The Universal Approximation Property Characterization, Construction, Representation, and Existence

The universal approximation property of various machine learning models is currently only understood on a case-by-case basis, limiting the rapid development of new theoretically justified neural network architectures and blurring our understanding of our current models’ potential. This paper works t...

Full description

Saved in:
Bibliographic Details
Published inAnnals of mathematics and artificial intelligence Vol. 89; no. 5-6; pp. 435 - 469
Main Author Kratsios, Anastasis
Format Journal Article
LanguageEnglish
Published Cham Springer International Publishing 01.06.2021
Springer
Subjects
Online AccessGet full text
ISSN1012-2443
1573-7470
DOI10.1007/s10472-020-09723-1

Cover

Abstract The universal approximation property of various machine learning models is currently only understood on a case-by-case basis, limiting the rapid development of new theoretically justified neural network architectures and blurring our understanding of our current models’ potential. This paper works towards overcoming these challenges by presenting a characterization, a representation, a construction method, and an existence result, each of which applies to any universal approximator on most function spaces of practical interest. Our characterization result is used to describe which activation functions allow the feed-forward architecture to maintain its universal approximation capabilities when multiple constraints are imposed on its final layers and its remaining layers are only sparsely connected. These include a rescaled and shifted Leaky ReLU activation function but not the ReLU activation function. Our construction and representation result is used to exhibit a simple modification of the feed-forward architecture, which can approximate any continuous function with non-pathological growth, uniformly on the entire Euclidean input space. This improves the known capabilities of the feed-forward architecture.
AbstractList The universal approximation property of various machine learning models is currently only understood on a case-by-case basis, limiting the rapid development of new theoretically justified neural network architectures and blurring our understanding of our current models’ potential. This paper works towards overcoming these challenges by presenting a characterization, a representation, a construction method, and an existence result, each of which applies to any universal approximator on most function spaces of practical interest. Our characterization result is used to describe which activation functions allow the feed-forward architecture to maintain its universal approximation capabilities when multiple constraints are imposed on its final layers and its remaining layers are only sparsely connected. These include a rescaled and shifted Leaky ReLU activation function but not the ReLU activation function. Our construction and representation result is used to exhibit a simple modification of the feed-forward architecture, which can approximate any continuous function with non-pathological growth, uniformly on the entire Euclidean input space. This improves the known capabilities of the feed-forward architecture.
The universal approximation property of various machine learning models is currently only understood on a case-by-case basis, limiting the rapid development of new theoretically justified neural network architectures and blurring our understanding of our current models' potential. This paper works towards overcoming these challenges by presenting a characterization, a representation, a construction method, and an existence result, each of which applies to any universal approximator on most function spaces of practical interest. Our characterization result is used to describe which activation functions allow the feed-forward architecture to maintain its universal approximation capabilities when multiple constraints are imposed on its final layers and its remaining layers are only sparsely connected. These include a rescaled and shifted Leaky ReLU activation function but not the ReLU activation function. Our construction and representation result is used to exhibit a simple modification of the feed-forward architecture, which can approximate any continuous function with non-pathological growth, uniformly on the entire Euclidean input space. This improves the known capabilities of the feed-forward architecture. Keywords Universal approximation * Constrained approximation * Uniform approximation * Deep learning * Topological transitivity * Composition operators Mathematics Subject Classification (2010) 68T07 47B33 * 47A16 * 68T05 * 30L05 * 46M40
Audience Academic
Author Kratsios, Anastasis
Author_xml – sequence: 1
  givenname: Anastasis
  orcidid: 0000-0001-6791-3371
