Federated Continual Learning to Detect Accounting Anomalies in Financial Auditing Marco Schreyer12Hamed Hemati1Damian Borth1Miklos A. Vasarhelyi2

2025-05-06 0 0 3.81MB 14 页 10玖币

侵权投诉

Federated Continual Learning to

Detect Accounting Anomalies in Financial Auditing

Marco Schreyer1,2∗Hamed Hemati1Damian Borth1Miklos A. Vasarhelyi2

1University of St.Gallen, Switzerland 2Rutgers, The State University of New Jersey, USA

{firstname.lastname}@unisg.ch,{firstname.lastname}@rutgers.edu

Abstract

The International Standards on Auditing require auditors to collect reasonable

assurance that ﬁnancial statements are free of material misstatement. At the same

time, a central objective of Continuous Assurance is the ‘real-time’ assessment of

digital accounting journal entries. Recently, driven by the advances in artiﬁcial

intelligence, Deep Learning techniques have emerged in ﬁnancial auditing to

examine vast quantities of accounting data. However, learning highly adaptive audit

models in decentralised and dynamic settings remains challenging. It requires the

study of data distribution shifts over multiple clients and time periods. In this work,

we propose a Federated Continual Learning framework enabling auditors to learn

audit models from decentral clients continuously. We evaluate the framework’s

ability to detect accounting anomalies in common scenarios of organizational

activity. Our empirical results, using real-world datasets and combined federated-

continual learning strategies, demonstrate the learned model’s ability to detect

anomalies in audit settings of data distribution shifts.

1 Introduction

The International Standards in Auditing (ISA) demand auditors to collect reasonable assurance that

ﬁnancial statements are free from material misstatements, whether caused by error or fraud [

The term fraud refers to ‘the abuse of one’s occupation for personal enrichment through the deliberate

misuse of an organisation’s resources or assets’ [

]. According to the Association of Certiﬁed Fraud

Examiners (ACFE), organisations lose 5% of their annual revenues due to fraud [

The ACFE

highlights that respondents experienced a median loss of USD 100K in the ﬁrst 7-12 months after a

fraud scheme begins. Detecting fraud and abuse quickly is critical since the longer a scheme remains

undetected, the more severe its ﬁnancial, reputational, and organisational impact [53].

In the last decade, the ongoing digital transformation has fundamentally changed the nature, recording,

and volume of audit evidence [

]. Nowadays, organizations record vast quantities of digital

accounting records, referred to as Journal Entries (JEs), in Enterprise Resource Planning (ERP)

systems [

]. For the audit profession, this unprecedented exposure to large volumes of accounting

records offers new opportunities to obtain audit-relevant insights [

]. Lately, accounting ﬁrms also

foster the development of Deep Learning (DL) capabilities [

] to learn advanced models to audit

digital journal entry data. To augment a human auditor’s capabilities DL models are applied in

various audit tasks, such as accounting anomaly detection [

] illustrated in Fig. 1, audit

sampling [

] or notes analysis [

]. However, learning DL-enabled models in auditing is

still in its infancy, exhibiting two main limitations. First, most of today’s audit models are trained

∗Work conducted while at Rutgers, The State University of New Jersey, Newark, NJ, USA.

2The ACFE study encompasses an analysis of 2,110 cases of occupational fraud surveyed between January

2020 and September 2021 in 133 countries.

Preprint. Under review.

arXiv:2210.15051v1 [cs.LG] 26 Oct 2022

p(·)

<latexit sha1_base64="B++gL5V+RwgtFMD0E8pZawk9eeA=">AAAB+HicbVBNSwMxEM3Wr1o/uurRS7AI9VJ2q6DHohePFWwtdJclm03b0GwSkqxQl/4SLx4U8epP8ea/MW33oK0PBh7vzTAzL5aMauN5305pbX1jc6u8XdnZ3duvugeHXS0yhUkHCyZUL0aaMMpJx1DDSE8qgtKYkYd4fDPzHx6J0lTwezORJEzRkNMBxchYKXKrMsoDOaLTeoATYc4it+Y1vDngKvELUgMF2pH7FSQCZynhBjOkdd/3pAlzpAzFjEwrQaaJRHiMhqRvKUcp0WE+P3wKT62SwIFQtriBc/X3RI5SrSdpbDtTZEZ62ZuJ/3n9zAyuwpxymRnC8WLRIGPQCDhLASZUEWzYxBKEFbW3QjxCCmFjs6rYEPzll1dJt9nwzxvNu4ta67qIowyOwQmoAx9cgha4BW3QARhk4Bm8gjfnyXlx3p2PRWvJKWaOwB84nz+Ix5MA</latexit>

§AB

§01.11.2021

§11:43

§31

§00404002

§8.350,12

§Import Inc.

§User 007

§TC 043

…

§RE

§01.11.2008

§11:43

§31

§00404002

§8.350,12

§Import Inc.

§User 007

§TC 043

…

...

…. …

q (·)

<latexit sha1_base64="x3ye1VwIxu0/tuqMjCjHtxHGTFE=">AAAB+HicbVDLSsNAFJ3UV62PRl26GSxC3ZSkCrosunFZwT6gCWEymbRDJzNxZiLU0C9x40IRt36KO//GaZuFth64cDjnXu69J0wZVdpxvq3S2vrG5lZ5u7Kzu7dftQ8Ou0pkEpMOFkzIfogUYZSTjqaakX4qCUpCRnrh+Gbm9x6JVFTwez1JiZ+gIacxxUgbKbCrD0HupYpO6x6OhD4L7JrTcOaAq8QtSA0UaAf2lxcJnCWEa8yQUgPXSbWfI6kpZmRa8TJFUoTHaEgGhnKUEOXn88On8NQoEYyFNMU1nKu/J3KUKDVJQtOZID1Sy95M/M8bZDq+8nPK00wTjheL4oxBLeAsBRhRSbBmE0MQltTcCvEISYS1yapiQnCXX14l3WbDPW807y5qresijjI4BiegDlxwCVrgFrRBB2CQgWfwCt6sJ+vFerc+Fq0lq5g5An9gff4Am2iTDA==</latexit>

<latexit sha1_base64="jpxWCzNs91O26GFSNTO+O/XBa6E=">AAAB7HicbVBNS8NAEJ3Ur1q/qh69LBbBU0mqoMeiF48VTFtoQ9lsp+3SzSbsbsQS+hu8eFDEqz/Im//GbZuDtj4YeLw3w8y8MBFcG9f9dgpr6xubW8Xt0s7u3v5B+fCoqeNUMfRZLGLVDqlGwSX6hhuB7UQhjUKBrXB8O/Nbj6g0j+WDmSQYRHQo+YAzaqzkP/UyPu2VK27VnYOsEi8nFcjR6JW/uv2YpRFKwwTVuuO5iQkyqgxnAqelbqoxoWxMh9ixVNIIdZDNj52SM6v0ySBWtqQhc/X3REYjrSdRaDsjakZ62ZuJ/3md1Ayug4zLJDUo2WLRIBXExGT2OelzhcyIiSWUKW5vJWxEFWXG5lOyIXjLL6+SZq3qXVRr95eV+k0eRxFO4BTOwYMrqMMdNMAHBhye4RXeHOm8OO/Ox6K14OQzx/AHzucPJruO6A==</latexit>

A"ri bute 1

A"ri bute k A"ri bute 2

…. …

…

…. …

…

Journal Entries

Rec. Error

0 20000 40000 60000

Buchung

Rekonstruktionsfehler

High Error

Low Error

LREC

<latexit sha1_base64="AKW1jwlAwyEJSzqGwRbPJNFZGm0=">AAAB+nicbVDLSsNAFL2pr1pfqS7dBIvgqiRV0GWxCC5cVLEPaEOYTKft0MkkzEyUEvMpblwo4tYvceffOGmz0NYDA4dz7uWeOX7EqFS2/W0UVlbX1jeKm6Wt7Z3dPbO835ZhLDBp4ZCFousjSRjlpKWoYqQbCYICn5GOP2lkfueBCElDfq+mEXEDNOJ0SDFSWvLMcj9AaowRS25SL7m7aqSeWbGr9gzWMnFyUoEcTc/86g9CHAeEK8yQlD3HjpSbIKEoZiQt9WNJIoQnaER6mnIUEOkms+ipdayVgTUMhX5cWTP190aCAimnga8ns6By0cvE/7xerIYXbkJ5FCvC8fzQMGaWCq2sB2tABcGKTTVBWFCd1cJjJBBWuq2SLsFZ/PIyadeqzmm1dntWqV/mdRThEI7gBBw4hzpcQxNagOERnuEV3own48V4Nz7mowUj3zmAPzA+fwBLKZQE</latexit>

Anomaly Detec-on

Journal Entries

§AB

§01.11.2021

§11:43

§31

§00404002

§8.350,12

§Import Inc.

