Prediction games: machine learning in the presence of an adversary
A main assumption in machine learning is that the data which are used to build a predictive model are governed by the same distribution as the data which the predictive model will be exposed to at application time. This condition is violated when future data are generated in response to the presence...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
Potsdam
Universitätsverl.
2012
|
| Subjects: | |
| Online Access: | Inhaltsverzeichnis |
| Summary: | A main assumption in machine learning is that the data which are used to build a predictive model are governed by the same distribution as the data which the predictive model will be exposed to at application time. This condition is violated when future data are generated in response to the presence of a predictive model which is the case, for instance, in email spam filtering. In this thesis, we establish the concept of prediction games to handle such tasks: We model the interaction between a learner, who builds the predictive model, and a data generator, who controls the process of data generation, as an one-shot game. The game-theoretic framework enables us to explicitly model the players‘ interests, their possible actions, and their level of knowledge about each other. We study three instances of prediction games which differ regarding the order in which the players decide for their action. In case studies on email spam filtering we empirically explore properties of all derived models. We show that spam filters resulting from prediction games in the majority of cases outperform other existing baseline methods. |
| Item Description: | Online-Ausg. im Internet |
| Physical Description: | x, 121 S. graph. Darst. |
| ISBN: | 9783869562032 |
Staff View
MARC
| LEADER | 00000nam a2200000 c 4500 | ||
|---|---|---|---|
| 001 | BV040488806 | ||
| 003 | DE-604 | ||
| 005 | 20130205 | ||
| 007 | t | ||
| 008 | 121017s2012 d||| m||| 00||| eng d | ||
| 020 | |a 9783869562032 |c Pb.: 12,50 EUR |9 978-3-86956-203-2 | ||
| 035 | |a (OCoLC)816256716 | ||
| 035 | |a (DE-599)GBV727092774 | ||
| 040 | |a DE-604 |b ger | ||
| 041 | 0 | |a eng | |
| 049 | |a DE-12 |a DE-473 |a DE-92 | ||
| 084 | |a ST 300 |0 (DE-625)143650: |2 rvk | ||
| 100 | 1 | |a Brückner, Michael |e Verfasser |4 aut | |
| 245 | 1 | 0 | |a Prediction games |b machine learning in the presence of an adversary |c Michael Brückner |
| 264 | 1 | |a Potsdam |b Universitätsverl. |c 2012 | |
| 300 | |a x, 121 S. |b graph. Darst. | ||
| 336 | |b txt |2 rdacontent | ||
| 337 | |b n |2 rdamedia | ||
| 338 | |b nc |2 rdacarrier | ||
| 500 | |a Online-Ausg. im Internet | ||
| 502 | |a Zugl.: Potsdam, Univ., Diss., 2012 | ||
| 520 | 1 | |a A main assumption in machine learning is that the data which are used to build a predictive model are governed by the same distribution as the data which the predictive model will be exposed to at application time. This condition is violated when future data are generated in response to the presence of a predictive model which is the case, for instance, in email spam filtering. In this thesis, we establish the concept of prediction games to handle such tasks: We model the interaction between a learner, who builds the predictive model, and a data generator, who controls the process of data generation, as an one-shot game. The game-theoretic framework enables us to explicitly model the players‘ interests, their possible actions, and their level of knowledge about each other. We study three instances of prediction games which differ regarding the order in which the players decide for their action. In case studies on email spam filtering we empirically explore properties of all derived models. We show that spam filters resulting from prediction games in the majority of cases outperform other existing baseline methods. | |