  surname: Kratsios
  fullname: Kratsios, Anastasis
  email: anastasis.kratsios@math.ethz.ch
  organization: (ETH) Eidgenössische Technische Hochschule Zürich
BookMark eNp9kDtPwzAQgC1UJNrCH2DKwOpythM7HquKl1QJhna2HMcurtIksgOi_x63YWKobrjTSd89vhmatF1rEbonsCAA4jESyAXFQAGDFJRhcoWmpBAMi1zAJNVAKKZ5zm7QLMY9AEhe8il62HzabNv6bxuibrJl34fuxx_04Ls2-whdb8NwvEXXTjfR3v3lOdo-P21Wr3j9_vK2Wq6xYQwGzFlueGFKZrQtNbWyErSuAbgBSSW1WoIuStBVQV0hdZk7VzkCpja1lLwq2Rwtxrk73VjlW9cNQZsUtT14kz52PvWXgghe5LIgCaAjYEIXY7BO9SEdH46KgDqJUaMYlcSosxh1gsp_kPHD-eG0zTeXUTaiMe1pdzaoffcV2uTkEvULIRJ5Dg
CitedBy_id crossref_primary_10_1007_s00500_024_09918_2
crossref_primary_10_3390_axioms11020080
crossref_primary_10_1016_j_neucom_2022_01_036
crossref_primary_10_1007_s41870_023_01432_2
crossref_primary_10_1016_j_jtbi_2024_111987
crossref_primary_10_1061_JAEEEZ_ASENG_4884
crossref_primary_10_1007_s00780_024_00549_x
crossref_primary_10_1063_5_0123466
crossref_primary_10_1007_s00366_023_01875_y
crossref_primary_10_1080_14697688_2024_2307493
crossref_primary_10_1016_j_asr_2021_10_051
crossref_primary_10_1109_JOE_2022_3173454
crossref_primary_10_1007_s41605_022_00368_8
crossref_primary_10_1051_epjconf_202225904003
crossref_primary_10_1007_s41365_023_01297_x
crossref_primary_10_1177_01492063231194968
crossref_primary_10_1038_s41467_022_30545_8
crossref_primary_10_1259_dmfr_20210437
crossref_primary_10_1063_5_0134984
crossref_primary_10_2139_ssrn_4657010
crossref_primary_10_1007_s10472_022_09824_z
crossref_primary_10_1016_j_amc_2023_128230
crossref_primary_10_1103_PhysRevD_106_114510
crossref_primary_10_1080_0952813X_2023_2270995
crossref_primary_10_1016_j_jat_2022_105716
crossref_primary_10_1016_j_ymssp_2022_109328
crossref_primary_10_1109_TIP_2022_3176220
crossref_primary_10_1016_j_commatsci_2022_111984
crossref_primary_10_2139_ssrn_3902402
crossref_primary_10_1016_j_combustflame_2023_112781
crossref_primary_10_1103_PhysRevD_105_014017
crossref_primary_10_1109_TCCN_2022_3177516
crossref_primary_10_1007_s00158_021_03005_y
crossref_primary_10_1137_21M1418794
crossref_primary_10_1109_TPAMI_2023_3341688
Cites_doi 10.3390/risks8020040
10.5802/aif.8
10.1080/14697688.2019.1571683
10.5486/PMD.2012.4814
10.1016/0893-6080(90)90005-6
10.1016/0893-6080(91)90009-T
10.1016/0893-6080(89)90020-8
10.1038/s41576-019-0122-6
10.4064/sm159-1-6
10.1515/crelle-2018-0037
10.1007/978-1-4471-2170-1
10.4064/sm180109-8-6
10.1017/S0962492900002919
10.1016/S0893-6080(05)80131-5
10.2140/pjm.1956.6.397
10.33232/BIMS.0075.45.57
10.1037/0033-295X.99.3.480
10.1145/168304.168357
10.1007/978-3-322-90559-8
10.1093/imrn/rnq127
10.1090/S0002-9904-1966-11541-0
10.1109/TNN.2008.2005605
10.1007/s11263-012-0591-y
10.3390/math7100992
10.1007/BF02551274
10.1016/j.neuroimage.2008.10.052
10.1007/BF01898849
10.1145/3374217
10.1037/h0042519
10.1007/s10208-020-09461-0
10.3390/risks8040101
10.1007/BF02478259
10.4064/sm-69-1-33-40
10.3389/fnins.2014.00229
10.1016/j.jmaa.2016.08.029
10.1016/j.jmaa.2018.04.063
10.1007/s00365-021-09543-4
10.1142/9911
10.1007/978-3-319-02045-7
10.1038/d41586-018-02174-z
10.1007/978-3-642-21298-7
10.1109/18.256500
10.1073/pnas.17.5.315
10.1007/BF02414945
ContentType Journal Article
Copyright The Author(s) 2021
COPYRIGHT 2021 Springer
Copyright_xml – notice: The Author(s) 2021
– notice: COPYRIGHT 2021 Springer
DBID C6C
AAYXX
CITATION
DOI 10.1007/s10472-020-09723-1
DatabaseName Springer Nature OA Free Journals
CrossRef
DatabaseTitle CrossRef
DatabaseTitleList

CrossRef
Database_xml – sequence: 1
  dbid: C6C
  name: Springer Nature OA Free Journals
  url: http://www.springeropen.com/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
Discipline Mathematics
Computer Science
EISSN 1573-7470
EndPage 469
ExternalDocumentID A717654951
10_1007_s10472_020_09723_1
GrantInformation_xml – fundername: ETH Zürich Foundation
  funderid: https://doi.org/10.13039/501100012652
GroupedDBID -4Z
-59
-5G
-BR
-EM
-Y2
-~C
.86
.DC
.VR
06D
0R~
0VY
1N0
1SB
2.D
203
23M
28-
2J2
2JN
2JY
2KG
2LR
2P1
2VQ
2~H
30V
4.4
406
408
409
40D
40E
5GY
5QI
5VS
67Z
6NX
8TC
8UJ
95-
95.
95~
96X
AAAVM
AABHQ
AACDK
AAHNG
AAIAL
AAJBT
AAJKR
AANZL
AAOBN
AARHV
AARTL
AASML
AATNV
AATVU
AAUYE
AAWCG
AAYIU
AAYQN
AAYTO
AAYZH
ABAKF
ABBBX
ABBXA
ABDZT
ABECU
ABFTD
ABFTV
ABHLI
ABHQN
ABJCF
ABJNI
ABJOX
ABKCH
ABKTR
ABMNI
ABMQK
ABNWP
ABQBU
ABQSL
ABSXP
ABTEG
ABTHY
ABTKH
ABTMW
ABULA
ABWNU
ABXPI
ACAOD
ACBXY
ACDTI
ACGFS
ACHSB
ACHXU
ACIWK
ACKNC
ACMDZ
ACMLO
ACOKC
ACOMO
ACPIV
ACSNA
ACZOJ
ADHHG
ADHIR
ADINQ
ADKNI
ADKPE
ADRFC
ADTPH
ADURQ
ADYFF
ADZKW
AEBTG
AEFIE
AEFQL
AEGAL
AEGNC
AEJHL
AEJRE
AEKMD
AEMSY
AENEX
AEOHA
AEPYU
AESKC
AETLH
AEVLU
AEXYK
AFBBN
AFEXP
AFGCZ
AFKRA
AFLOW
AFQWF
AFWTZ
AFZKB
AGAYW
AGDGC
AGGDS
AGJBK
AGMZJ
AGQEE
AGQMX
AGRTI
AGWIL
AGWZB
AGYKE
AHAVH
AHBYD
AHKAY
AHSBF
AHYZX
AIAKS
AIGIU
AIIXL
AILAN
AITGF
AJBLW
AJRNO
AJZVZ
ALMA_UNASSIGNED_HOLDINGS
ALWAN
AMKLP
AMXSW
AMYLF
AMYQR
AOCGG
ARAPS
ARMRJ
ASPBG
AVWKF
AXYYD
AYJHY
AZFZN
B-.