§User 007

§TC 043

…

§RE

§01.11.2008

§11:43

§31

§00404002

§8.350,12

§Import Inc.

§User 011

§TC 043

…

...

Reconstruc-ons

…

...

No Audit

Detailed Audit

...

…

🕵""

˜xi

<latexit sha1_base64="ojcEmDv5NjofrG2pV0jlfYlgn38=">AAAB9HicbVDLSgNBEOyNrxhfUY9eBoPgKexGQY9BLx4jmAckS5id7SRDZh/OzAbDst/hxYMiXv0Yb/6Nk2QPmljQUFR1093lxYIrbdvfVmFtfWNzq7hd2tnd2z8oHx61VJRIhk0WiUh2PKpQ8BCbmmuBnVgiDTyBbW98O/PbE5SKR+GDnsboBnQY8gFnVBvJ7WkufEyfsn7Ks365YlftOcgqcXJSgRyNfvmr50csCTDUTFCluo4dazelUnMmMCv1EoUxZWM6xK6hIQ1Quen86IycGcUng0iaCjWZq78nUhooNQ080xlQPVLL3kz8z+smenDtpjyME40hWywaJILoiMwSID6XyLSYGkKZ5OZWwkZUUqZNTiUTgrP88ipp1arORbV2f1mp3+RxFOEETuEcHLiCOtxBA5rA4BGe4RXerIn1Yr1bH4vWgpXPHMMfWJ8/kaiSng==</latexit>

<latexit sha1_base64="7/C/Bz4Ujb9U9aWUvkvjo7s3DtA=">AAAB8HicbVBNSwMxEJ2tX7V+VT16CRbBU9mtgh6LXjxWsB/SriWbZtvQJLskWbEs+yu8eFDEqz/Hm//GtN2Dtj4YeLw3w8y8IOZMG9f9dgorq2vrG8XN0tb2zu5eef+gpaNEEdokEY9UJ8CaciZp0zDDaSdWFIuA03Ywvp767UeqNIvknZnE1Bd4KFnICDZWun96SL2sn7KsX664VXcGtEy8nFQgR6Nf/uoNIpIIKg3hWOuu58bGT7EyjHCalXqJpjEmYzykXUslFlT76ezgDJ1YZYDCSNmSBs3U3xMpFlpPRGA7BTYjvehNxf+8bmLCSz9lMk4MlWS+KEw4MhGafo8GTFFi+MQSTBSztyIywgoTYzMq2RC8xZeXSatW9c6qtdvzSv0qj6MIR3AMp+DBBdThBhrQBAICnuEV3hzlvDjvzse8teDkM4fwB87nDxV9kJc=</latexit>

<latexit sha1_base64="y/xuiVIcPopYG2/0jWCLg+nSmJs=">AAAB8HicbVBNSwMxEJ2tX7V+VT16CRbBU9mtgh6LXjxWsB/SriWbZtvQJLskWbEs+yu8eFDEqz/Hm//GtN2Dtj4YeLw3w8y8IOZMG9f9dgorq2vrG8XN0tb2zu5eef+gpaNEEdokEY9UJ8CaciZp0zDDaSdWFIuA03Ywvp767UeqNIvknZnE1Bd4KFnICDZWun96SGtZP2VZv1xxq+4MaJl4OalAjka//NUbRCQRVBrCsdZdz42Nn2JlGOE0K/USTWNMxnhIu5ZKLKj209nBGTqxygCFkbIlDZqpvydSLLSeiMB2CmxGetGbiv953cSEl37KZJwYKsl8UZhwZCI0/R4NmKLE8IklmChmb0VkhBUmxmZUsiF4iy8vk1at6p1Va7fnlfpVHkcRjuAYTsGDC6jDDTSgCQQEPMMrvDnKeXHenY95a8HJZw7hD5zPHxcGkJg=</latexit>

<latexit sha1_base64="/JplDnJeJAesbeKAQu3Rtw486DY=">AAAB8HicbVBNSwMxEJ31s9avqkcvwSJ4KrtV0GPRi8cK9kPatWTTbBubZJckK5Zlf4UXD4p49ed489+YtnvQ1gcDj/dmmJkXxJxp47rfztLyyuraemGjuLm1vbNb2ttv6ihRhDZIxCPVDrCmnEnaMMxw2o4VxSLgtBWMriZ+65EqzSJ5a8Yx9QUeSBYygo2V7p7u04esl7KsVyq7FXcKtEi8nJQhR71X+ur2I5IIKg3hWOuO58bGT7EyjHCaFbuJpjEmIzygHUslFlT76fTgDB1bpY/CSNmSBk3V3xMpFlqPRWA7BTZDPe9NxP+8TmLCCz9lMk4MlWS2KEw4MhGafI/6TFFi+NgSTBSztyIyxAoTYzMq2hC8+ZcXSbNa8U4r1Zuzcu0yj6MAh3AEJ+DBOdTgGurQAAICnuEV3hzlvDjvzsesdcnJZw7gD5zPH2z+kNA=</latexit>

<latexit sha1_base64="UcTy3SCh5+pJrCxLDlz2vv0GAxc=">AAAB8HicbVBNSwMxEJ31s9avqkcvwSJ4KrtV0GPRi+Clgv2Qdi3ZNNuGJtklyYpl2V/hxYMiXv053vw3pu0etPXBwOO9GWbmBTFn2rjut7O0vLK6tl7YKG5ube/slvb2mzpKFKENEvFItQOsKWeSNgwznLZjRbEIOG0Fo6uJ33qkSrNI3plxTH2BB5KFjGBjpfunh/Qm66Us65XKbsWdAi0SLydlyFHvlb66/YgkgkpDONa647mx8VOsDCOcZsVuommMyQgPaMdSiQXVfjo9OEPHVumjMFK2pEFT9fdEioXWYxHYToHNUM97E/E/r5OY8MJPmYwTQyWZLQoTjkyEJt+jPlOUGD62BBPF7K2IDLHCxNiMijYEb/7lRdKsVrzTSvX2rFy7zOMowCEcwQl4cA41uIY6NICAgGd4hTdHOS/Ou/Mxa11y8pkD+APn8wc9Z5Cx</latexit>

<latexit sha1_base64="qO9eTzGFHZSl41/FB+Ta5GakNaI=">AAAB8HicbVBNSwMxEJ2tX7V+VT16CRbBU9mtgh6LXjxWsB/SriWbZtvQJLskWbEs+yu8eFDEqz/Hm//GtN2Dtj4YeLw3w8y8IOZMG9f9dgorq2vrG8XN0tb2zu5eef+gpaNEEdokEY9UJ8CaciZp0zDDaSdWFIuA03Ywvp767UeqNIvknZnE1Bd4KFnICDZWun96SMdZP2VZv1xxq+4MaJl4OalAjka//NUbRCQRVBrCsdZdz42Nn2JlGOE0K/USTWNMxnhIu5ZKLKj209nBGTqxygCFkbIlDZqpvydSLLSeiMB2CmxGetGbiv953cSEl37KZJwYKsl8UZhwZCI0/R4NmKLE8IklmChmb0VkhBUmxmZUsiF4iy8vk1at6p1Va7fnlfpVHkcRjuAYTsGDC6jDDTSgCQQEPMMrvDnKeXHenY95a8HJZw7hD5zPH26HkNE=</latexit>