| 650 | 0 | 7 | |a Spam-Mail |0 (DE-588)4528688-7 |2 gnd |9 rswk-swf |
| 650 | 0 | 7 | |a Maschinelles Lernen |0 (DE-588)4193754-5 |2 gnd |9 rswk-swf |
| 650 | 0 | 7 | |a Mail-Filter |0 (DE-588)4792631-4 |2 gnd |9 rswk-swf |
| 650 | 0 | 7 | |a Vorhersagetheorie |0 (DE-588)4188671-9 |2 gnd |9 rswk-swf |
| 650 | 0 | 7 | |a Spieltheorie |0 (DE-588)4056243-8 |2 gnd |9 rswk-swf |
| 655 | 7 | |0 (DE-588)4113937-9 |a Hochschulschrift |2 gnd-content | |
| 689 | 0 | 0 | |a Maschinelles Lernen |0 (DE-588)4193754-5 |D s |
| 689 | 0 | 1 | |a Vorhersagetheorie |0 (DE-588)4188671-9 |D s |
| 689 | 0 | 2 | |a Spieltheorie |0 (DE-588)4056243-8 |D s |
| 689 | 0 | 3 | |a Spam-Mail |0 (DE-588)4528688-7 |D s |
| 689 | 0 | 4 | |a Mail-Filter |0 (DE-588)4792631-4 |D s |
| 689 | 0 | |5 DE-604 | |
| 856 | 4 | 2 | |m DNB Datenaustausch |q application/pdf |u http://bvbr.bib-bvb.de:8991/F?func=service&doc_library=BVB01&local_base=BVB01&doc_number=025335858&sequence=000001&line_number=0001&func_code=DB_RECORDS&service_type=MEDIA |3 Inhaltsverzeichnis |
| 999 | |a oai:aleph.bib-bvb.de:BVB01-025335858 | ||
| 942 | 1 | 1 | |c 004 |e 22/bsb |
| 942 | 1 | 1 | |c 302.2 |e 22/bsb |
Record in the Search Index
| _version_ | 1804149551316074496 |
|---|---|
| adam_text | IMAGE 1
C O N T E N T S
1 I N T R O D U C T I O N 1
1.1 MOTIVATING EXAMPLES 2
1.2 AN ARMS RACE BETWEEN LEARNER A N D D A T A GENERATOR 4
1.3 CONTRIBUTIONS 5
1.4 OWN PREVIOUSLY PUBLISHED WORK 8
1.5 OUTLINE 10
2 L E A R N I N G P R E D I C T I V E M O D E L S 1 1
2.1 RISK MINIMIZATION 12
2.2 LEARNING IN T H E BAYESIAN FRAMEWORK 14
2.3 LINEAR DECISION FUNCTIONS 17
2.4 LOSS FUNCTIONS 19
2.5 REGULARIZATION 22
2.6 FEATURE REPRESENTATION A N D KERNELS 23
2.7 P A R A M E T E R ESTIMATION 26
3 T H E P R E D I C T I O N G A M E 2 9
3.1 AN INTRODUCTION T O G A M E THEORY 29
3.1.1 BASIC TERMS A N D DEFINITIONS 30
3.1.2 SOLUTION CONCEPTS 32
3.2 ADVERSARIAL PREDICTION PROBLEMS 35
3.2.1 N U M B E R OF PLAYERS 35
3.2.2 N U M B E R OF REPETITIONS 36
3.3 MODELING T H E PREDICTION G A M E 37
3.4 CLASSES OF PREDICTION G A M E S 39
4 N A S H P R E D I C T I O N G A M E S 4 1
4.1 NASH SOLUTION TO PREDICTION GAMES 41
4.1.1 EXISTENCE OF A NASH EQUILIBRIUM 42
4.1.2 UNIQUENESS OF T H E NASH EQUILIBRIUM 4 3
4.2 FINDING T H E UNIQUE NASH EQUILIBRIUM 48
4.2.1 AN INEXACT LINESEARCH APPROACH 48
4.2.2 A MODIFIED EXTRAGRADIENT APPROACH 50
4.3 APPLYING KERNELS 50
4.4 INSTANCES OF T H E NASH PREDICTION G A M E 52
4.4.1 NASH LOGISTIC REGRESSION 53
HTTP://D-NB.INFO/1026785863
IMAGE 2
X
4.4.2 NASH S U P P O R T VECTOR MACHINE 54
4.5 RELATED WORK 55
4.6 EMPIRICAL EVALUATION 56
4.6.1 CONVERGENCE 57
4.6.2 R.EGULARIZATION PARAMETERS 58
4.6.3 EVALUATION FOR NASH-PLAVING ADVERSARY 60
4.6.4 A CASE STUDY ON EMAIL SPAM FILTERING 61
4.6.5 EFFICIENCY VERSUS EFFECTIVENESS 63
4.6.6 NASH-EQUILIBRIAL TRANSFORMATION 63
5 S T A C K E L B E R G P R E D I C T I O N G A M E S 6 7
5.1 STACKELBERG SOLUTION TO PREDICTION G A M E S 67
5.2 AN S Q P METHOD FOR STACKELBERG PREDICTION G A M E S 71
5.3 APPLYING KERNELS 72
5.4 INSTANCES OF THE STACKELBERG PREDICTION G A M E 73
5.4.1 WORST-CASE LOSS 73
5.4.2 LINEAR LOSS 74
5.4.3 LOGISTIC LOSS 75
5.5 RELATED WORK 76
5.6 EMPIRICAL EVALUATION 77
5.6.1 A CASE STUDY ON EMAIL SPAM FILTERING 77
5.6.2 EFFICIENCY VERSUS EFFECTIVENESS 79
5.6.3 TRANSFORMATION 79
6 C O V A R I A T E S H I F T 8 1
6.1 LEARNING UNDER COVARIATE SHIFT 82
6.1.1 M A P ESTIMATION UNDER COVARIATE SHIFT 8 3
6.1.2 AN INTEGRATED MODEL 86
6.2 LOGISTIC REGRESSION I M P O R T A N C E ESTIMATION 88
6.3 A TWO-STAGE APPROXIMATION 91
6.4 APPLYING KERNELS 92
6.5 RELATED WORK 9 3