BA0
BBWZM
BDATZ
BENPR
BGLVJ
BGNMA
BSONS
C6C
CAG
CCPQU
COF
CS3
CSCUP
DDRTE
DL5
DNIVK
DPUIP
EBLON
EBS
EIOEI
EJD
ESBYG
F5P
FEDTE
FERAY
FFXSO
FIGPU
FINBP
FNLPD
FRRFC
FSGXE
FWDCC
GGCAI
GGRSB
GJIRD
GNWQR
GQ6
GQ7
GQ8
GXS
HCIFZ
HF~
HG5
HG6
HMJXF
HQYDN
HRMNR
HVGLF
HZ~
I09
IAO
IHE
IJ-
IKXTQ
ITM
IWAJR
IXC
IZIGR
IZQ
I~X
I~Z
J-C
J0Z
JBSCW
JCJTX
JZLTJ
K7-
KDC
KOV
KOW
LAK
LLZTM
M4Y
M7S
MA-
N2Q
NB0
NDZJH
NPVJJ
NQJWS
NU0
O9-
O93
O9G
O9I
O9J
OAM
OVD
P19
P2P
P9O
PF0
PT4
PT5
PTHSS
QOK
QOS
R4E
R89
R9I
RHV
RNI
RNS
ROL
RPX
RSV
RZC
RZE
RZK
S16
S1Z
S26
S27
S28
S3B
SAP
SCJ
SCLPG
SCO
SDH
SDM
SHX
SISQX
SJYHP
SNE
SNPRN
SNX
SOHCF
SOJ
SPISZ
SRMVM
SSLCW
STPWE
SZN
T13
T16
TEORI
TN5
TSG
TSK
TSV
TUC
U2A
UG4
UOJIU
UTJUX
UZXMN
VC2
VFIZW
W23
W48
WK8
YLTOR
Z45
Z7R
Z7X
Z81
Z83
Z88
Z92
ZMTXR
~A9
~EX
AAPKM
AAYXX
ABBRH
ABDBE
ABFSG
ABRTQ
ACSTC
ADHKG
ADKFA
AEZWR
AFDZB
AFHIU
AFOHR
AGQPQ
AHPBZ
AHWEU
AIXLP
ATHPR
AYFIA
CITATION
PHGZM
PHGZT
PQGLB
PUEGO
ID FETCH-LOGICAL-c330t-634c65c83cae8a2e9b72dd006c09292ea90a580ab52f59a84ffbf10cdcd996b83
IEDL.DBID AGYKE
ISSN 1012-2443
IngestDate Mon Oct 20 16:20:48 EDT 2025
Thu Apr 24 23:08:18 EDT 2025
Wed Oct 01 03:59:00 EDT 2025
Fri Feb 21 02:49:12 EST 2025
IsDoiOpenAccess true
IsOpenAccess true
IsPeerReviewed true
IsScholarly true
Issue 5-6
Keywords Deep learning
30L05
46M40
68T05
47A16
Universal approximation
Uniform approximation
Composition operators
Constrained approximation
68T07 47B33
Topological transitivity
Language English
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-c330t-634c65c83cae8a2e9b72dd006c09292ea90a580ab52f59a84ffbf10cdcd996b83
ORCID 0000-0001-6791-3371
OpenAccessLink https://proxy.k.utb.cz/login?url=https://link.springer.com/10.1007/s10472-020-09723-1
PageCount 35
ParticipantIDs gale_infotracacademiconefile_A717654951
crossref_primary_10_1007_s10472_020_09723_1
crossref_citationtrail_10_1007_s10472_020_09723_1
springer_journals_10_1007_s10472_020_09723_1
ProviderPackageCode CITATION
AAYXX
PublicationCentury 2000
PublicationDate 20210600
2021-06-00
20210601
PublicationDateYYYYMMDD 2021-06-01
PublicationDate_xml – month: 6
  year: 2021
  text: 20210600
PublicationDecade 2020
PublicationPlace Cham
PublicationPlace_xml – name: Cham
PublicationTitle Annals of mathematics and artificial intelligence
PublicationTitleAbbrev Ann Math Artif Intell
PublicationYear 2021
Publisher Springer International Publishing
Springer
Publisher_xml – name: Springer International Publishing
– name: Springer
References Munkres (CR37) 2000
Osborne (CR45) 2014
Hornik (CR5) 1991; 4
Gelfand (CR49) 1941; 9
Cybenko (CR4) 1989; 2
Grigoryeva, Ortega (CR41) 2019; 20
Dieudonné, Schwartz (CR84) 1949; 1
CR35
Bayart, Darji, Pires (CR80) 2018; 465
Kratsios, Hyndman (CR14) 2020; 8
CR32
CR76
CR31
CR75
CR30
Scarselli, Gori, Tsoi, Hagenbuchner, Monfardini (CR59) 2009; 20
CR74
Hornik, Stinchcombe, White (CR3) 1990; 3
Fletcher (CR53) 2013; 105
Fletcher (CR43) 2013; 105
Ambrosio, Puglisi (CR28) 2020; 764
Phelps (CR72) 1957; 8
von Luxburg, Bousquet (CR27) 2003; 5
Prolla (CR70) 1971; 89
Bourbaki (CR77) 1981
Horvath, Muguruza, Tomas (CR15) 2020; 0
CR7
Maas, Hannun, Ng (CR63) 2013; 30
CR48
Isbell (CR50) 1966; 72
Fletcher, Venkatasubramanian, Joshi (CR22) 2009; 45
CR47
CR46
CR44
Buehler, Gonon, Teichmann, Wood (CR11) 2019; 19
CR42
CR83
Godefroy (CR33) 2015; 55
CR81
Behrends, Schmidt-Bichler (CR82) 1980; 69
Bourbaki (CR71) 1971
Rosenblatt (CR2) 1958; 65
Arens, Eells (CR26) 1956; 6
Jost (CR34) 2011
Becker, Cheridito, Jentzen (CR12) 2019; 20
Meyer, Bonnabel, Sepulchre (CR54) 2011; 12
McCulloch, Pitts (CR1) 1943; 5
Plis (CR9) 2014; 8
CR19
Przestacki (CR79) 2017; 445
CR17
Nagata (CR36) 1974
Hornik, Stinchcombe, White (CR18) 1989; 2
CR58
CR57
CR55
Hummel, Biederman (CR56) 1992; 99
CR10
CR52
Cuchiero, Khosrawi, Teichmann (CR13) 2020; 8
Leshno, Lin, Pinkus, Schocken (CR16) 1993; 6
Caponnetto, Micchelli, Pontil, Ying (CR40) 2008; 9
Eraslan, Avsec, Gagneur, Theis (CR8) 2019; 20
CR29
Keller-Ressel, Nargang (CR23) 2020; 8
Kontorovich, Nadler (CR39) 2009; 10
Micchelli, Xu, Zhang (CR38) 2006; 7
CR25
CR69
CR24
CR67
CR66
CR21
CR65
CR20
Koopman (CR62) 1931; 17
Singh, Manhas (CR64) 1993
Kolmogorov (CR6) 1957; 114
Dimov (CR51) 2012; 80
CR61
CR60
Kalmes (CR78) 2019; 249
Barron (CR68) 1993; 39
Kadec (CR73) 1967; 1
RK Singh (9723_CR64) 1993
PT Fletcher (9723_CR22) 2009; 45