˜x1

<latexit sha1_base64="em0fdO5fKF2gSIuUzqzFUWUWNj4=">AAAB+nicbVBNS8NAEJ3Ur1q/Uj16CRbBU0mqoMeiF48V7Ae0MWw2m3bpZhN2N2qJ+SlePCji1V/izX/jts1BWx8MPN6bYWaenzAqlW1/G6WV1bX1jfJmZWt7Z3fPrO53ZJwKTNo4ZrHo+UgSRjlpK6oY6SWCoMhnpOuPr6Z+954ISWN+qyYJcSM05DSkGCkteWZ1oCgLSPaY32VO7mU098yaXbdnsJaJU5AaFGh55tcgiHEaEa4wQ1L2HTtRboaEopiRvDJIJUkQHqMh6WvKUUSkm81Oz61jrQRWGAtdXFkz9fdEhiIpJ5GvOyOkRnLRm4r/ef1UhRduRnmSKsLxfFGYMkvF1jQHK6CCYMUmmiAsqL7VwiMkEFY6rYoOwVl8eZl0GnXntN64Oas1L4s4ynAIR3ACDpxDE66hBW3A8ADP8ApvxpPxYrwbH/PWklHMHMAfGJ8/BkSUfg==</latexit>

˜x2

<latexit sha1_base64="yl/Ko7c0bpejaB758Ww9Xmh+Dtk=">AAAB+nicbVBNS8NAEN3Ur1q/Uj16CRbBU0mqoMeiF48V7Ae0MWw2k3bpZhN2N2qJ+SlePCji1V/izX/jts1BWx8MPN6bYWaenzAqlW1/G6WV1bX1jfJmZWt7Z3fPrO53ZJwKAm0Ss1j0fCyBUQ5tRRWDXiIARz6Drj++mvrdexCSxvxWTRJwIzzkNKQEKy15ZnWgKAsge8zvskbuZTT3zJpdt2ewlolTkBoq0PLMr0EQkzQCrgjDUvYdO1FuhoWihEFeGaQSEkzGeAh9TTmOQLrZ7PTcOtZKYIWx0MWVNVN/T2Q4knIS+bozwmokF72p+J/XT1V44WaUJ6kCTuaLwpRZKramOVgBFUAUm2iCiaD6VouMsMBE6bQqOgRn8eVl0mnUndN64+as1rws4iijQ3SETpCDzlETXaMWaiOCHtAzekVvxpPxYrwbH/PWklHMHKA/MD5/AAfNlH8=</latexit>

˜xj

<latexit sha1_base64="c0hcVd8ZqkiRHRCVulE5tG+VK3s=">AAAB+nicbVBNS8NAEJ3Ur1q/Uj16CRbBU0mqoMeiF48V7Ae0MWw2m3btZhN2N2qJ+SlePCji1V/izX/jts1Bqw8GHu/NMDPPTxiVyra/jNLS8srqWnm9srG5tb1jVnc7Mk4FJm0cs1j0fCQJo5y0FVWM9BJBUOQz0vXHF1O/e0eEpDG/VpOEuBEachpSjJSWPLM6UJQFJHvIb7Lb3Mto7pk1u27PYP0lTkFqUKDlmZ+DIMZpRLjCDEnZd+xEuRkSimJG8soglSRBeIyGpK8pRxGRbjY7PbcOtRJYYSx0cWXN1J8TGYqknES+7oyQGslFbyr+5/VTFZ65GeVJqgjH80VhyiwVW9McrIAKghWbaIKwoPpWC4+QQFjptCo6BGfx5b+k06g7x/XG1UmteV7EUYZ9OIAjcOAUmnAJLWgDhnt4ghd4NR6NZ+PNeJ+3loxiZg9+wfj4Bl3FlLc=</latexit>

˜xk

<latexit sha1_base64="emtxvbX73bJGn6qg9x2sSHA7bpA=">AAAB+nicbVBNS8NAEN3Ur1q/Uj16CRbBU0mqoMeiF48V7Ae0MWw2k3bpZhN2N2qJ+SlePCji1V/izX/jts1BWx8MPN6bYWaenzAqlW1/G6WV1bX1jfJmZWt7Z3fPrO53ZJwKAm0Ss1j0fCyBUQ5tRRWDXiIARz6Drj++mvrdexCSxvxWTRJwIzzkNKQEKy15ZnWgKAsge8zvsnHuZTT3zJpdt2ewlolTkBoq0PLMr0EQkzQCrgjDUvYdO1FuhoWihEFeGaQSEkzGeAh9TTmOQLrZ7PTcOtZKYIWx0MWVNVN/T2Q4knIS+bozwmokF72p+J/XT1V44WaUJ6kCTuaLwpRZKramOVgBFUAUm2iCiaD6VouMsMBE6bQqOgRn8eVl0mnUndN64+as1rws4iijQ3SETpCDzlETXaMWaiOCHtAzekVvxpPxYrwbH/PWklHMHKA/MD5/AF9OlLg=</latexit>

˜xK

<latexit sha1_base64="UM0bDloStC8RSbuex70PynGQ0Zk=">AAAB+nicbVBNS8NAEJ34WetXqkcvwSJ4KkkV9Fj0InipYD+gjWGz2bRLN5uwu1FLzE/x4kERr/4Sb/4bt20O2vpg4PHeDDPz/IRRqWz721haXlldWy9tlDe3tnd2zcpeW8apwKSFYxaLro8kYZSTlqKKkW4iCIp8Rjr+6HLid+6JkDTmt2qcEDdCA05DipHSkmdW+oqygGSP+V12nXsZzT2zatfsKaxF4hSkCgWanvnVD2KcRoQrzJCUPcdOlJshoShmJC/3U0kShEdoQHqachQR6WbT03PrSCuBFcZCF1fWVP09kaFIynHk684IqaGc9ybif14vVeG5m1GepIpwPFsUpsxSsTXJwQqoIFixsSYIC6pvtfAQCYSVTqusQ3DmX14k7XrNOanVb06rjYsijhIcwCEcgwNn0IAraEILMDzAM7zCm/FkvBjvxsesdckoZvbhD4zPHy4ulJg=</latexit>

Audit

Autoencoder Network

Figure 1: Overview autoencoder (AEN) based accounting anomaly detection setting [

]. Journal

entries corresponding to a high reconstruction error are selected for detailed audit procedures.

from scratch on stationary client data of the in-scope audit period, e.g., a ﬁnancial quarter or year [

Disregarding that organizations operate in environments in which activities rapidly and dynamically

evolve [

]. New business processes, models, or departments are constantly introduced, while

current ones are redesigned or discontinued. Second, audit models are often trained centrally on data

of a single client, e.g., the organization ‘in-scope’ of the audit [

]. Although large audit ﬁrms audit

multiple organisations operating in the same industry [

]. Such ‘peer audit clients’ [

] are affected

by similar economic and societal factors, e.g., supply-chains, market cycles, or ﬁscal policy [10].

In 1995 Thrun et al. proposed a machine learning setting in which a model incrementally learns from

a stream of experiences [

]. The underlying idea of Continual Learning (CL) refers to a progressive

model adaption; once new information becomes available [

]. Furthermore, in 2017 McMahan

proposed Federated Learning (FL), a learning setting enabling distributed clients to collaboratively

train models under the orchestration of a central server [

]. A key idea of FL is creating learning

synergies while preserving data privacy [

]. Ultimately, the Federated Continual Learning (FCL) of

highly adaptive audit models can be viewed as a central objective of Continuous Assurance [

However, learning such models in dynamic and decentralised audit settings remains challenging due

to the stability-plasticity dilemma [

]. It requires to cope with two types of data distribution shifts

[20, 9, 30], as described in the following:

•

A temporal distribution shift between time step

and

denotes a change in distributions

pt0(X)

6=pt1(X)

. In the CL setting, such shifts potentially results in the loss of a model’s ability to

perform on pt0(X)after learning from pt1(X)referred to as catastrophic forgetting [35].

•

A client distribution shift between client

ω0

and

ω1

denotes a divergence of the distributions

pω0(X)6=pω1(X)

. In the FL setting, such shifts cause the divergence of client models that

eventually result in non-convergence of the server model referred to as model interference [32].

In this work, we investigate the practical application of FCL in ﬁnancial auditing to improve the

assurance of ﬁnancial statements. In summary, we present the following contributions:

•

We propose a novel federated continual learning framework that enables auditors to incrementally

learn industry-speciﬁc models from distributed data of multiple audit clients.