6.6 EMPIRICAL EVALUATION 94
6.6.1 A CASE S T U D Y ON EMAIL SPAM FILTERING 94
6.6.2 INSPECTION OF THE RESAMPLING WEIGHTS 96
7 C O N C L U S I O N S 9 9
B I B L I O G R A P H Y 1 0 5
A P P E N D I X 1 1 1
N O T A T I O N 1 1 9
|
| any_adam_object | 1 |
| author | Brückner, Michael |
| author_facet | Brückner, Michael |
| author_role | aut |
| author_sort | Brückner, Michael |
| author_variant | m b mb |
| building | Verbundindex |
| bvnumber | BV040488806 |
| classification_rvk | ST 300 |
| ctrlnum | (OCoLC)816256716 (DE-599)GBV727092774 |
| discipline | Informatik |
| format | Thesis Book |
| fullrecord | <?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>03042nam a2200481 c 4500</leader><controlfield tag="001">BV040488806</controlfield><controlfield tag="003">DE-604</controlfield><controlfield tag="005">20130205 </controlfield><controlfield tag="007">t</controlfield><controlfield tag="008">121017s2012 d||| m||| 00||| eng d</controlfield><datafield tag="020" ind1=" " ind2=" "><subfield code="a">9783869562032</subfield><subfield code="c">Pb.: 12,50 EUR</subfield><subfield code="9">978-3-86956-203-2</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(OCoLC)816256716</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-599)GBV727092774</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-604</subfield><subfield code="b">ger</subfield></datafield><datafield tag="041" ind1="0" ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="049" ind1=" " ind2=" "><subfield code="a">DE-12</subfield><subfield code="a">DE-473</subfield><subfield code="a">DE-92</subfield></datafield><datafield tag="084" ind1=" " ind2=" "><subfield code="a">ST 300</subfield><subfield code="0">(DE-625)143650:</subfield><subfield code="2">rvk</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Brückner, Michael</subfield><subfield code="e">Verfasser</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Prediction games</subfield><subfield code="b">machine learning in the presence of an adversary</subfield><subfield code="c">Michael Brückner</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="a">Potsdam</subfield><subfield code="b">Universitätsverl.</subfield><subfield code="c">2012</subfield></datafield><datafield tag="300" ind1=" " ind2=" "><subfield code="a">x, 121 S.</subfield><subfield code="b">graph. Darst.</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">Online-Ausg. im Internet</subfield></datafield><datafield tag="502" ind1=" " ind2=" "><subfield code="a">Zugl.: Potsdam, Univ., Diss., 2012</subfield></datafield><datafield tag="520" ind1="1" ind2=" "><subfield code="a">A main assumption in machine learning is that the data which are used to build a predictive model are governed by the same distribution as the data which the predictive model will be exposed to at application time. This condition is violated when future data are generated in response to the presence of a predictive model which is the case, for instance, in email spam filtering. In this thesis, we establish the concept of prediction games to handle such tasks: We model the interaction between a learner, who builds the predictive model, and a data generator, who controls the process of data generation, as an one-shot game. The game-theoretic framework enables us to explicitly model the players&#x2018; interests, their possible actions, and their level of knowledge about each other. We study three instances of prediction games which differ regarding the order in which the players decide for their action. In case studies on email spam filtering we empirically explore properties of all derived models. We show that spam filters resulting from prediction games in the majority of cases outperform other existing baseline methods.