L Ambrosio (9723_CR28) 2020; 764
E Behrends (9723_CR82) 1980; 69
9723_CR35
T Kalmes (9723_CR78) 2019; 249
I Gelfand (9723_CR49) 1941; 9
GD Dimov (9723_CR51) 2012; 80
L Kontorovich (9723_CR39) 2009; 10
G Meyer (9723_CR54) 2011; 12
A Caponnetto (9723_CR40) 2008; 9
AR Barron (9723_CR68) 1993; 39
J Dieudonné (9723_CR84) 1949; 1
N Bourbaki (9723_CR77) 1981
9723_CR32
9723_CR76
9723_CR30
9723_CR74
9723_CR31
9723_CR75
9723_CR29
RR Phelps (9723_CR72) 1957; 8
M Leshno (9723_CR16) 1993; 6
9723_CR25
G Godefroy (9723_CR33) 2015; 55
9723_CR69
JR Munkres (9723_CR37) 2000
PT Fletcher (9723_CR43) 2013; 105
9723_CR67
9723_CR24
A Przestacki (9723_CR79) 2017; 445
AN Kolmogorov (9723_CR6) 1957; 114
MI Kadec (9723_CR73) 1967; 1
PT Fletcher (9723_CR53) 2013; 105
9723_CR7
B Horvath (9723_CR15) 2020; 0
K Hornik (9723_CR5) 1991; 4
9723_CR21
9723_CR65
F Rosenblatt (9723_CR2) 1958; 65
MS Osborne (9723_CR45) 2014
9723_CR66
9723_CR20
9723_CR61
9723_CR60
9723_CR19
JR Isbell (9723_CR50) 1966; 72
9723_CR17
9723_CR58
A Kratsios (9723_CR14) 2020; 8
9723_CR57
J Jost (9723_CR34) 2011
N Bourbaki (9723_CR71) 1971
C Cuchiero (9723_CR13) 2020; 8
J- Nagata (9723_CR36) 1974
F Bayart (9723_CR80) 2018; 465
H Buehler (9723_CR11) 2019; 19
U von Luxburg (9723_CR27) 2003; 5
F Scarselli (9723_CR59) 2009; 20
JB Prolla (9723_CR70) 1971; 89
WS McCulloch (9723_CR1) 1943; 5
L Grigoryeva (9723_CR41) 2019; 20
9723_CR10
9723_CR55
S Becker (9723_CR12) 2019; 20
9723_CR52
G Cybenko (9723_CR4) 1989; 2
M Keller-Ressel (9723_CR23) 2020; 8
K Hornik (9723_CR18) 1989; 2
K Hornik (9723_CR3) 1990; 3
9723_CR47
9723_CR48
J Hummel (9723_CR56) 1992; 99
9723_CR46
SM Plis (9723_CR9) 2014; 8
G Eraslan (9723_CR8) 2019; 20
AL Maas (9723_CR63) 2013; 30
RF Arens (9723_CR26) 1956; 6
CA Micchelli (9723_CR38) 2006; 7
9723_CR44
9723_CR42
9723_CR83
BO Koopman (9723_CR62) 1931; 17
9723_CR81
References_xml – year: 1974
  ident: CR36
  publication-title: Modern general topology
– volume: 8
  start-page: 101
  issue: 4
  year: 2020
  ident: CR13
  article-title: A generative adversarial network approach to calibration of local stochastic volatility models
  publication-title: Risks
– ident: CR74
– volume: 12
  start-page: 593
  year: 2011
  end-page: 625
  ident: CR54
  article-title: Regression on fixed-rank positive semidefinite matrices: a Riemannian approach
  publication-title: J. Mach. Learn. Res.
– volume: 30
  start-page: 3
  issue: 1
  year: 2013
  ident: CR63
  article-title: Rectifier nonlinearities improve neural network acoustic models
  publication-title: ICML
– volume: 72
  start-page: 619
  year: 1966
  end-page: 655
  ident: CR50
  article-title: Structure of categories
  publication-title: Bull. Amer. Math. Soc.
– year: 1981
  ident: CR77
  publication-title: Espaces vectoriels topologiques. Chapitres 1 à 5
– volume: 9
  start-page: 1615
  year: 2008
  end-page: 1646
  ident: CR40
  article-title: Universal multi-task kernels
  publication-title: J. Mach. Learn. Res.
– volume: 17
  start-page: 315
  issue: 5
  year: 1931
  end-page: 318
  ident: CR62
  article-title: Hamiltonian systems and transformation in hilbert space
  publication-title: Proc. Natl. Acad. Sci.
– volume: 20
  start-page: 61
  issue: 1
  year: 2009
  end-page: 80
  ident: CR59
  article-title: The graph neural network model
  publication-title: Neural Netw. Learn Syst.
– volume: 1
  start-page: 61
  year: 1967
  end-page: 70
  ident: CR73
  article-title: A proof of the topological equivalence of all separable infinite-dimensional Banach spaces
  publication-title: Funkcional. Anal. i Priložen.
– ident: CR35
– volume: 99
  start-page: 480
  year: 1992
  end-page: 517
  ident: CR56
  article-title: Dynamic binding in a neural network for shape recognition
  publication-title: Psych. Rev.
– volume: 69
  start-page: 33
  issue: 1
  year: 1980
  end-page: 40
  ident: CR82
  article-title: M-structure and the Banach-Stone theorem
  publication-title: Studia Math.
– ident: CR29
– ident: CR61
– volume: 465
  start-page: 125
  issue: 1
  year: 2018
  end-page: 139
  ident: CR80
  article-title: Topological transitivity and mixing of composition operators
  publication-title: J. Math. Anal. Appl.
– volume: 20
  start-page: Paper No. 74, 25
  year: 2019
  ident: CR12
  article-title: Deep optimal stopping
  publication-title: J. Mach. Learn. Res.