•

We demonstrate that the framework enables audit models to retain previously learned knowledge

that is still relevant and replace obsolete knowledge with novel information.

•

We conduct an extensive evaluation of scenarios of common organizational activity patterns using

real-world datasets to illustrate the setting’s beneﬁt in detecting accounting anomalies.

Ultimately, we view FCL as a promising, under-explored learning setting in contemporary auditing.

The remainder of this work is structured as follows: In Section 2, we provide an overview of related

work. Section 3 follows with a description of the proposed framework to learn federated and privacy-

preserving models from vast quantities of JE data. The experimental setup and results are outlined in

Section 4 and Section 5. In Section 6, the work concludes with a summary.

2 Related Work

The application of ML in ﬁnancial audits triggered a sizable body of research by academia [

]

and practitioners [

]. This section presents our literature study focusing on the unsupervised,

federated, and continual learning of accounting data representations.

Representation Learning

denotes a machine learning technique that allows systems to discover

relevant data features to solve a given task [

]. Nowadays, most ML methods used in ﬁnancial audits

depend on ‘human’ engineered data representations [

]. Such techniques encompass, Naive Bayes

classiﬁcation [

], network analysis [

], univariate and multivariate attribute analysis [

], cluster

Audit Firm

Audit Client …

ˆx1

<latexit sha1_base64="+P4Qui9+eSbkTTsZnKRyYvNaQns=">AAAB+HicbVBNS8NAEN34WetHox69LBbBU0mqoMeiF48V7Ae0MWy2m3bpZhN2J2IN+SVePCji1Z/izX/jts1BWx8MPN6bYWZekAiuwXG+rZXVtfWNzdJWeXtnd69i7x+0dZwqylo0FrHqBkQzwSVrAQfBuoliJAoE6wTj66nfeWBK81jewSRhXkSGkoecEjCSb1f6IwLZY+5nPL/P3Ny3q07NmQEvE7cgVVSg6dtf/UFM04hJoIJo3XOdBLyMKOBUsLzcTzVLCB2TIesZKknEtJfNDs/xiVEGOIyVKQl4pv6eyEik9SQKTGdEYKQXvan4n9dLIbz0Mi6TFJik80VhKjDEeJoCHnDFKIiJIYQqbm7FdEQUoWCyKpsQ3MWXl0m7XnPPavXb82rjqoijhI7QMTpFLrpADXSDmqiFKErRM3pFb9aT9WK9Wx/z1hWrmDlEf2B9/gBtAZOV</latexit>

ˆx2

<latexit sha1_base64="RrodfGjavDLRK8cEVA6hkwxtafY=">AAAB+HicbVBNS8NAEN34WetHox69LBbBU0mqoMeiF48V7Ae0MWy2m3bpZhN2J2IN+SVePCji1Z/izX/jts1BWx8MPN6bYWZekAiuwXG+rZXVtfWNzdJWeXtnd69i7x+0dZwqylo0FrHqBkQzwSVrAQfBuoliJAoE6wTj66nfeWBK81jewSRhXkSGkoecEjCSb1f6IwLZY+5nPL/P6rlvV52aMwNeJm5BqqhA07e/+oOYphGTQAXRuuc6CXgZUcCpYHm5n2qWEDomQ9YzVJKIaS+bHZ7jE6MMcBgrUxLwTP09kZFI60kUmM6IwEgvelPxP6+XQnjpZVwmKTBJ54vCVGCI8TQFPOCKURATQwhV3NyK6YgoQsFkVTYhuIsvL5N2veae1eq359XGVRFHCR2hY3SKXHSBGugGNVELUZSiZ/SK3qwn68V6tz7mrStWMXOI/sD6/AFuhpOW</latexit>

ˆxj

<latexit sha1_base64="I3uiCGaUcfM+evvK3Pjw1uKDngY=">AAAB+HicbVBNS8NAEN3Ur1o/GvXoZbEInkpSBT0WvXisYD+grWGz3bRrN5uwOxFryC/x4kERr/4Ub/4bt20O2vpg4PHeDDPz/FhwDY7zbRVWVtfWN4qbpa3tnd2yvbff0lGiKGvSSESq4xPNBJesCRwE68SKkdAXrO2Pr6Z++4EpzSN5C5OY9UMylDzglICRPLvcGxFIHzMv5dldep95dsWpOjPgZeLmpIJyNDz7qzeIaBIyCVQQrbuuE0M/JQo4FSwr9RLNYkLHZMi6hkoSMt1PZ4dn+NgoAxxEypQEPFN/T6Qk1HoS+qYzJDDSi95U/M/rJhBc9FMu4wSYpPNFQSIwRHiaAh5wxSiIiSGEKm5uxXREFKFgsiqZENzFl5dJq1Z1T6u1m7NK/TKPo4gO0RE6QS46R3V0jRqoiShK0DN6RW/Wk/VivVsf89aClc8coD+wPn8Aw56Tzg==</latexit>

ˆxK

<latexit sha1_base64="gZQT+JZMei8Zbqaecb42Ivrqz4c=">AAAB+HicbVBNS8NAEN3Ur1o/GvXoZbEInkpSBT0WvQheKtgPaGPYbLft0s0m7E7EGvJLvHhQxKs/xZv/xm2bg7Y+GHi8N8PMvCAWXIPjfFuFldW19Y3iZmlre2e3bO/tt3SUKMqaNBKR6gREM8ElawIHwTqxYiQMBGsH46up335gSvNI3sEkZl5IhpIPOCVgJN8u90YE0sfMT3l2n95kvl1xqs4MeJm4OamgHA3f/ur1I5qETAIVROuu68TgpUQBp4JlpV6iWUzomAxZ11BJQqa9dHZ4ho+N0seDSJmSgGfq74mUhFpPwsB0hgRGetGbiv953QQGF17KZZwAk3S+aJAIDBGepoD7XDEKYmIIoYqbWzEdEUUomKxKJgR38eVl0qpV3dNq7fasUr/M4yiiQ3SETpCLzlEdXaMGaiKKEvSMXtGb9WS9WO/Wx7y1YOUzB+gPrM8flIOTrw==</latexit>

ˆxl

<latexit sha1_base64="ZPJAgvJLL/1mMrK4XOq3RcyiW54=">AAAB+HicbVBNS8NAEN34WetHox69LBbBU0mqoMeiF48V7Ae0MWy2m3bpZhN2J2IN+SVePCji1Z/izX/jts1BWx8MPN6bYWZekAiuwXG+rZXVtfWNzdJWeXtnd69i7x+0dZwqylo0FrHqBkQzwSVrAQfBuoliJAoE6wTj66nfeWBK81jewSRhXkSGkoecEjCSb1f6IwLZY+5nPL/PRO7bVafmzICXiVuQKirQ9O2v/iCmacQkUEG07rlOAl5GFHAqWF7up5olhI7JkPUMlSRi2stmh+f4xCgDHMbKlAQ8U39PZCTSehIFpjMiMNKL3lT8z+ulEF56GZdJCkzS+aIwFRhiPE0BD7hiFMTEEEIVN7diOiKKUDBZlU0I7uLLy6Rdr7lntfrtebVxVcRRQkfoGJ0iF12gBrpBTdRCFKXoGb2iN+vJerHerY9564pVzByiP7A+fwDGqJPQ</latexit>

˜xl

<latexit sha1_base64="DEmOPjZ8cK0qlVAi+McmXMjZtkI=">AAAB+nicbVBNS8NAEN3Ur1q/Uj16CRbBU0mqoMeiF48V7Ae0MWw2m3bpZhN2J2qJ+SlePCji1V/izX/jts1BWx8MPN6bYWaen3CmwLa/jdLK6tr6RnmzsrW9s7tnVvc7Kk4loW0S81j2fKwoZ4K2gQGnvURSHPmcdv3x1dTv3lOpWCxuYZJQN8JDwUJGMGjJM6sDYDyg2WPuZSy/y3jumTW7bs9gLROnIDVUoOWZX4MgJmlEBRCOleo7dgJuhiUwwmleGaSKJpiM8ZD2NRU4osrNZqfn1rFWAiuMpS4B1kz9PZHhSKlJ5OvOCMNILXpT8T+vn0J44WZMJClQQeaLwpRbEFvTHKyASUqATzTBRDJ9q0VGWGICOq2KDsFZfHmZdBp157TeuDmrNS+LOMroEB2hE+Sgc9RE16iF2oigB/SMXtGb8WS8GO/Gx7y1ZBQzB+gPjM8fYM+UuQ==</latexit>

