IPCC.2005.1494243
Posted online: 20050815 08:24:47.0




This paper analyzes how total project costs are split
into operative costs and distribution costs. This split
depends on the collaboration model being applied and the
level of maturity of the collaboration. The lower the
collaboration maturity, the more significant the
socalled offshoring cost barrier, i.e. the resources
and time needed to progress toward more costeffective
forms of collaboration. The lowcost offshoring problem
is stated as a cost optimization problem, with customer
requirements and internal requirements as boundary
conditions. The more degree of freedom is available for
project team composition, the better the result will be.
The project team composition is always a tradeoff
between including the best available experts from
different sites and the corresponding communication and
coordination problems. 
Practices and Supporting
Structures for Mature Inquiry Culture in Distributed Software
Development Projects
Mikulovic, V.
Heiss, M.
Herbsleb, J.D.
Program & Syst. Eng., Siemens AG Austria, Vienna
This paper appears in:
Global Software Engineering, 2006. ICGSE '06.
International Conference on
Publication Date: Oct. 2006
On page(s): 245  246
Number of Pages: 245  246
Location: Florianopolis
ISBN: 0769526632
Digital Object Identifier: 10.1109/ICGSE.2006.261242
Posted online: 20061219 09:37:20.0




As software specifications for complex systems are practically
never entirely complete and consistent, the recipient of the
specification needs domain knowledge in order to decide which
parts of the system are specified clearly and which parts are
specified ambiguously and thus need inquiry to get a more
detailed specification. By analyzing the evidence gained in
multiplecase study, the necessary components for achieving a
mature inquiry culture in distributed software development
derived from the practices at Siemens Program and System
Engineering (PSE) are identified. These components are presented
in three categoriespillars: project communication, requirements
communication and inquiry practices
The
bottomup/topdown pattern: an organizational pattern
for a balanced management system
Heiss,
M.
Stoeckl, S.
Hausknotz, C.
Dept. of Innovation & Technol. Manage., Siemens Program
& Syst. Eng. PSE, Vienna, Austria
This paper appears in:
Engineering Management Conference, 2004.
Proceedings. 2004 IEEE International
Publication Date: 1821 Oct. 2004
Volume: 1
On page(s): 317  323 Vol.1
Number of Pages: 3 vol. (xix+1352)
ISSN:
ISBN: 0780385195
INSPEC Accession Number:8331184
Digital Object Identifier: 10.1109/IEMC.2004.1407127
Posted online: 20050404 09:25:41.0




A balanced management system requires a balanced
composition of bottomup and topdown activities. The
larger the company/division/organization is, the less
manageable and the less effective is an unstructured
interaction of bottomup and topdown activities. The
bottomup/topdown interaction usually is hierarchically
structured, lacking in interaction between different
business units and therefore lacking in synergy. The
paper proposes the concept of management components as a
topicbased structure of bottomup/topdown interaction.
The concept is presented in form of an organizational
pattern. The pattern is called bottomup/topdown
pattern. Further on in this paper, a short case study
shows how this concept is implemented at Siemens program
and system engineering PSE, a 5000 engineers R&D
division of the Siemens AG.
Distributed
facetoface communication in bottomup driven
technology managementa model for optimizing
communication topologies
Kubasa, G.
Heiss,
M.
Siemens AG Austria, Vienna, Austria
This paper appears in:
Engineering Management Conference, 2002.
IEMC '02. 2002 IEEE International
Publication Date: 2002
Volume: 1
On page(s): 234  238 vol.1
Number of Pages: 2 vol.xx+935
ISSN:
ISBN: 0780373855
INSPEC Accession Number:7529208
Digital Object Identifier:
10.1109/IEMC.2002.1038426
Posted online: 20030709 09:46:56.0




Understanding the communication process in
software development organizations has been
recognized as the key element to improve the
development performance. The paper analyses how
different communication topologieswho talks
directly to whom and who is informed only via
third personsresult in different communication
costs and different communication efficiencies.
The paper presents a simple cost and efficiency
model for facetoface communication. Based on
this model and based on a desired communication
benchmark, we are able to compute the optimal
communication topology by minimizing the
cost/efficiency ratio. The goal is to apply the
model for optimizing the knowledge transfer
within the knowledge networks at Siemens PSE.
These geographically distributed knowledge
networks are the key ingredients representing
the bottom up component of the technology
management process at Siemens PSE.
The
technology tree conceptan evolutionary
approach to technology management in a
rapidly changing market
Heiss,
M.
Jankowsky, J.
This paper appears in:
Change Management and the New
Industrial Revolution, 2001. IEMC '01
Proceedings.
Publication Date: 79 Oct. 2001
On page(s): 37  43
Number of Pages: x+461
Meeting Date: 10/07/2001  10/09/2001
Location: Albany, NY
ISBN: 0780372603
INSPEC Accession Number:7176214
Digital Object Identifier:
10.1109/IEMC.2001.960477
Posted online: 20020807 00:34:11.0




Technological knowledge is changing very
rapidly, especially in the information
and communication business. Due to the
natural limited knowledge processing
capacity of management, classical
topdown driven technology management
approaches have reached their limits.
This paper presents a bottomup driven
technology management concept, the
socalled technology tree concept, which
is capable of integrating the capacity
of all engineers into its technology
management process. It represents an
open process allowing all engineers the
possibility of initiating their own
knowledge network for a promising
technology of their own choice. Driven
by the evolutionary processsurvival of
the fittestsome of these technology
networks develop into strong business
branches, others die off and many remain
on a readytostartlevel without
stranded investments up to that point.
As soon as the market is ready, the
corresponding network is able to quickly
expand and apply its technology. Further
on in this paper, an example shows how
the concept is implemented at an R&D
division of Siemens AG Austria, having
5000 engineers. The technology tree
concept is a driving force for cultural
changes towards knowledge sharing,
cooperating across the borders of
business units, and opening topdown
bottlenecks 



and others 