– volume: 105
  start-page: 171
  issue: 2
  year: 2013
  end-page: 185
  ident: CR53
  article-title: Geodesic regression and the theory of least squares on Riemannian manifolds
  publication-title: Int. J. Comput. Vis.
– ident: CR58
– volume: 105
  start-page: 171
  issue: 2
  year: 2013
  end-page: 185
  ident: CR43
  article-title: Geodesic regression and the theory of least squares on Riemannian manifolds
  publication-title: Int. J. Comput. Vis.
– ident: CR25
– ident: CR42
– volume: 8
  start-page: cnaa002, 18
  issue: 1
  year: 2020
  ident: CR23
  article-title: Hydra: a method for strain-minimizing hyperbolic embedding of network- and distance-based data
  publication-title: J. Complex Netw.
– volume: 4
  start-page: 251
  issue: 2
  year: 1991
  end-page: 257
  ident: CR5
  article-title: Approximation capabilities of multilayer feedforward networks
  publication-title: Neural Netw.
– ident: CR21
– ident: CR46
– ident: CR19
– volume: 80
  start-page: 255
  issue: 3-4
  year: 2012
  end-page: 293
  ident: CR51
  article-title: Some generalizations of the Stone duality theorem
  publication-title: Publ. Math. Debrecen
– ident: CR67
– ident: CR75
– volume: 445
  start-page: 1097
  issue: 1
  year: 2017
  end-page: 1113
  ident: CR79
  article-title: Dynamical properties of weighted composition operators on the space of smooth functions
  publication-title: J. Math. Anal. Appl.
– volume: 8
  start-page: 40
  issue: 2
  year: 2020
  ident: CR14
  article-title: Deep arbitrage-free learning in a generalized HJM framework via arbitrage-regularization
  publication-title: Risks
– ident: CR57
– volume: 5
  start-page: 115
  year: 1943
  end-page: 133
  ident: CR1
  article-title: A logical calculus of the ideas immanent in nervous activity
  publication-title: Bull. Math. Biophys.
– ident: CR32
– ident: CR60
– ident: CR81
– volume: 7
  start-page: 2651
  year: 2006
  end-page: 2667
  ident: CR38
  article-title: Universal kernels
  publication-title: J. Mach. Learn. Res.
– volume: 249
  start-page: 259
  issue: 3
  year: 2019
  end-page: 301
  ident: CR78
  article-title: Dynamics of weighted composition operators on function spaces defined by local properties
  publication-title: Studia Math.
– volume: 65
  start-page: 386
  issue: 6
  year: 1958
  ident: CR2
  article-title: The perceptron: a probabilistic model for information storage and organization in the brain.
  publication-title: Psych. Rev.
– volume: 89
  start-page: 145
  year: 1971
  end-page: 157
  ident: CR70
  article-title: Weighted spaces of vector-valued continuous functions
  publication-title: Ann. Mat. Pura Appl. (4)
– volume: 55
  start-page: 89
  issue: 2
  year: 2015
  end-page: 118
  ident: CR33
  article-title: A survey on Lipschitz-free Banach spaces
  publication-title: Comment. Math.
– year: 2011
  ident: CR34
  publication-title: Riemannian Geometry and Geometric Analysis
– volume: 114
  start-page: 953
  year: 1957
  end-page: 956
  ident: CR6
  article-title: On the representation of continuous functions of many variables by superposition of continuous functions of one variable and addition
  publication-title: Dokl. Akad. Nauk SSSR
– ident: CR66
– ident: CR47
– volume: 9
  start-page: 3
  issue: 51
  year: 1941
  end-page: 24
  ident: CR49
  article-title: Normierte Ringe
  publication-title: Rec. Math. N. S.
– year: 1993
  ident: CR64
  publication-title: Composition operators on function spaces, North-Holland Mathematics Studies, vol. 179
– year: 2014
  ident: CR45
  publication-title: Locally convex spaces, Graduate Texts in Mathematics, vol. 269
– ident: CR30
– volume: 20
  start-page: 389
  issue: 7
  year: 2019
  end-page: 403
  ident: CR8
  article-title: Deep learning: new computational modelling techniques for genomics
  publication-title: Nat. Rev. Genet.
– volume: 45
  start-page: S143
  issue: 1
  year: 2009
  end-page: S152
  ident: CR22
  article-title: The geometric median on riemannian manifolds with application to robust atlas estimation
  publication-title: Neuroimage
– ident: CR10
– volume: 2
  start-page: 359
  issue: 5
  year: 1989
  end-page: 366
  ident: CR18
  article-title: Multilayer feedforward networks are universal approximators
  publication-title: Neural Netw.
– volume: 6
  start-page: 397
  year: 1956
  end-page: 403
  ident: CR26
  article-title: On embedding uniform and topological spaces
  publication-title: Pacific J. Math.
– volume: 3
  start-page: 551
  issue: 5
  year: 1990
  end-page: 560
  ident: CR3
  article-title: Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks
  publication-title: Neural Netw.
– volume: 20
  start-page: Paper No. 179, 62
  year: 2019
  ident: CR41
  article-title: Differentiable reservoir computing
  publication-title: J. Mach. Learn. Res.
– year: 1971
  ident: CR71
  publication-title: Éléments de mathématique. Topologie générale. Chapitres 1 à 4
– ident: CR69
– ident: CR44
– ident: CR48
– ident: CR65
– volume: 2
  start-page: 303
  issue: 4
  year: 1989
  end-page: 314
  ident: CR4
  article-title: Approximation by superpositions of a sigmoidal function
  publication-title: Math. Control Signals Syst.
– volume: 5
  start-page: 669
  year: 2003
  end-page: 695
  ident: CR27
  article-title: Distance-based classification with Lipschitz functions
  publication-title: J. Mach. Learn. Res.
– ident: CR52
– ident: CR17
– year: 2000
  ident: CR37
  publication-title: Topology
– volume: 6
  start-page: 861
  issue: 6
  year: 1993
  end-page: 867
  ident: CR16
  article-title: Multilayer feedforward networks with a nonpolynomial activation function can approximate any function
  publication-title: Neural Netw.
– ident: CR31
– volume: 764
  start-page: 1
  year: 2020
  end-page: 21
  ident: CR28
  article-title: Linear extension operators between spaces of Lipschitz maps and optimal transport
  publication-title: J. Reine Angew. Math.