˜xK

<latexit sha1_base64="e0R5+tmX+frFJiOT4SrximhIXOw=">AAAB+nicbVBNS8NAEJ34WetXqkcvwSJ4KkkV9Fj0InipYD+gjWGz2bRLN5uwu1FLzE/x4kERr/4Sb/4bt20O2vpg4PHeDDPz/IRRqWz721haXlldWy9tlDe3tnd2zcpeW8apwKSFYxaLro8kYZSTlqKKkW4iCIp8Rjr+6HLid+6JkDTmt2qcEDdCA05DipHSkmdW+oqygGSPuZfR/C67zj2zatfsKaxF4hSkCgWanvnVD2KcRoQrzJCUPcdOlJshoShmJC/3U0kShEdoQHqachQR6WbT03PrSCuBFcZCF1fWVP09kaFIynHk684IqaGc9ybif14vVeG5m1GepIpwPFsUpsxSsTXJwQqoIFixsSYIC6pvtfAQCYSVTqusQ3DmX14k7XrNOanVb06rjYsijhIcwCEcgwNn0IAraEILMDzAM7zCm/FkvBjvxsesdckoZvbhD4zPHy6qlJg=</latexit>

˜xj

<latexit sha1_base64="hOKW/Mo4DhTti5P2Iy+7PC85XJc=">AAAB+nicbVBNS8NAEJ3Ur1q/Uj16CRbBU0mqoMeiF48V7Ae0MWw2m3btZhN2N2qJ+SlePCji1V/izX/jts1Bqw8GHu/NMDPPTxiVyra/jNLS8srqWnm9srG5tb1jVnc7Mk4FJm0cs1j0fCQJo5y0FVWM9BJBUOQz0vXHF1O/e0eEpDG/VpOEuBEachpSjJSWPLM6UJQFJHvIvYzmN9lt7pk1u27PYP0lTkFqUKDlmZ+DIMZpRLjCDEnZd+xEuRkSimJG8soglSRBeIyGpK8pRxGRbjY7PbcOtRJYYSx0cWXN1J8TGYqknES+7oyQGslFbyr+5/VTFZ65GeVJqgjH80VhyiwVW9McrIAKghWbaIKwoPpWC4+QQFjptCo6BGfx5b+k06g7x/XG1UmteV7EUYZ9OIAjcOAUmnAJLWgDhnt4ghd4NR6NZ+PNeJ+3loxiZg9+wfj4Bl3FlLc=</latexit>

˜x1

<latexit sha1_base64="VKBtMr6uKq+0yY5TskVmLdmn0E4=">AAAB+nicbVBNS8NAEJ3Ur1q/Uj16CRbBU0mqoMeiF48V7Ae0MWw2m3bpZhN2N2qJ+SlePCji1V/izX/jts1BWx8MPN6bYWaenzAqlW1/G6WV1bX1jfJmZWt7Z3fPrO53ZJwKTNo4ZrHo+UgSRjlpK6oY6SWCoMhnpOuPr6Z+954ISWN+qyYJcSM05DSkGCkteWZ1oCgLSPaYexnN7zIn98yaXbdnsJaJU5AaFGh55tcgiHEaEa4wQ1L2HTtRboaEopiRvDJIJUkQHqMh6WvKUUSkm81Oz61jrQRWGAtdXFkz9fdEhiIpJ5GvOyOkRnLRm4r/ef1UhRduRnmSKsLxfFGYMkvF1jQHK6CCYMUmmiAsqL7VwiMkEFY6rYoOwVl8eZl0GnXntN64Oas1L4s4ynAIR3ACDpxDE66hBW3A8ADP8ApvxpPxYrwbH/PWklHMHMAfGJ8/ByiUfg==</latexit>

˜x2

<latexit sha1_base64="clyNAqPL2A1fakQuS1inRIFrZ74=">AAAB+nicbVBNS8NAEN3Ur1q/Uj16CRbBU0mqoMeiF48V7Ae0MWw2k3bpZhN2N2qJ+SlePCji1V/izX/jts1BWx8MPN6bYWaenzAqlW1/G6WV1bX1jfJmZWt7Z3fPrO53ZJwKAm0Ss1j0fCyBUQ5tRRWDXiIARz6Drj++mvrdexCSxvxWTRJwIzzkNKQEKy15ZnWgKAsge8y9jOZ3WSP3zJpdt2ewlolTkBoq0PLMr0EQkzQCrgjDUvYdO1FuhoWihEFeGaQSEkzGeAh9TTmOQLrZ7PTcOtZKYIWx0MWVNVN/T2Q4knIS+bozwmokF72p+J/XT1V44WaUJ6kCTuaLwpRZKramOVgBFUAUm2iCiaD6VouMsMBE6bQqOgRn8eVl0mnUndN64+as1rws4iijQ3SETpCDzlETXaMWaiOCHtAzekVvxpPxYrwbH/PWklHMHKA/MD5/AAitlH8=</latexit>

˜xi

…

ˆx1

ˆx2

ˆxj

ˆxK

ˆxl

˜xl

˜xK

˜xj

˜x1

˜x2

˜xi

…

ˆx1

ˆx2

ˆxj

ˆxK

ˆxl

˜xl

˜xK

˜xj

˜x1

˜x2

˜xi

…

ˆx1

ˆx2

ˆxj

ˆxK

ˆxl

˜xl

˜xK

˜xj

˜x1

˜x2

˜xi

…

ˆx1

ˆx2

ˆxj

ˆxK

ˆxl

˜xl

˜xK

˜xj

˜x1

˜x2

˜xi

…

ˆx1

ˆx2

ˆxj

ˆxK

ˆxl

˜xl

˜xK

˜xj

˜x1

˜x2

˜xi

…

Collaborative Client Collaborative Client

Local Models

Global Model

…

Scenario 1

Scenario 2Scenario 3

E(1)

<latexit sha1_base64="OoFa0F0//z4diD6V5Xy3n5iF0Bo=">AAAB7nicbVDLSgNBEOz1GeMr6tHLYBDiJexGQY9BETxGMA9I1jA7mSRDZmeXmV4hLPkILx4U8er3ePNvnCR70MSChqKqm+6uIJbCoOt+Oyura+sbm7mt/PbO7t5+4eCwYaJEM15nkYx0K6CGS6F4HQVK3oo1p2EgeTMY3Uz95hPXRkTqAccx90M6UKIvGEUrNW8f05J3NukWim7ZnYEsEy8jRchQ6xa+Or2IJSFXyCQ1pu25Mfop1SiY5JN8JzE8pmxEB7xtqaIhN346O3dCTq3SI/1I21JIZurviZSGxozDwHaGFIdm0ZuK/3ntBPtXfipUnCBXbL6on0iCEZn+TnpCc4ZybAllWthbCRtSTRnahPI2BG/x5WXSqJS983Ll/qJYvc7iyMExnEAJPLiEKtxBDerAYATP8ApvTuy8OO/Ox7x1xclmjuAPnM8fSbCO4Q==</latexit>

E(2)

<latexit sha1_base64="UfEMmDTxhwPEKSXCXKHHdscYiPE=">AAAB7nicbVDLSgNBEOz1GeMr6tHLYBDiJexGQY9BETxGMA9I1jA7mSRDZmeXmV4hLPkILx4U8er3ePNvnCR70MSChqKqm+6uIJbCoOt+Oyura+sbm7mt/PbO7t5+4eCwYaJEM15nkYx0K6CGS6F4HQVK3oo1p2EgeTMY3Uz95hPXRkTqAccx90M6UKIvGEUrNW8f01LlbNItFN2yOwNZJl5GipCh1i18dXoRS0KukElqTNtzY/RTqlEwySf5TmJ4TNmIDnjbUkVDbvx0du6EnFqlR/qRtqWQzNTfEykNjRmHge0MKQ7NojcV//PaCfav/FSoOEGu2HxRP5EEIzL9nfSE5gzl2BLKtLC3EjakmjK0CeVtCN7iy8ukUSl75+XK/UWxep3FkYNjOIESeHAJVbiDGtSBwQie4RXenNh5cd6dj3nripPNHMEfOJ8/SzaO4g==</latexit>