</subfield></datafield><datafield tag="650" ind1="0" ind2="7"><subfield code="a">Spam-Mail</subfield><subfield code="0">(DE-588)4528688-7</subfield><subfield code="2">gnd</subfield><subfield code="9">rswk-swf</subfield></datafield><datafield tag="650" ind1="0" ind2="7"><subfield code="a">Maschinelles Lernen</subfield><subfield code="0">(DE-588)4193754-5</subfield><subfield code="2">gnd</subfield><subfield code="9">rswk-swf</subfield></datafield><datafield tag="650" ind1="0" ind2="7"><subfield code="a">Mail-Filter</subfield><subfield code="0">(DE-588)4792631-4</subfield><subfield code="2">gnd</subfield><subfield code="9">rswk-swf</subfield></datafield><datafield tag="650" ind1="0" ind2="7"><subfield code="a">Vorhersagetheorie</subfield><subfield code="0">(DE-588)4188671-9</subfield><subfield code="2">gnd</subfield><subfield code="9">rswk-swf</subfield></datafield><datafield tag="650" ind1="0" ind2="7"><subfield code="a">Spieltheorie</subfield><subfield code="0">(DE-588)4056243-8</subfield><subfield code="2">gnd</subfield><subfield code="9">rswk-swf</subfield></datafield><datafield tag="655" ind1=" " ind2="7"><subfield code="0">(DE-588)4113937-9</subfield><subfield code="a">Hochschulschrift</subfield><subfield code="2">gnd-content</subfield></datafield><datafield tag="689" ind1="0" ind2="0"><subfield code="a">Maschinelles Lernen</subfield><subfield code="0">(DE-588)4193754-5</subfield><subfield code="D">s</subfield></datafield><datafield tag="689" ind1="0" ind2="1"><subfield code="a">Vorhersagetheorie</subfield><subfield code="0">(DE-588)4188671-9</subfield><subfield code="D">s</subfield></datafield><datafield tag="689" ind1="0" ind2="2"><subfield code="a">Spieltheorie</subfield><subfield code="0">(DE-588)4056243-8</subfield><subfield code="D">s</subfield></datafield><datafield tag="689" ind1="0" ind2="3"><subfield code="a">Spam-Mail</subfield><subfield code="0">(DE-588)4528688-7</subfield><subfield code="D">s</subfield></datafield><datafield tag="689" ind1="0" ind2="4"><subfield code="a">Mail-Filter</subfield><subfield code="0">(DE-588)4792631-4</subfield><subfield code="D">s</subfield></datafield><datafield tag="689" ind1="0" ind2=" "><subfield code="5">DE-604</subfield></datafield><datafield tag="856" ind1="4" ind2="2"><subfield code="m">DNB Datenaustausch</subfield><subfield code="q">application/pdf</subfield><subfield code="u">http://bvbr.bib-bvb.de:8991/F?func=service&doc_library=BVB01&local_base=BVB01&doc_number=025335858&sequence=000001&line_number=0001&func_code=DB_RECORDS&service_type=MEDIA</subfield><subfield code="3">Inhaltsverzeichnis</subfield></datafield><datafield tag="999" ind1=" " ind2=" "><subfield code="a">oai:aleph.bib-bvb.de:BVB01-025335858</subfield></datafield><datafield tag="942" ind1="1" ind2="1"><subfield code="c">004</subfield><subfield code="e">22/bsb</subfield></datafield><datafield tag="942" ind1="1" ind2="1"><subfield code="c">302.2</subfield><subfield code="e">22/bsb</subfield></datafield></record></collection> |
| genre | (DE-588)4113937-9 Hochschulschrift gnd-content |
| genre_facet | Hochschulschrift |
| id | DE-604.BV040488806 |
| illustrated | Illustrated |
| indexdate | 2024-07-10T00:24:50Z |
| institution | BVB |
| isbn | 9783869562032 |
| language | English |
| oai_aleph_id | oai:aleph.bib-bvb.de:BVB01-025335858 |
| oclc_num | 816256716 |
| open_access_boolean | |
| owner | DE-12 DE-473 DE-BY-UBG DE-92 |
| owner_facet | DE-12 DE-473 DE-BY-UBG DE-92 |
| physical | x, 121 S. graph. Darst. |
| publishDate | 2012 |
| publishDateSearch | 2012 |
| publishDateSort | 2012 |
| publisher | Universitätsverl. |
| record_format | marc |