– volume: 1
  start-page: 61
  year: 1949
  end-page: 101
  ident: CR84
  article-title: La dualité dans les espaces F et LF
  publication-title: Ann. Inst. Fourier (Grenoble)
– volume: 39
  start-page: 930
  issue: 3
  year: 1993
  end-page: 945
  ident: CR68
  article-title: Universal approximation bounds for superpositions of a sigmoidal function
  publication-title: IEEE Trans. Inform. Theory
– volume: 8
  start-page: 229
  year: 2014
  ident: CR9
  article-title: Deep learning for neuroimaging: a validation study
  publication-title: Front. Neurosci.
– volume: 10
  start-page: 1095
  year: 2009
  end-page: 1129
  ident: CR39
  article-title: Universal kernel-based learning with applications to regular languages
  publication-title: J. Mach. Learn. Res.
– volume: 19
  start-page: 1271
  issue: 8
  year: 2019
  end-page: 1291
  ident: CR11
  article-title: Deep hedging
  publication-title: Quant. Finance
– ident: CR55
– ident: CR7
– ident: CR76
– ident: CR83
– volume: 8
  start-page: 444
  year: 1957
  end-page: 450
  ident: CR72
  article-title: Subreflexive normed linear spaces
  publication-title: Arch. Math. (Basel)
– ident: CR24
– volume: 0
  start-page: 1
  issue: 0
  year: 2020
  end-page: 17
  ident: CR15
  article-title: Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models
  publication-title: Quant. Finance
– ident: CR20
– volume: 8
  start-page: 40
  issue: 2
  year: 2020
  ident: 9723_CR14
  publication-title: Risks
  doi: 10.3390/risks8020040
– volume: 0
  start-page: 1
  issue: 0
  year: 2020
  ident: 9723_CR15
  publication-title: Quant. Finance
– volume: 1
  start-page: 61
  year: 1949
  ident: 9723_CR84
  publication-title: Ann. Inst. Fourier (Grenoble)
  doi: 10.5802/aif.8
– volume: 19
  start-page: 1271
  issue: 8
  year: 2019
  ident: 9723_CR11
  publication-title: Quant. Finance
  doi: 10.1080/14697688.2019.1571683
– volume: 5
  start-page: 669
  year: 2003
  ident: 9723_CR27
  publication-title: J. Mach. Learn. Res.
– volume: 80
  start-page: 255
  issue: 3-4
  year: 2012
  ident: 9723_CR51
  publication-title: Publ. Math. Debrecen
  doi: 10.5486/PMD.2012.4814
– volume: 3
  start-page: 551
  issue: 5
  year: 1990
  ident: 9723_CR3
  publication-title: Neural Netw.
  doi: 10.1016/0893-6080(90)90005-6
– volume: 4
  start-page: 251
  issue: 2
  year: 1991
  ident: 9723_CR5
  publication-title: Neural Netw.
  doi: 10.1016/0893-6080(91)90009-T
– volume: 2
  start-page: 359
  issue: 5
  year: 1989
  ident: 9723_CR18
  publication-title: Neural Netw.
  doi: 10.1016/0893-6080(89)90020-8
– volume: 20
  start-page: 389
  issue: 7
  year: 2019
  ident: 9723_CR8
  publication-title: Nat. Rev. Genet.
  doi: 10.1038/s41576-019-0122-6
– ident: 9723_CR31
  doi: 10.4064/sm159-1-6
– ident: 9723_CR29
– ident: 9723_CR55
– ident: 9723_CR65
– volume: 764
  start-page: 1
  year: 2020
  ident: 9723_CR28
  publication-title: J. Reine Angew. Math.
  doi: 10.1515/crelle-2018-0037
– ident: 9723_CR17
– ident: 9723_CR74
  doi: 10.1007/978-1-4471-2170-1
– volume: 249
  start-page: 259
  issue: 3
  year: 2019
  ident: 9723_CR78
  publication-title: Studia Math.
  doi: 10.4064/sm180109-8-6
– volume-title: Modern general topology
  year: 1974
  ident: 9723_CR36
– ident: 9723_CR21
– ident: 9723_CR61
  doi: 10.1017/S0962492900002919
– volume: 6
  start-page: 861
  issue: 6
  year: 1993
  ident: 9723_CR16
  publication-title: Neural Netw.
  doi: 10.1016/S0893-6080(05)80131-5
– volume-title: Éléments de mathématique. Topologie générale. Chapitres 1 à 4
  year: 1971
  ident: 9723_CR71
– ident: 9723_CR25
– volume: 6
  start-page: 397
  year: 1956
  ident: 9723_CR26
  publication-title: Pacific J. Math.
  doi: 10.2140/pjm.1956.6.397
– ident: 9723_CR81
  doi: 10.33232/BIMS.0075.45.57
– ident: 9723_CR67
– volume: 99
  start-page: 480
  year: 1992
  ident: 9723_CR56
  publication-title: Psych. Rev.
  doi: 10.1037/0033-295X.99.3.480
– volume: 7
  start-page: 2651
  year: 2006
  ident: 9723_CR38
  publication-title: J. Mach. Learn. Res.
– ident: 9723_CR69
  doi: 10.1145/168304.168357
– ident: 9723_CR42
– volume: 20
  start-page: Paper No. 179,
  year: 2019
  ident: 9723_CR41
  publication-title: J. Mach. Learn. Res.
– ident: 9723_CR83
  doi: 10.1007/978-3-322-90559-8
– volume: 8
  start-page: cnaa002, 18
  issue: 1
  year: 2020
  ident: 9723_CR23
  publication-title: J. Complex Netw.
– ident: 9723_CR35
– volume-title: Composition operators on function spaces, North-Holland Mathematics Studies, vol. 179
  year: 1993
  ident: 9723_CR64
– ident: 9723_CR52
  doi: 10.1093/imrn/rnq127
– volume: 1
  start-page: 61
  year: 1967
  ident: 9723_CR73
  publication-title: Funkcional. Anal. i Priložen.