E(1)

E(2)

E(3)

<latexit sha1_base64="tb6v5DtyhbgCTU3xonVdRt9jt+4=">AAAB7nicbVBNSwMxEJ2tX7V+VT16CRahXspuK+ixKILHCvYD2rVk02wbmmSXJCuUpT/CiwdFvPp7vPlvTNs9aOuDgcd7M8zMC2LOtHHdbye3tr6xuZXfLuzs7u0fFA+PWjpKFKFNEvFIdQKsKWeSNg0znHZiRbEIOG0H45uZ336iSrNIPphJTH2Bh5KFjGBjpfbtY1qunU/7xZJbcedAq8TLSAkyNPrFr94gIomg0hCOte56bmz8FCvDCKfTQi/RNMZkjIe0a6nEgmo/nZ87RWdWGaAwUrakQXP190SKhdYTEdhOgc1IL3sz8T+vm5jwyk+ZjBNDJVksChOOTIRmv6MBU5QYPrEEE8XsrYiMsMLE2IQKNgRv+eVV0qpWvFqlen9Rql9nceThBE6hDB5cQh3uoAFNIDCGZ3iFNyd2Xpx352PRmnOymWP4A+fzB0y8juM=</latexit>

E(1)

E(2)

E(3)

E(1)

E(2)

E(3)

E(1)

E(2)

E(3)

E(1)

E(2)

D(1)

<latexit sha1_base64="2ZIlrcomPUwt49u2hI6M9rxRVl4=">AAAB+nicbVDLSsNAFL2pr1pfqS7dBItQNyWpgi6LunBZwT6gjWUynbRDJ5MwM1FKzKe4caGIW7/EnX/jpM1CWw8MHM65l3vmeBGjUtn2t1FYWV1b3yhulra2d3b3zPJ+W4axwKSFQxaKrockYZSTlqKKkW4kCAo8Rjre5CrzOw9ESBryOzWNiBugEac+xUhpaWCW+wFSY4xYcp3eJ1XnJB2YFbtmz2AtEycnFcjRHJhf/WGI44BwhRmSsufYkXITJBTFjKSlfixJhPAEjUhPU44CIt1kFj21jrUytPxQ6MeVNVN/byQokHIaeHoyCyoXvUz8z+vFyr9wE8qjWBGO54f8mFkqtLIerCEVBCs21QRhQXVWC4+RQFjptkq6BGfxy8ukXa85p7X67VmlcZnXUYRDOIIqOHAODbiBJrQAwyM8wyu8GU/Gi/FufMxHC0a+cwB/YHz+ALcZk6M=</latexit>

D(2)

<latexit sha1_base64="mklB/6++oknRJycIWGO9q+sZ7r0=">AAAB+nicbVDLSsNAFL2pr1pfqS7dBItQNyWpgi6LunBZwT6gjWUynbRDJ5MwM1FKzKe4caGIW7/EnX/jpM1CWw8MHM65l3vmeBGjUtn2t1FYWV1b3yhulra2d3b3zPJ+W4axwKSFQxaKrockYZSTlqKKkW4kCAo8Rjre5CrzOw9ESBryOzWNiBugEac+xUhpaWCW+wFSY4xYcp3eJ9X6STowK3bNnsFaJk5OKpCjOTC/+sMQxwHhCjMkZc+xI+UmSCiKGUlL/ViSCOEJGpGephwFRLrJLHpqHWtlaPmh0I8ra6b+3khQIOU08PRkFlQuepn4n9eLlX/hJpRHsSIczw/5MbNUaGU9WEMqCFZsqgnCguqsFh4jgbDSbZV0Cc7il5dJu15zTmv127NK4zKvowiHcARVcOAcGnADTWgBhkd4hld4M56MF+Pd+JiPFox85wD+wPj8Abifk6Q=</latexit>

D(1)

D(2)

D(1)

ˆx1

ˆx2

ˆxj

ˆxK

ˆxl

˜xl

˜xK

˜xj

˜x1

˜x2

˜xi

…

D(3)

<latexit sha1_base64="gN0zhCNoCOvhoj/Mid+QPqTF6R0=">AAAB+nicbVDLSsNAFL3xWesr1aWbYBHqpiStoMuiLlxWsA9oY5lMJ+3QySTMTJQS8yluXCji1i9x5984abPQ1gMDh3Pu5Z45XsSoVLb9baysrq1vbBa2its7u3v7ZumgLcNYYNLCIQtF10OSMMpJS1HFSDcSBAUeIx1vcpX5nQciJA35nZpGxA3QiFOfYqS0NDBL/QCpMUYsuU7vk0r9NB2YZbtqz2AtEycnZcjRHJhf/WGI44BwhRmSsufYkXITJBTFjKTFfixJhPAEjUhPU44CIt1kFj21TrQytPxQ6MeVNVN/byQokHIaeHoyCyoXvUz8z+vFyr9wE8qjWBGO54f8mFkqtLIerCEVBCs21QRhQXVWC4+RQFjptoq6BGfxy8ukXas69Wrt9qzcuMzrKMARHEMFHDiHBtxAE1qA4RGe4RXejCfjxXg3PuajK0a+cwh/YHz+ALolk6U=</latexit>

E(3)

ˆx1

ˆx2

ˆxj

ˆxK

ˆxl

˜xl

˜xK

˜xj

˜x1

˜x2

˜xi

…

E(2)

D(2)

D(3)

E(3)

E(1)

E(2)

E(3)

E(1)

E(2)

E(t)

<latexit sha1_base64="9vkV1n9Jkgta1IU/JBPvrmHaVO8=">AAAB7nicbVDLSgNBEOz1GeMr6tHLYBDiJexGQY9BETxGMA9I1jA7mSRDZmeXmV4hLPkILx4U8er3ePNvnCR70MSChqKqm+6uIJbCoOt+Oyura+sbm7mt/PbO7t5+4eCwYaJEM15nkYx0K6CGS6F4HQVK3oo1p2EgeTMY3Uz95hPXRkTqAccx90M6UKIvGEUrNW8f0xKeTbqFolt2ZyDLxMtIETLUuoWvTi9iScgVMkmNaXtujH5KNQom+STfSQyPKRvRAW9bqmjIjZ/Ozp2QU6v0SD/SthSSmfp7IqWhMeMwsJ0hxaFZ9Kbif147wf6VnwoVJ8gVmy/qJ5JgRKa/k57QnKEcW0KZFvZWwoZUU4Y2obwNwVt8eZk0KmXvvFy5vyhWr7M4cnAMJ1ACDy6hCndQgzowGMEzvMKbEzsvzrvzMW9dcbKZI/gD5/MHr8KPJA==</latexit>

E(t)

⇠0

<latexit sha1_base64="AC+vadD1fu5mfo09PGvKyJv4CAw=">AAAB73icjVDLSgNBEOyNrxhfUY9eBoPoKewmgh6DXjxGMA9IljA76U2GzM6uM7NiWPITXjwo4tXf8ebfOHkcVBQsaCiquunuChLBtXHdDye3tLyyupZfL2xsbm3vFHf3mjpOFcMGi0Ws2gHVKLjEhuFGYDtRSKNAYCsYXU791h0qzWN5Y8YJ+hEdSB5yRo2V2t173su8yXGvWPLK7gzkb1KCBeq94nu3H7M0QmmYoFp3PDcxfkaV4UzgpNBNNSaUjegAO5ZKGqH2s9m9E3JklT4JY2VLGjJTv05kNNJ6HAW2M6JmqH96U/E3r5Oa8NzPuExSg5LNF4WpICYm0+dJnytkRowtoUxxeythQ6ooMzaiwv9CaFbKXrVcuT4t1S4WceThAA7hBDw4gxpcQR0awEDAAzzBs3PrPDovzuu8NecsZvbhG5y3T7Gaj70=</latexit>