| spelling | Brückner, Michael Verfasser aut Prediction games machine learning in the presence of an adversary Michael Brückner Potsdam Universitätsverl. 2012 x, 121 S. graph. Darst. txt rdacontent n rdamedia nc rdacarrier Online-Ausg. im Internet Zugl.: Potsdam, Univ., Diss., 2012 A main assumption in machine learning is that the data which are used to build a predictive model are governed by the same distribution as the data which the predictive model will be exposed to at application time. This condition is violated when future data are generated in response to the presence of a predictive model which is the case, for instance, in email spam filtering. In this thesis, we establish the concept of prediction games to handle such tasks: We model the interaction between a learner, who builds the predictive model, and a data generator, who controls the process of data generation, as an one-shot game. The game-theoretic framework enables us to explicitly model the players‘ interests, their possible actions, and their level of knowledge about each other. We study three instances of prediction games which differ regarding the order in which the players decide for their action. In case studies on email spam filtering we empirically explore properties of all derived models. We show that spam filters resulting from prediction games in the majority of cases outperform other existing baseline methods. Spam-Mail (DE-588)4528688-7 gnd rswk-swf Maschinelles Lernen (DE-588)4193754-5 gnd rswk-swf Mail-Filter (DE-588)4792631-4 gnd rswk-swf Vorhersagetheorie (DE-588)4188671-9 gnd rswk-swf Spieltheorie (DE-588)4056243-8 gnd rswk-swf (DE-588)4113937-9 Hochschulschrift gnd-content Maschinelles Lernen (DE-588)4193754-5 s Vorhersagetheorie (DE-588)4188671-9 s Spieltheorie (DE-588)4056243-8 s Spam-Mail (DE-588)4528688-7 s Mail-Filter (DE-588)4792631-4 s DE-604 DNB Datenaustausch application/pdf http://bvbr.bib-bvb.de:8991/F?func=service&doc_library=BVB01&local_base=BVB01&doc_number=025335858&sequence=000001&line_number=0001&func_code=DB_RECORDS&service_type=MEDIA Inhaltsverzeichnis |
| spellingShingle | Brückner, Michael Prediction games machine learning in the presence of an adversary Spam-Mail (DE-588)4528688-7 gnd Maschinelles Lernen (DE-588)4193754-5 gnd Mail-Filter (DE-588)4792631-4 gnd Vorhersagetheorie (DE-588)4188671-9 gnd Spieltheorie (DE-588)4056243-8 gnd |
| subject_GND | (DE-588)4528688-7 (DE-588)4193754-5 (DE-588)4792631-4 (DE-588)4188671-9 (DE-588)4056243-8 (DE-588)4113937-9 |
| title | Prediction games machine learning in the presence of an adversary |
| title_auth | Prediction games machine learning in the presence of an adversary |
| title_exact_search | Prediction games machine learning in the presence of an adversary |
| title_full | Prediction games machine learning in the presence of an adversary Michael Brückner |
| title_fullStr | Prediction games machine learning in the presence of an adversary Michael Brückner |
| title_full_unstemmed | Prediction games machine learning in the presence of an adversary Michael Brückner |
| title_short | Prediction games |
| title_sort | prediction games machine learning in the presence of an adversary |
| title_sub | machine learning in the presence of an adversary |
| topic | Spam-Mail (DE-588)4528688-7 gnd Maschinelles Lernen (DE-588)4193754-5 gnd Mail-Filter (DE-588)4792631-4 gnd Vorhersagetheorie (DE-588)4188671-9 gnd Spieltheorie (DE-588)4056243-8 gnd |
| topic_facet | Spam-Mail Maschinelles Lernen Mail-Filter Vorhersagetheorie Spieltheorie Hochschulschrift |
| url | http://bvbr.bib-bvb.de:8991/F?func=service&doc_library=BVB01&local_base=BVB01&doc_number=025335858&sequence=000001&line_number=0001&func_code=DB_RECORDS&service_type=MEDIA |
| work_keys_str_mv | AT brucknermichael predictiongamesmachinelearninginthepresenceofanadversary |