– volume: 72
  start-page: 619
  year: 1966
  ident: 9723_CR50
  publication-title: Bull. Amer. Math. Soc.
  doi: 10.1090/S0002-9904-1966-11541-0
– volume: 20
  start-page: 61
  issue: 1
  year: 2009
  ident: 9723_CR59
  publication-title: Neural Netw. Learn Syst.
  doi: 10.1109/TNN.2008.2005605
– volume: 105
  start-page: 171
  issue: 2
  year: 2013
  ident: 9723_CR53
  publication-title: Int. J. Comput. Vis.
  doi: 10.1007/s11263-012-0591-y
– volume: 114
  start-page: 953
  year: 1957
  ident: 9723_CR6
  publication-title: Dokl. Akad. Nauk SSSR
– ident: 9723_CR20
  doi: 10.3390/math7100992
– volume: 10
  start-page: 1095
  year: 2009
  ident: 9723_CR39
  publication-title: J. Mach. Learn. Res.
– ident: 9723_CR24
– ident: 9723_CR66
– volume: 2
  start-page: 303
  issue: 4
  year: 1989
  ident: 9723_CR4
  publication-title: Math. Control Signals Syst.
  doi: 10.1007/BF02551274
– ident: 9723_CR76
– volume: 12
  start-page: 593
  year: 2011
  ident: 9723_CR54
  publication-title: J. Mach. Learn. Res.
– volume: 45
  start-page: S143
  issue: 1
  year: 2009
  ident: 9723_CR22
  publication-title: Neuroimage
  doi: 10.1016/j.neuroimage.2008.10.052
– volume-title: Espaces vectoriels topologiques. Chapitres 1 à 5
  year: 1981
  ident: 9723_CR77
– volume-title: Topology
  year: 2000
  ident: 9723_CR37
– volume: 30
  start-page: 3
  issue: 1
  year: 2013
  ident: 9723_CR63
  publication-title: ICML
– volume: 8
  start-page: 444
  year: 1957
  ident: 9723_CR72
  publication-title: Arch. Math. (Basel)
  doi: 10.1007/BF01898849
– ident: 9723_CR10
  doi: 10.1145/3374217
– volume: 9
  start-page: 3
  issue: 51
  year: 1941
  ident: 9723_CR49
  publication-title: Rec. Math. N. S.
– ident: 9723_CR57
– volume: 65
  start-page: 386
  issue: 6
  year: 1958
  ident: 9723_CR2
  publication-title: Psych. Rev.
  doi: 10.1037/h0042519
– ident: 9723_CR46
  doi: 10.1007/s10208-020-09461-0
– ident: 9723_CR30
– volume: 8
  start-page: 101
  issue: 4
  year: 2020
  ident: 9723_CR13
  publication-title: Risks
  doi: 10.3390/risks8040101
– volume: 5
  start-page: 115
  year: 1943
  ident: 9723_CR1
  publication-title: Bull. Math. Biophys.
  doi: 10.1007/BF02478259
– volume: 69
  start-page: 33
  issue: 1
  year: 1980
  ident: 9723_CR82
  publication-title: Studia Math.
  doi: 10.4064/sm-69-1-33-40
– ident: 9723_CR19
– volume: 8
  start-page: 229
  year: 2014
  ident: 9723_CR9
  publication-title: Front. Neurosci.
  doi: 10.3389/fnins.2014.00229
– volume: 20
  start-page: Paper No. 74, 2
  year: 2019
  ident: 9723_CR12
  publication-title: J. Mach. Learn. Res.
– ident: 9723_CR44
– ident: 9723_CR75
– ident: 9723_CR48
– volume: 445
  start-page: 1097
  issue: 1
  year: 2017
  ident: 9723_CR79
  publication-title: J. Math. Anal. Appl.
  doi: 10.1016/j.jmaa.2016.08.029
– volume: 465
  start-page: 125
  issue: 1
  year: 2018
  ident: 9723_CR80
  publication-title: J. Math. Anal. Appl.
  doi: 10.1016/j.jmaa.2018.04.063
– ident: 9723_CR47
  doi: 10.1007/s00365-021-09543-4
– ident: 9723_CR32
  doi: 10.1142/9911
– ident: 9723_CR58
– ident: 9723_CR60
– volume-title: Locally convex spaces, Graduate Texts in Mathematics, vol. 269
  year: 2014
  ident: 9723_CR45
  doi: 10.1007/978-3-319-02045-7
– ident: 9723_CR7
  doi: 10.1038/d41586-018-02174-z
– volume-title: Riemannian Geometry and Geometric Analysis
  year: 2011
  ident: 9723_CR34
  doi: 10.1007/978-3-642-21298-7
– volume: 39
  start-page: 930
  issue: 3
  year: 1993
  ident: 9723_CR68
  publication-title: IEEE Trans. Inform. Theory
  doi: 10.1109/18.256500
– volume: 9
  start-page: 1615
  year: 2008
  ident: 9723_CR40
  publication-title: J. Mach. Learn. Res.
– volume: 17
  start-page: 315
  issue: 5
  year: 1931
  ident: 9723_CR62
  publication-title: Proc. Natl. Acad. Sci.
  doi: 10.1073/pnas.17.5.315
– volume: 55
  start-page: 89
  issue: 2
  year: 2015
  ident: 9723_CR33
  publication-title: Comment. Math.
– volume: 89
  start-page: 145
  year: 1971
  ident: 9723_CR70
  publication-title: Ann. Mat. Pura Appl. (4)
  doi: 10.1007/BF02414945
– volume: 105
  start-page: 171
  issue: 2
  year: 2013
  ident: 9723_CR43
  publication-title: Int. J. Comput. Vis.
  doi: 10.1007/s11263-012-0591-y
SSID ssj0009686
Score 2.525324
Snippet The universal approximation property of various machine learning models is currently only understood on a case-by-case basis, limiting the rapid development of...