⇠2

<latexit sha1_base64="mCnEWqtiGI/1FZNC8g2B3xWW23Q=">AAAB7nicjVDLSgNBEOz1GeMr6tHLYBA8hd0o6DHoxWME84BkCbOT3mTIzOwyMyuGJR/hxYMiXv0eb/6Nk8dBRcGChqKqm+6uKBXcWN//8JaWV1bX1gsbxc2t7Z3d0t5+0ySZZthgiUh0O6IGBVfYsNwKbKcaqYwEtqLR1dRv3aE2PFG3dpxiKOlA8Zgzap3U6t7zXl6d9ErloOLPQP4mZVig3iu9d/sJyyQqywQ1phP4qQ1zqi1nAifFbmYwpWxEB9hxVFGJJsxn507IsVP6JE60K2XJTP06kVNpzFhGrlNSOzQ/van4m9fJbHwR5lylmUXF5oviTBCbkOnvpM81MivGjlCmubuVsCHVlFmXUPF/ITSrleC0Ur05K9cuF3EU4BCO4AQCOIcaXEMdGsBgBA/wBM9e6j16L97rvHXJW8wcwDd4b59QTo+N</latexit>

⇠M

<latexit sha1_base64="GJxvDS+l1S5KXdaz54WdcsMwpUg=">AAAB+nicjVDLSsNAFL3xWesr1aWbwSK4KkkVdFl040aoYB/QhjCZTtqhk0mYmagl5lPcuFDErV/izr9x0nahouCBC4dz7uUeTpBwprTjfFgLi0vLK6ultfL6xubWtl3Zaas4lYS2SMxj2Q2wopwJ2tJMc9pNJMVRwGknGJ8XfueGSsVica0nCfUiPBQsZARrI/l2pX/H/KwfYT0imGeXee7bVbfmTIH+JlWYo+nb7/1BTNKICk04VqrnOon2Miw1I5zm5X6qaILJGA9pz1CBI6q8bBo9RwdGGaAwlmaERlP160WGI6UmUWA2i4zqp1eIv3m9VIenXsZEkmoqyOxRmHKkY1T0gAZMUqL5xBBMJDNZERlhiYk2bZX/V0K7XnOPavWr42rjbF5HCfZgHw7BhRNowAU0oQUEbuEBnuDZurcerRfrdba6YM1vduEbrLdP6FGUaw==</latexit>

✓!: ![!

<latexit sha1_base64="2a9D0KvFlyy4PlJkFsPGDEI1hqE=">AAACGXicbVDJSgNBEO2JW4xb1KOXxiB4CjNRUDwFvXiMYBbIhNDTqSRNemaa7hohDPkNL/6KFw+KeNSTf2NnwSXxQcHr96roqhcoKQy67qeTWVpeWV3Lruc2Nre2d/K7ezUTJ5pDlccy1o2AGZAigioKlNBQGlgYSKgHg6uxX78DbUQc3eJQQStkvUh0BWdopXbe9bEPyNqpH4fQY6MLXxnx/aI+TxT1Vf9HaucLbtGdgC4Sb0YKZIZKO//ud2KehBAhl8yYpucqbKVMo+ASRjk/MaAYH7AeNC2NWAimlU4uG9Ejq3RoN9a2IqQT9fdEykJjhmFgO0OGfTPvjcX/vGaC3fNWKiKVIER8+lE3kRRjOo6JdoQGjnJoCeNa2F0p7zPNONowczYEb/7kRVIrFb2TYunmtFC+nMWRJQfkkBwTj5yRMrkmFVIlnNyTR/JMXpwH58l5dd6mrRlnNrNP/sD5+AIYuqGY</latexit>

✓: [

<latexit sha1_base64="6c7COxd2mOOAyslfSciH6lUxCVg=">AAACB3icbVDLSgMxFM34rPU16lKQYBFclZkqKK6KblxWsA/olJJJM21oJhOSO0IZunPjr7hxoYhbf8Gdf2PazkJbD4ScnHMvN/eESnADnvftLC2vrK6tFzaKm1vbO7vu3n7DJKmmrE4TkehWSAwTXLI6cBCspTQjcShYMxzeTPzmA9OGJ/IeRop1YtKXPOKUgJW67lEAAwakm42vAmW4vXFAU4UDNZg8um7JK3tT4EXi56SEctS67lfQS2gaMwlUEGPavqegkxENnAo2LgapYYrQIemztqWSxMx0sukeY3xilR6OEm2PBDxVf3dkJDZmFIe2MiYwMPPeRPzPa6cQXXYyLlUKTNLZoCgVGBI8CQX3uGYUxMgSQjW3f8V0QDShYKMr2hD8+ZUXSaNS9s/KlbvzUvU6j6OADtExOkU+ukBVdItqqI4oekTP6BW9OU/Oi/PufMxKl5y85wD9gfP5A4BambU=</latexit>

Audit Client

⇠0

Collaborative Client(s)

{⇠!}M

!=2

<latexit sha1_base64="KPVW4CUueSWq4xRHgTq5XNLCDr4=">AAACEnicjVBNS8NAEJ3Ur1q/oh69BIugl9JUQS9C0YsXoYL9gCaGzXbTLt1swu5GLCG/wYt/xYsHRbx68ua/cdNWUFHwwcDjvRlm5vkxo1JVq+9GYWZ2bn6huFhaWl5ZXTPXN1oySgQmTRyxSHR8JAmjnDQVVYx0YkFQ6DPS9oenud++JkLSiF+qUUzcEPU5DShGSkueueekzg31UicKSR9lTnaVOiFSA4xYep5ln8ZxLfPMsl2pjmH9TcowRcMz35xehJOQcIUZkrJrV2PlpkgoihnJSk4iSYzwEPVJV1OOQiLddPxSZu1opWcFkdDFlTVWv06kKJRyFPq6Mz9X/vRy8Tevm6jgyE0pjxNFOJ4sChJmqcjK87F6VBCs2EgThAXVt1p4gATCSqdY+l8IrVrF3q/ULg7K9ZNpHEXYgm3YBRsOoQ5n0IAmYLiFe3iEJ+POeDCejZdJa8GYzmzCNxivHwmNnvE=</latexit>

✗

…

<latexit sha1_base64="nzVWE0EHabdgG8fOVbsLCxFpiIw=">AAAB+HicbVDLSsNAFL2pr1ofrbp0M1gEVyWpgi6L3bisYB/QhjCZTtqhk0mYmQg15EvcuFDErZ/izr9x0mahrQcGDufcyz1z/JgzpW372yptbG5t75R3K3v7B4fV2tFxT0WJJLRLIh7JgY8V5UzQrmaa00EsKQ59Tvv+rJ37/UcqFYvEg57H1A3xRLCAEayN5NWqoxDrKcE8bWde6mRerW437AXQOnEKUocCHa/2NRpHJAmp0IRjpYaOHWs3xVIzwmlWGSWKxpjM8IQODRU4pMpNF8EzdG6UMQoiaZ7QaKH+3khxqNQ89M1kHlOtern4nzdMdHDjpkzEiaaCLA8FCUc6QnkLaMwkJZrPDcFEMpMVkSmWmGjTVcWU4Kx+eZ30mg3nstG8v6q3bos6ynAKZ3ABDlxDC+6gA10gkMAzvMKb9WS9WO/Wx3K0ZBU7J/AH1ucP5pyTPg==</latexit>

<latexit sha1_base64="X6IObxQunDoz0Nr1hkjvqtVpRcI=">AAAB+HicbVDLSsNAFL2pr1ofjbp0M1gEVyWpgi6L3bisYB/QhjCZTtqhk0mYmQg19EvcuFDErZ/izr9x0mahrQcGDufcyz1zgoQzpR3n2yptbG5t75R3K3v7B4dV++i4q+JUEtohMY9lP8CKciZoRzPNaT+RFEcBp71g2sr93iOVisXiQc8S6kV4LFjICNZG8u3qMMJ6QjDPWnM/a8x9u+bUnQXQOnELUoMCbd/+Go5ikkZUaMKxUgPXSbSXYakZ4XReGaaKJphM8ZgODBU4osrLFsHn6NwoIxTG0jyh0UL9vZHhSKlZFJjJPKZa9XLxP2+Q6vDGy5hIUk0FWR4KU450jPIW0IhJSjSfGYKJZCYrIhMsMdGmq4opwV398jrpNuruZb1xf1Vr3hZ1lOEUzuACXLiGJtxBGzpAIIVneIU368l6sd6tj+VoySp2TuAPrM8f6CGTPw==</latexit>