SourceID gale
crossref
springer
SourceType Aggregation Database
Enrichment Source
Index Database
Publisher
StartPage 435
SubjectTerms Artificial Intelligence
Complex Systems
Computer Science
Costs (Law)
Machine learning
Mathematics
Neural networks
Subtitle Characterization, Construction, Representation, and Existence
Title The Universal Approximation Property
URI https://link.springer.com/article/10.1007/s10472-020-09723-1
Volume 89
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVLSH
  databaseName: SpringerLink Journals
  customDbUrl:
  mediaType: online
  eissn: 1573-7470
  dateEnd: 99991231
  omitProxy: false
  ssIdentifier: ssj0009686
  issn: 1012-2443
  databaseCode: AFBBN
  dateStart: 19900301
  isFulltext: true
  providerName: Library Specific Holdings
– providerCode: PRVPQU
  databaseName: ProQuest Central
  customDbUrl: http://www.proquest.com/pqcentral?accountid=15518
  eissn: 1573-7470
  dateEnd: 20241105
  omitProxy: true
  ssIdentifier: ssj0009686
  issn: 1012-2443
  databaseCode: BENPR
  dateStart: 19970301
  isFulltext: true
  titleUrlDefault: https://www.proquest.com/central
  providerName: ProQuest
– providerCode: PRVAVX
  databaseName: SpringerLINK - Czech Republic Consortium
  customDbUrl:
  eissn: 1573-7470
  dateEnd: 99991231
  omitProxy: false
  ssIdentifier: ssj0009686
  issn: 1012-2443
  databaseCode: AGYKE
  dateStart: 19970101
  isFulltext: true
  titleUrlDefault: http://link.springer.com
  providerName: Springer Nature
– providerCode: PRVAVX
  databaseName: SpringerLink Journals (ICM)
  customDbUrl:
  eissn: 1573-7470
  dateEnd: 99991231
  omitProxy: true
  ssIdentifier: ssj0009686
  issn: 1012-2443
  databaseCode: U2A
  dateStart: 19970101
  isFulltext: true
  titleUrlDefault: http://www.springerlink.com/journals/
  providerName: Springer Nature
link http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3NT8IwFH9RuOhBFDXiB9mBxIOWbN3WbUcgINFAPEiCp6UfW2I0aGAk6l_v6-hAjCHhslP70r2-1_bXvvd7AA1FhRKBrYiTcoEARUgSUe6TwOaa4IsJJw_GHAxZf-Tdj_2xSQqbFdHuxZNkvlL_SnbzAko03NGUMy5BzFPO-bZKUG7dPT90V2S7LK_wqKmrCG5frkmW-V_K2oZULMvrj6L5XtOrwKgY5SLE5LU5z0RTfv8hcNz2Nw7hwBw-rdbCWo5gJ5lUoVIUdrCMn1dhf7Akc50dQwNNyTIBHLq3ZiH_fFmkPFqP-jJ_mn2dwKjXfer0iSmuQKTr2hlhrieZL0NX8iTkNIlEQJVCH5Q2nphowiOb-6HNhU9TP-Khl6YidWyppEKIJEL3FEqT90lyBpYQqZewyFGOcDwe2KGHjh0yRaVMNGCrgVNoOJaGeVwXwHiLV5zJWikxKiXOlRI7NbhZ9vlY8G5sbH2tJy7WTomSJTe5BTg-TW8VtxC0MkTCPra8LaYlNt462yD4fLvmF7BHdcxLfktzCaVsOk-u8NCSiTraaK_dHtaNrdZht8M6-B3R1g8RPeIB
linkProvider Springer Nature
linkToHtml http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV1LSwMxEA5SD-rBR1Wszz0UPGggyWazu8ciStW2eGiht5DXgiBV2hX035vZZlsLUvCeDMvMTma-ZOYbhNqWaatTYjEtlPYARRucM5XglCgg-BKaVsWY_YHojvjTOBmHprBZXe1eP0lWJ_WvZjeeMgxwByhnYuwxzyYQWAFj_oh1llS7oprvCMRV2AevOLTK_C1jJRzVh_Lqk2gVaR720W5IEaPO3KYHaMNNmmivHr8QBW9sop3-gnJ1doja3uBRKLOA3cAV_vU6b0yMXuDKfVp-H6HRw_3wrovDCARs4piUWMTciMRksVEuU8zlOmXWek8xxOc1zKmcqCQjSiesSHKV8aLQBSXGGuuBjM7iY9SYvE_cCYq0LrgTObVUU65SknHvfpmwzBgHsKqFaK0JaQI_OIypeJNLZmPQnvTak5X2JG2hm8Wejzk7xtrV16BgCa7jJRsVOgD89wEJlex4aCk8Xk38ytvaBjL41GyN4NP_Lb9CW91hvyd7j4PnM7TNoEqlulc5R41y-ukufJpR6svqr_oB6IDEsA
linkToPdf http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV1LSwMxEA5SQfTgoyrW5x4KHjQ0yWazu8dSLfXR0oOF3kJeC4KspV1B_73JNtsHSMF7MiwzO5n5kplvAGhqIrWMkYY4E9ICFKlgSkQEYyQcwReTuCzG7A9Yb0Sfx9F4pYu_rHavniTnPQ2OpSkvWhOdtVYa32hMoIM-jn4mhBb_bFMb3dwMgw7rLGl3WTnr0ZFYQRvIQt8287eMtdBUHdDrz6Nl1Okegn2fLgbtuX2PwJbJ6-CgGsUQeM-sg73-gn51dgya1viBL7lwux1v-Pf7vEkxGLrr92nxcwJG3ce3Tg_6cQhQhSEqIAupYpFKQiVMIohJZUy0tl6jkM1xiBEpElGChIxIFqUioVkmM4yUVtqCGpmEp6CWf-bmDARSZtSwFGssMRUxSqh1xYRpopRxEKsBcKUJrjxXuBtZ8cGXLMdOe9xqj5fa47gB7hZ7JnOmjI2rb52CuXMjK1kJ3w1gv88RUvG2hZnMYtfIrryvbMC9f802CD7_3_IbsDN86PLXp8HLBdglrmClvGK5BLVi-mWubMZRyOvyp_oFpLDI1g
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=The+Universal+Approximation+Property&rft.jtitle=Annals+of+mathematics+and+artificial+intelligence&rft.au=Kratsios%2C+Anastasis&rft.date=2021-06-01&rft.issn=1012-2443&rft.eissn=1573-7470&rft.volume=89&rft.issue=5-6&rft.spage=435&rft.epage=469&rft_id=info:doi/10.1007%2Fs10472-020-09723-1&rft.externalDBID=n%2Fa&rft.externalDocID=10_1007_s10472_020_09723_1
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1012-2443&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1012-2443&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1012-2443&client=summon