<latexit sha1_base64="RV2EhQ79FQ4v8n9+VygBPQteg8s=">AAAB+HicbVBNS8NAFHypX7V+tOrRy2IRPJWkFfRY7MVjBdsKbQib7aZdutmE3Y1QQ36JFw+KePWnePPfuGlz0NaBhWHmPd7s+DFnStv2t1Xa2Nza3invVvb2Dw6rtaPjvooSSWiPRDySDz5WlDNBe5ppTh9iSXHoczrwZ53cHzxSqVgk7vU8pm6IJ4IFjGBtJK9WHYVYTwnmaSfz0lbm1ep2w14ArROnIHUo0PVqX6NxRJKQCk04Vmro2LF2Uyw1I5xmlVGiaIzJDE/o0FCBQ6rcdBE8Q+dGGaMgkuYJjRbq740Uh0rNQ99M5jHVqpeL/3nDRAfXbspEnGgqyPJQkHCkI5S3gMZMUqL53BBMJDNZEZliiYk2XVVMCc7ql9dJv9lwWo3m3WW9fVPUUYZTOIMLcOAK2nALXegBgQSe4RXerCfrxXq3PpajJavYOYE/sD5/AOmmk0A=</latexit>

<latexit sha1_base64="d90T7On5S9j8qikUucg99QwGAJo=">AAAB+HicbVDLSsNAFL3xWeujUZduBovgqiRV0GWxGxcuKtgHtCFMppN26GQSZiZCDfkSNy4UceunuPNvnLZZaOuBgcM593LPnCDhTGnH+bbW1jc2t7ZLO+Xdvf2Din141FFxKgltk5jHshdgRTkTtK2Z5rSXSIqjgNNuMGnO/O4jlYrF4kFPE+pFeCRYyAjWRvLtyiDCekwwz5q5n93lvl11as4caJW4BalCgZZvfw2GMUkjKjThWKm+6yTay7DUjHCalwepogkmEzyifUMFjqjysnnwHJ0ZZYjCWJonNJqrvzcyHCk1jQIzOYuplr2Z+J/XT3V47WVMJKmmgiwOhSlHOkazFtCQSUo0nxqCiWQmKyJjLDHRpquyKcFd/vIq6dRr7kWtfn9ZbdwUdZTgBE7hHFy4ggbcQgvaQCCFZ3iFN+vJerHerY/F6JpV7BzDH1ifPw+yk1k=</latexit>

…

<latexit sha1_base64="GoIIiNb7nmzBR22zSmng9CT6ZzM=">AAAB+HicbVDLSsNAFL3xWeujUZduBovgqiS1oMtiNy4r2Ae0IUymk3boZBJmJkIN/RI3LhRx66e482+ctFlo64GBwzn3cs+cIOFMacf5tjY2t7Z3dkt75f2Dw6OKfXzSVXEqCe2QmMeyH2BFORO0o5nmtJ9IiqOA014wbeV+75FKxWLxoGcJ9SI8FixkBGsj+XZlGGE9IZhnrbmfNea+XXVqzgJonbgFqUKBtm9/DUcxSSMqNOFYqYHrJNrLsNSMcDovD1NFE0ymeEwHhgocUeVli+BzdGGUEQpjaZ7QaKH+3shwpNQsCsxkHlOtern4nzdIdXjjZUwkqaaCLA+FKUc6RnkLaMQkJZrPDMFEMpMVkQmWmGjTVdmU4K5+eZ106zX3qla/b1Sbt0UdJTiDc7gEF66hCXfQhg4QSOEZXuHNerJerHfrYzm6YRU7p/AH1ucP6yuTQQ==</latexit>

…

✗✗

✗ ✗

✗

…

✗✗

✗

✗✗

…

✗

Figure 2: Schematic overview of the proposed Federated Continual Learning (FCL) setting [

]

(left) to detect accounting anomalies and simulated organisational activity scenarios (right).

analysis [

], transaction log mining [

], or business process mining [

]. With the advent of

Deep Learning (DL), representation learning techniques have emerged in ﬁnancial audits [

Nowadays, the application of DL in auditing JEs encompasses, autoencoder neural networks [

adversarial autoencoders [

], or variational autoencoders [

]. Lately, self-supervised learning

techniques have been proposed to learn representations for multiple audit tasks [61].

Continual Learning (CL)

describes a setting where a model continuously trains on a sequence of

experiences [

]. The challenge of forgetting in model learning has been studied extensively [

]. Rehearsal techniques replay samples from past experiences [

] to conduct knowledge

distillation [

], while generative rehearsal techniques replay synthesised samples [

]. Recent

methods use gradients of previous tasks to mitigate gradient interference [

]. Regularisation

techniques consolidate previously acquired knowledge and restrict parameter updates in model

optimization. LwF [

], uses knowledge distillation to regularize parameter updates. EWC [

]

directly regularises the parameters based on their previous task importance. In SI [

] regularization is

applied in separate parameter optimization step. Dynamic architecture techniques prevent forgetting

by parameter reuse or increase [

] or. Lately, CL has been deployed in a variety of application

scenarios, such as, healthcare [38], machine translation [4], and ﬁnancial auditing [24].

Federated Learning (FL)

enables entities to collaboratively train ML models under the orchestration

of a central trusted entity under differential privacy [

]. FedAvg [

] aggregates the client’s

model parameters by computing a weighted average based on the number of training samples. PATE

[

] aggregates knowledge transferred from a teacher model that is trained on separated data of

student models. FedAdagrad,FedYogi, and FedAdam [

] use adaptive optimizer variations to

improve FL convergence. FedProx [

] trains the local with a proximal term that restricts the updates

to the central model. FedCurv [

] aims to minimise the model dispersity across clients by adopting

a modiﬁed version of EWC [

]. Recent works [

] also introduce Bayesian non-parametric

aggregation policies. Nowadays, FL learning is applied in sensitive application scenarios, such as,

healthcare [14], ﬁnancial risk modeling [85], and ﬁnancial auditing [62].

Federated Continual Learning (FCL)

addresses the challenge of distributed distribution shifts over

time and has only been preliminary studied [

]. Scaffold [

] applies variance reduction in the local

updates to correct for a client’s concept drifts. FedWeIT [

] decompose a model into global shared

and sparse task-speciﬁc parameters to prevent inter-client interference. CDA-FedAvg [

] extends

FedAVG by distribution based drift detection and a long-short term memory for rehearsal. VER [

]

conducts an additional server-side rehearsal based training using a VAE’s [

] embedding statistics

or actual embeddings. FedDrift [

] uses local drift detection and hierarchical clustering to learn

multiple global concept models. To the best of our knowledge, this work presents the ﬁrst step

towards the federated continual learning of highly adaptive audit models in ﬁnancial auditing.

3 Methodology

We consider a unsupervised

audit learning setting

where a dataset

formally deﬁnes a population of

i= 1,2, ..., I

JEs. Each entry, denoted by

xi={x1

i, x2

i, ..., xJ

i;x1

i, x2

i, ..., xK

, consists of

j= 1,2, ..., J

categorical accounting attributes and

k= 1,2, ..., K

numerical accounting attributes. The individual

attributes encompass the journal entry details, such as posting date, amount, or general ledger. Each

JE is generated by a particular organizational activity

,e.g., an organizational unit, entity or business

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

FederatedContinualLearningtoDetectAccountingAnomaliesinFinancialAuditingMarcoSchreyer1;2HamedHemati1DamianBorth1MiklosA.Vasarhelyi21UniversityofSt.Gallen,Switzerland2Rutgers,TheStateUniversityofNewJersey,USA{firstname.lastname}@unisg.ch,{firstname.lastname}@rutgers.eduAbstractTheInternationalStanda...

展开>> 收起<<

Federated Continual Learning to Detect Accounting Anomalies in Financial Auditing Marco Schreyer12Hamed Hemati1Damian Borth1Miklos A. Vasarhelyi2.pdf

共14页,预览3页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Federated Continual Learning to Detect Accounting Anomalies in Financial Auditing Marco Schreyer12Hamed Hemati1Damian Borth1Miklos A. Vasarhelyi2

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: