Position Paper: Quality Assurance in Deep Learning Systems

Domingos F. Oliveira

1,2 a

and Miguel A. Brito

2,3 b

Department of Informatics and Computing, Mandume Ya Ndemufaio University, Lubango, Angola

Algoritmi Centre, University of Minho, Guimar

aes, Portugal

Information Systems Department, University of Minho, Guimar

aes, Portugal

Keywords:

Deep Learning, Software Quality Assurance, Software Quality Assurance Standards, Quality Assurance in

DL Systems.

Abstract:

The use of DL as a driving force for new and next-generation technological innovation plays a vital role in the

success of organisations. Its penetration in almost all domains requires improving the quality of such systems

using quality assurance models. It has been widely explored in DM and SD projects, hence the need to resort

to methodology like KDD, SEMMA and the CRISP-DM. In this way, the reuse of standards and methods to

guarantee the quality of these systems presents itself as an opportunity. In this way, the position paper has

the fundamental objective of giving an idea about the form of a structure that facilitates the application of

quality assurance in DL systems. Creating a framework that enables quality assurance of DL systems involves

adjusting the development process of traditional methods since the challenge lies in the different programming

paradigms and the logical representation of DL software.

1 INTRODUCTION

Organisations frequently change their business re-

quirements due to their needs. Technologies, one of

the components of Information Systems (IS), play an

essential role in changing these requirements.

The use of Deep Learning (DL), as a technique

of Machine Learning (ML), and this one of the areas

of Artiﬁcial Intelligence (AI), has presented advances

in solving organisational problems, contributing to

solutions such as intelligent systems, assisted real-

ity, autonomous vehicles, medical diagnostics, and

fundamentally in solutions for Data Mining (DM),

more speciﬁcally in Data Science (DS), among oth-

ers. Moreover, for (Ma et al., 2018), the penetration

of DL in almost all domains revolutionises our daily

lives, leading to opportunities for study on how to im-

prove the assurance and quality control of such sys-

tems.

Taking into account the problems associated with

quality assurance (QA) and quality control (QC), to

lead to the success of IS in organisation with forme

states (Delone and McLean, 1992), it was deﬁned as

a central research question for the present work: how

https://orcid.org/0000-0002-2890-0655

https://orcid.org/0000-0003-4235-9700

to guarantee quality in DL system? To answer the

question, that is according to the problem presented,

it has been deﬁned as the objectives of this position

paper to give and make an analysis of some norms

of guaranteeing the quality of systems, as well as to

present and analyse some methodologies of systems

development that use DL.

To meet this objective, we have deﬁned the

methodology for this work as a search for biblio-

graphic references about relevant and current jobs in

information systems, software engineering and com-

puter science, which deal with the theme of quality

assurance in DL systems.

The present position paper comprises four sec-

tions, dismissed as follows. The ﬁrst section is the In-

troduction, the second section the Study of the Art, the

third will make a General Approach about the quality

assurance standards and some DL systems develop-

ment methodologies, as well as we present our idea

about a quality assurance framework in DL systems.

Finally, we will ﬁnish with the fourth section, corre-

sponding to the general Conclusions of this work.

Oliveira, D. and Brito, M.

Position Paper: Quality Assurance in Deep Learning Systems.

DOI: 10.5220/0011107100003269

In Proceedings of the 11th International Conference on Data Science, Technology and Applications (DATA 2022), pages 203-210

ISBN: 978-989-758-583-8; ISSN: 2184-285X

203

2 STATE OF ART

The DL technique is beneﬁcial in DM and DS design;

hence to make a study on quality assurance in systems

of the kind, one must present the concepts and a set

of norms and standards for quality assurance of tech-

niques, as well as the development methodologies for

DM and DS design, offered by the industry.

2.1 Software Quality Assurance (SQA)

The concept associated with software quality is often

difﬁcult to present. Still, the same goes through effec-

tive quality management applied to create a valuable

product that provides measurable value for producers’

users, according to (Pressman and Maxim, 2020) and

also referenced by the (Sommerville, 2016). For this

fact to be evidence, it is necessary to make the man-

agement of it.

According to (Institute, 2017), project quality

management is a process that determines quality poli-

cies, objectives and responsibilities to meet the needs

for which it was performed and, supports continuous

improvement activities, goes through a plan. Sys-

tem projects that use DL must also go through the

same phases. Otherwise, they will not meet customer

needs. That is why (Suryn, 2014) stated that quality

had become a critical attribute of software products

because its absence produces ﬁnancial, material and

sometimes life losses.

Hence (Sommerville, 2016), consider that soft-

ware quality is not simple, and the result of good

project management pass high standards, which in-

volves software engineering methods, management

techniques and software quality control and assurance

actions.

Quality control is one of the essential activities for

management. Its concept relates to a set of activities

designed to assess the quality of a developed product,

according to (Galin, 2018). It ensures that the product

complies with what was stipulated, emphasizing the

result and leaving a gap in the process because it is

only possible to have a ﬁnal quality product if the pro-

cedure for its development is rigorous as advocated by

(Society, 2014).

To obtain a quality product, you must have a rig-

orous process. To speak of a quality process is to say

of assurance, as a set of activities that deﬁne and eval-

uate the suitability of methods, for our case a, soft-

ware, to provide evidence that establishes conﬁdence

that the processes are appropriate and produce prod-

ucts of adequate quality for their intended purposes,

this according to (Society, 2014).

While (Galin, 2018), considers SQA as a set

of systematically planned actions required to pro-

vide adequate conﬁdence that software development

conforms to established functional technical require-

ments and also to the management requirements of

meeting deadlines and operating on a budget.

In this way is that the quality of software, if

present as fundamental for systems considered criti-

cal or that involve DL and help teams to create these

systems that meet the needs of users, this according

to a set of standards planned and systematic actions

required to support this need, such as (Pressman and

Maxim, 2020).

According to (Suryn, 2014), quality has become a

critical attribute of products as its absence produces

ﬁnancial, health and sometimes life losses. QA in-

volves a set of activities. Therefore, it should be em-

ployed in another set of activities being the software

development process or in the entire development cy-

cle.

Requirements

assessment

Requirements

analysis

Project Implementation Te st Deployment

Software Quality Assurance

Figure 1: The perspective of QA in the IS development cy-

cle adapted from (Suryn, 2014).

Figure 1 is a simple demonstration of what has

been addressed so far about how the quality assurance

process should be represented in the development cy-

cle of a system. Thus we consider quality assurance

to be necessary, as it establishes an infrastructure that

supports sound software engineering methods, project

management, and QC actions, all of which are essen-

tial to building software, according to (Pressman and

Maxim, 2020).

For this, the Plan-Do-Check-Act (PDCA) can

present itself with a fundamental element for the qual-

ity assurance process since the same is a quality man-

agement system, which allows the continuous im-

provement of processes and products through a ﬂow

of activities, used to pay attention and adjust the de-

viations that may occur in the process, as the ﬁgure 2

this according to (Isniah et al., 2020).

Software quality has been an approach in system

projects for a long time, but when it comes to DL sys-

tems, it is still a dilemma, according to (Ma et al.,

2018). In this way, we ﬁrst seek to understand the

norms and standards that make QA possible on sys-

tems.

DATA 2022 - 11th International Conference on Data Science, Technology and Applications

204

Plan

DoCheck

Act

Figure 2: PDCA adapted from (Isniah et al., 2020).

2.2 Software Quality Assurance

Standards

Quality standards focus on quality assurance, depend-

ing on the system and concentrate ﬁrst on accepting

or responding to user needs. Their use of means leads

to some beneﬁts, according to (Galin, 2018) develop-

ment and maintenance, better understanding and mu-

tual coordination between development and mainte-

nance teams.

Still, (Galin, 2018), in the last two decades, there

has been a rapid development of international SQA

standards. This is due to the increased coverage of

related topics, which has led to a greater understand-

ing of the standards and their need. As a result, four

standards have had wide acceptance in the commu-

nity, namely ISO 9000-3, ISO/IEC 12207, ISO/IEC

15504, and CMMI.

2.2.1 ISO 9000-3

The ISO 9000 - 3 arises from other standards such

as ISO 9000, being a series of standards for qual-

ity management, which allows for identifying errors

and streamlining operation, provides a design, devel-

opment, production, installation and services used as

criteria to qualify development organizations for con-

tracts and gives a signiﬁcant impact to the software

industry (Inoue et al., 1994).

2.2.2 ISO/IEC 15504

ISO/IEC 15504 is a set of standards which propose

models to improve and evaluate processes related to

IS and software products (Pat

on-Romero et al., 2018),

allowing the improvement of its development process.

2.2.3 CMMI Standard

CMMI provide guidance, facilitating the develop-

ment of solutions by improving your ability to man-

age product or service development, acquisition and

maintenance. It has three categories, CMMI-SVC for

services, CMMI-ACQ for purchase, and CMMI-DEV

for growth, which focus on best practices for devel-

oping quality products/services that meet or exceed

customer expectations (Ayyagari and Atoum, 2019).

2.2.4 ISO/IEC 12207

The standard, ISO/IEC 12207, is a framework that de-

ﬁnes a software life cycle process, with well-deﬁned

terms that can be referenced in the software industry

according to (Anwer et al., 2018).

2.3 Comparison of Quality Assurance

Standards

For a better analysis, table 1 was prepared to com-

pare the standards, considering an overview approach,

considering ﬁve perspectives: scope, focus, compati-

bility, implementation model, and process.

Table 2 presents the advantages and disadvantages

of quality assurance standards.

An approach has been made to the norms and

standards of quality assurance, and the assumption

of seeking to apply quality assurance to DL systems

should be taken into account. To this end, the follow-

ing sections describe the concepts of DL systems. In

addition, I am trying to understand its scope and way

of development to analyse how its quality should be

guaranteed.

2.4 Quality Assurance in DL Systems

According to (Santhanam et al., 2019), recent ad-

vances in AI using DL techniques have triggered their

large-scale use in a wide range of applications, plus

the current levels of maturity, of using an AI compo-

nent in applications considered critical can have unex-

pected consequences, leading to concerns about their

reliability, repeatability, trustworthiness, and main-

tainability. In this way, we intend to approach the

concepts of employing QA in DL systems.

2.4.1 Deep Learning Systems

DL has emerged as a new research area in ML (Deng

and Yu, 2013), allowing computational models com-

posed of several processing layers to learn data repre-

sentations with various levels of abstraction, making

Position Paper: Quality Assurance in Deep Learning Systems

205

Table 1: Comparison of quality assurance standards adapted from (Konttinen, 2016).

ISO 90003 ISO/IEC 15504 CMMI ISO/IEC 12207

Scope General Software develop-

ment process

Software develop-

ment process

Software development

process

Focus Customers and pro-

cesses

Process assessment Business and process

improvement

Framework for soft-

ware life cycle

processes

Compatibility CMMI level 3 compli-

ant

CMMI and ISO

90003 compliant

ISO 15504 compliant

Implementation Full and ﬂexible com-

pliance is required

Flexible continuous

improvement model

Focused, phased and

continuous improve-

ment models

——————

Approach Veriﬁcation of docu-

mented standards.

Evaluates the project

on capability levels.

Evaluates maturity

levels.

Focuses on pro-

curement, delivery,

operation and mainte-

nance.

Table 2: Advantages and disadvantages of quality assurance standards.

Standards Advantages Disadvantages

ISO 9000-3 Is independent of the life cycle, technology, pro-

cesses and organizational structure.

It is not used as an evaluation criterion in certiﬁ-

cations.

ISO/IEC 15504 Provides a framework for assessment, maintains

the improvement process, assesses risks, and de-

termines capabilities.

It is a more comprehensive model, has its com-

plexity, does not deﬁne a precise assessment

method.

CMMI Maturity levels; Process improvement; On-time

delivery; increased customer satisfaction; quality

culture in the programmers.

Overload on documentation, time/effort to imple-

ment, culture changes and does not integrate with

other models.

ISO/IEC 12207 To help organizations understand the components

present in the acquisition and delivery of software

effectively.

It only provides a framework of software pro-

cesses, activities, and tasks that can be identiﬁed,

planned, and executed.

information available in traditional and new scopes

and extending key aspects of artiﬁcial intelligence

(Lecun et al., 2015).

Thus, it is essential to understand its concept to

study quality assurance in systems of this nature.

According to (Johnson, 2021), DL systems can be

considered as systems that use multi-layered neural

networks to perform learning tasks, including regres-

sion, classiﬁcation, clustering, and automatic cod-

ing, applying ML techniques that exploit many layers

of non-linear information processing for supervised

or unsupervised feature extraction and transformation

and pattern analysis and classiﬁcation, following a set

of 5 phases.

The application of these techniques means that DL

systems are highly exploited nowadays. Hence qual-

ity assurance is essential, as they can be classiﬁed

as critical systems and require more excellent care in

their use.

This is why (Xiao et al., 2018), consider that the

fundamental characteristics of DL systems involve

feedback-driven exploration, where a user performs a

set of tasks to get the best result for a speciﬁc mission

based on a set of characteristics.

These features are associated with models that go

through Convolution Neural Network (CNN), Deep

Belief Networks (DBNs), Recurrent Neural Network

(RNNs) and Self Encoders (AE), where each of these

is used to solve a speciﬁc problem, as they can be

trained in a supervised, unsupervised and reinforce-

ment manner.

The characteristic is related to the explanatory

variable, and its extraction involves a set of data that

lacks greater precision. Hence, the DL systems de-

velopment process is very delicate, and this leads us

to look at the DL systems development life cycle, as

presented in ﬁgure 3.

1 – Understand the

problem

2 – Identify

data

3 – Select deep

learning algorithms

4 – Training th e

model

5 – Te st the

model

Figure 3: DL process adapted from (Johnson, 2021).

DATA 2022 - 11th International Conference on Data Science, Technology and Applications

206

2.4.2 DL Systems Development Process

According to (Zhang et al., 2019a), (Zhang et al.,

2019b), DL system development adopts a different

programming paradigm and practice from conven-

tional software applications. Hence, software engi-

neering for traditional system development cannot be-

come effective for DL system development.

The ﬁgure 4 presents a simple demonstration of

the disparity that exists between the conventional ap-

plication development process and the process used

for the development of DL systems, adapted from

(Simard et al., 2017) and (Amershi et al., 2019), as,

for DL systems, it is presented as some fundamental

data layers and lack training.

1 – Understand the

problem

2 – Identify

data

3 – Select deep

learning algorithms

4 – Tra ini ng t he

model

5 – Te st th e

model

1 – Requirements

assessment

2 – Requirements

analysis

3 – Project 4 – Implementation 5 – Te s t

6 –

Deployment

Traditional software development

process

DL system development process

Figure 4: Conventional development process vs DL systems

development process.

In this way, it becomes fundamental to make an

in-depth analysis of the same development process

aimed at the quality assurance of DL systems since

it acts on a set of activities that form the same process

since DL systems differ from traditional ones. How-

ever, the difference is visible in different activities.

Quality assurance has long been addressed for

conventional systems. For DL systems, it becomes a

challenge due to problems that can cause misuse and

avoiding normal accidents that technologies some-

times cause. Proof of this are the various works

(Masuda et al., 2018), (Ma et al., 2018), (Nakajima,

2018), (Liu et al., 2019), (Nakajima, 2019), , (Fujii

et al., 2020) that place this area as a challenge for the

scientiﬁc community and professionals.

There are some works that make an approach

on DL systems development processes (Amershi

et al., 2019), (Zhang et al., 2020), (Zhang et al.,

2019a), (Zhang et al., 2019b), where they consider

that it, adopts a data-driven programming paradigm

and practice. Hence, doing an analysis on DM/DS

projects (Foroughi and Luksch, 2018) is fundamental

using methodologies.

2.5 Methodologies for Developing DL

Systems

The management of DM and DS projects is a complex

task; hence a methodology is needed, as the develop-

ment process of these projects adopts the methods like

KDD (Fayyad et al., 1996), SEMMA (Olson and De-

len, 2008), and CRISP-DM (Schr

oer et al., 2021).

2.5.1 Knowledge Discovery in Databases - KDD

It refers to the overall process of knowledge discov-

ery from data, using an extraction process that ap-

plies speciﬁc algorithms for extracting patterns from

data (Fayyad et al., 1996). This methodology com-

prises steps for knowledge discovery from the exis-

tence of data, and DM is one of the steps in this pro-

cess (F

avero, 2019).

2.5.2 Sample, Explore, Modify, Model, and

Access - SEMMA

Developed by the SAS Institute (Institute, 2021),

it starts with a statistically representative sample of

data, facilitating the application of exploratory statis-

tics and visualisation techniques to select and trans-

form the most signiﬁcant predictive variables to the

model, predicts outcomes and assesses the model ac-

curacy, this as stated (Olson and Delen, 2008).

2.5.3 Cross-Industry Standard Process for Data

Mining - CRISP-DM

It is a process model that describes the life cycle of

DS projects, allowing planning, organising, and im-

plementation of a project involving ML or DL. It con-

sists of six phases with arrows indicating the most

important and frequent dependencies between steps

(Wehrstein and Bachmann, 2020).

2.6 Comparison of DL Systems

Development Methodologies

We present a comparison in table 3, taking into ac-

count some attributes of the KDD, SEMMA and

CRISP-DM methodologies, to give a better perspec-

tive on their use in projects involving the develop-

ment of DL systems, more speciﬁcally in the area of

DM and DS, taking as a starting point the work of

(Molero-Castillo et al., 2018).

Table 4 gives an overview of the strengths and

limitations and the advantages and disadvantages of

KDD, SEMMA and CRISP-DM methodologies.

By the table 4, it is possible to notice that the

CRISP-DM presents itself better because it allows its

stages can be reversed, facilitating the correction of

possible errors without having to ﬁnish the whole cy-

cle this as it afﬁrms (Pyvovar et al., 2019). Still, it

Position Paper: Quality Assurance in Deep Learning Systems

207

Table 3: Comparison of KDD, SEMMA and CRISP-DM methodologies adapted from (Molero-Castillo et al., 2018),

aderman and Rosander, 2018).

Description KDD SEMMA CRISP-DM

Purpose Extract hidden knowledge from

data

Guide the implementation of data

mining applications

Create a reliable, incremental

process for delivering value

Strategy Map low-level data to other forms

that may be more compact, more

abstract, or more valuable

Provide an easy to understand the

process, allowing for organised

project development and mainte-

nance

Make software development

lightweight

Phases

———— ———— Business knowledge

Data selection Sample

Data understanding

Data pre-processing Exploration

Data transformation Modiﬁcation Data preparation

Data mining Model Modelling

Interpretation and Evaluation Evaluation Evaluation

———— ———— Implementation

Table 4: Advantages and disadvantages of the KDD, SEMMA and CRISP-DM methodologies.

Methodologies Advantages Disadvantages

KDD It is iterative (Molero-Castillo et al.,

2018).

It does not describe the tasks and activities that should be per-

formed in each phase (Molero-Castillo et al., 2018). Leaves gap

in interpretation and visualization (D

aderman and Rosander,

2018).

SEMMA It is an iterative process, focuses on

data management and DM model as-

pects, supports user and various DM

techniques (D

aderman and Rosander,

2018)

It does not describe the activities carried out in each phase

(Molero-Castillo et al., 2018). Ignores the evaluation phase of

the work, designed to work with SAS Enterprise (D

aderman

and Rosander, 2018).

CRISP-DM Division of phases, tasks and activities,

iterative, supported on DM techniques

and has documentation (D

aderman and

Rosander, 2018). Stages can be re-

versed, as any problem can be cor-

rected without interfering with the cycle

(Pyvovar et al., 2019)

Does not include control, monitoring, communication, knowl-

edge management or quality, change and team management ac-

tivities, ignores maintenance or updating of models (Michal-

czyk and Scheu, 2020). Lacks some maturity model (Molero-

Castillo et al., 2018). Long and complicated process (D

aderman

and Rosander, 2018). Difﬁculty in preparing and modelling

streaming data (Kalgotra and Sharda, 2016).

presents with disadvantages the fact of not provid-

ing a sound vision of management on communica-

tion, knowledge and aspects of quality management

as it afﬁrms (Michalczyk and Scheu, 2020). Hence it

is fundamental to our idea.

2.7 Focus on Quality Assurance in DL

Systems

Assuming that a DL system is the application of tech-

niques on a set of data to be trained to learn by itself

through patterns in several layers of processing, their

quality assurance involves applying a set of parame-

ters to the development process of these systems to

have a product that meets customer needs.

Proof said is what they considered (Ma et al.,

2018), where they deﬁned DL systems assurance as

to the accuracy that training data plays in modelling

the learning process and decision logic of systems and

that the same systematic approach lacks quality con-

trol, inspection and evaluation as the same data can be

introduced as malicious.

In our view, quality assurance should be seen

throughout the development process of DL systems to

eliminate possible errors that can already be detected

in the initial phase of the process itself.

3 APPROACH

As we have presented above, the DM and DS project

methodologies are fundamental to the DL system de-

velopment process. Hence, associated with PDCA

may be helpful in the implementation of quality as-

DATA 2022 - 11th International Conference on Data Science, Technology and Applications

208

surance of these systems. In this way, our perspec-

tive on quality assurance of DL systems, based on the

combination of the same, is our focus.

3.1 Quality Assurance Perspective in

DL Systems

Given the various opinions on quality assurance in DL

systems, we present our vision on a proposal to create

a framework that can assist the process of develop-

ing DL systems safely and with excellent quality. Our

concept aims ﬁrst to deﬁne a set of activities that al-

low the development of these systems with quality.

The model devised by us is based on the use of

two tools already proven and widely used in industry

and academia, which is the use of PDCA, applied to

CRISP-DM, giving a framework perspective on qual-

ity assurance in DL system as illustrated by 5, which

presents itself as an opportunity to be used in DM and

DS projects, as is the case of CRISP-ML(Q) (Studer

et al., 2020).

Figure 5: CRISP-DL(Q).

We intend to develop a framework called CRISP-

DL(Q), which will be applied to the development

of DL systems, enabling quality assurance, combin-

ing CRISP-DM and PDCA to overcome the limita-

tions presented by CRISP-DM, which does not in-

clude quality management.

4 CONCLUSIONS

The present position paper approaches SQA and

SQC, where some standards for software develop-

ment process are presented, namely ISO 90 003,

ISO/IEC 15 504, CMMI and ISO/IEC 12 207, their

advantages and disadvantages, as well as a compar-

ison of them. Also, we present the concepts of DL

systems, where one of the most signiﬁcant applica-

tions is in the development of DM and DS projects,

using methodologies such as KDD, SEMMA and the

CRISP-DM, which were also presented and analysed.

We end the document by presenting a perspective

on our idea about a framework that enables quality

assurance of DL systems, all by the literature. In

this way, the development and implementation of a

quality assurance framework for DL systems, namely

CRISP-DL(Q), presents a challenge for future work

to be developed.

ACKNOWLEDGEMENTS

This work has been supported by FCT – Fundac¸

para a Ci

encia e Tecnologia within the R&D Units

Project Scope: UIDB/00319/2020.

REFERENCES

Amershi, S., Begel, A., Bird, C., DeLine, R., Gall, H., Ka-

mar, E., Nagappan, N., Nushi, B., and Zimmermann,

T. (2019). Software Engineering for Machine Learn-

ing: A Case Study. Proceedings - 2019 IEEE/ACM

41st International Conference on Software Engineer-

ing: Software Engineering in Practice, ICSE-SEIP

2019, pages 291–300.

Anwer, S., Wen, L., Rout, T., and Wang, Z. (2018). In-

troducing requirements change management process

into ISO/IEC 12207. Communications in Computer

and Information Science, 918:185–199.

Ayyagari, M. R. and Atoum, I. (2019). CMMI-DEV Im-

plementation Simpliﬁed: A Spiral Software Model.

International Journal of Advanced Computer Science

and Applications, 10(4):445–450.

aderman, A. and Rosander, S. (2018). Evaluating Frame-

works for Implementing Machine Learning in Sig-

nal Processing: A Comparative Study of CRISP-DM,

SEMMA and KDD. Technical report.

Delone, W. H. and McLean, E. R. (1992). The quest for the

dependent variable. Information Systems Research.

Information System Research, 3(1):60–95.

Deng, L. and Yu, D. (2013). Deep learning: Methods and

applications. Foundations and Trends in Signal Pro-

cessing, 7(3-4):197–387.

avero, L. P. (2019). KDD e Data Mining: mais do que

apenas conceitos.

Fayyad, U., Piatetsky-Shapiro, G., and Smyth, P. (1996).

From Data Mining to Knowledge Discovery in

Databases. American Association for Artiﬁcial Intel-

ligence, 17(3):37–54.

Foroughi, F. and Luksch, P. (2018). Data Science Methodol-

ogy for Cybersecurity Projects. 5th International Con-

ference on Artiﬁcial Intelligence and Applications,

pages 1–14.

Fujii, G., Hamada, K., Ishikawa, F., Masuda, S., Matsuya,

M., Myojin, T., Nishi, Y., Ogawa, H., Toku, T., Toku-

moto, S., Tsuchiya, K., and Ujita, Y. (2020). Guide-

Position Paper: Quality Assurance in Deep Learning Systems

209

lines for Quality Assurance of Machine Learning-

Based Artiﬁcial Intelligence. International Journal

of Software Engineering and Knowledge Engineering,

30(11-12):1589–1606.

Galin, D. (2018). Software Quality: Concepts and Practice.

John Wiley & Sons, Inc., 1th edition.

Inoue, K., Watanabe, A., Iida, H., and Torii, K. (1994).

Modeling Method for Management Process and Its

Application to CMM and ISO9000-3. Proceedings

of the Third International Conference on the Software

Process. Applying the Software Process, pages 1–14.

Institute, P. M. (2017). A Guide to the Project Management

Body of Knowledge. Project Management Institute,

Pennsylvania, 6th edition.

Institute, T. S. (2021). About SAS.

Isniah, S., Hardi Purba, H., and Debora, F. (2020). Plan do

check action (PDCA) method: literature review and

research issues. Jurnal Sistem dan Manajemen Indus-

tri, 4(1):72–81.

Johnson, D. (2021). Deep Learning Tutorial: Neural Net-

work Basics for Beginners.

Kalgotra, P. and Sharda, R. (2016). Progression analysis

of signals: Extending CRISP-DM to stream analyt-

ics. 2016 IEEE International Conference on Big Data,

pages 2880–2885.

Konttinen, V. (2016). Towards Disciplined Software Devel-

opment: ISO 9001:2008 Based Software Process Im-

provement in SME. Technical Report May, Helsinki

Metropolia University of Applied Sciences.

Lecun, Y., Bengio, Y., and Hinton, G. (2015). Deep learn-

ing. Nature, 521(7553):436–444.

Liu, Y., Ma, L., and Zhao, J. (2019). Secure Deep Learn-

ing Engineering: A Road Towards Quality Assurance

of Intelligent Systems. Lecture Notes in Computer

Science (including subseries Lecture Notes in Artiﬁ-

cial Intelligence and Lecture Notes in Bioinformatics),

11852 LNCS:3–15.

Ma, L., Juefei-Xu, F., Xue, M., Hu, Q., Chen, S., Li, B., Liu,

Y., Zhao, J., Yin, J., and See, S. (2018). Secure Deep

Learning Engineering: A Software Quality Assurance

Perspective. arXiv, (Dl):1–11.

Masuda, S., Ono, K., Yasue, T., and Hosokawa, N. (2018).

A survey of software quality for machine learning ap-

plications. 11th International Conference on Software

Testing, Veriﬁcation and Validation Workshops, pages

279–284.

Michalczyk, S. and Scheu, S. (2020). Designing an An-

alytical Information Systems Engineering Method.

Twenty-Eighth European Conference on Information

Systems (ECIS2020), (June).

Molero-Castillo, G., B

arcenas, E., S

anchez, G., and

Antonio-Aquino, A. (2018). User-centered data min-

ing tool for survival-mortality classiﬁcation of breast

cancer in mexican-origin women. Intelligent Data

Sensing and Processing for Health and Well-being Ap-

plications, (January):223–242.

Nakajima, S. (2018). [ Invited ] Quality Assurance of Ma-

chine Learning Software. 7th Global Conference on

Consumer Electronics (GCCE 2018).

Nakajima, S. (2019). Quality Evaluation Assurance Levels

for Deep Neural Networks Software. 2019 Interna-

tional Conference on Technologies and Applications

of Artiﬁcial Intelligence.

Olson, D. L. and Delen, D. (2008). Advanced Data Mining

Techniques. Springer-Verlag Berlin Heidelberg.

Pat

on-Romero, J. D., Baldassarre, M. T., Rodr

ıguez, M.,

and Piattini, M. (2018). Green IT Governance and

Management based on ISO/IEC 15504. Computer

Standards and Interfaces, 60(May):26–36.

Pressman, R. S. and Maxim, B. R. (2020). Software En-

gineering - A Proctitioner’s Approach. McGraw-Hill

Education, New York, 9th edition.

Pyvovar, N., Yamkovyi, K., and Vechirko, M. (2019). Data

Science project management methodologies.

Santhanam, P., Farchi, E., and Pankratius, V. (2019). Engi-

neering Reliable Deep Learning Systems. AAAI Fall

Symposium Series on AI in Government & Public Sec-

tor, pages 1–8.

Schr

oer, C., Kruse, F., and G

omez, J. M. (2021). A system-

atic literature review on applying CRISP-DM process

model. Procedia Computer Science, 181(2019):526–

534.

Simard, P. Y., Amershi, S., Chickering, D. M., Pelton, A. E.,

Ghorashi, S., Meek, C., Ramos, G., Suh, J., Ver-

wey, J., Wang, M., and Wernsing, J. (2017). Ma-

chine Teaching: A New Paradigm for Building Ma-

chine Learning Systems. arXiv, pages 1–14.

Society, I. C. (2014). Guide to the Software Engineering

Body of Knowledge. IEEE Computer Society, pages

1–335.

Sommerville, I. (2016). Software Engineering. Pearson

Education Limited, 10th edition.

Studer, S., Bui, T. B., Drescher, C., Hanuschkin, A., Win-

kler, L., Peters, S., and Mueller, K.-R. (2020). To-

wards CRISP-ML(Q): A Machine Learning Process

Model with Quality Assurance Methodology. arXiv.

Suryn, W. (2014). Software Quality Engineering: A Prac-

titioner’s Approach. John Wiley & Sons, Inc, New

Jersey.

Wehrstein, L. and Bachmann, B. (2020). CRISP-DM ready

for Machine Learning Projects.

Xiao, W., Bhardwaj, R., Ramjee, R., Sivathanu, M., Kwatra,

N., Han, Z., Patel, P., Peng, X., Zhao, H., Zhang, Q.,

Yang, F., and Zhou, L. (2018). Gandiva: Introspec-

tive cluster scheduling for deep learning. 3th USENIX

Symposium on Operating Systems Design and Imple-

mentation, OSDI 2018, pages 595–610.

Zhang, J. M., Harman, M., Ma, L., and Liu, Y. (2020).

Machine Learning Testing: Survey, Landscapes and

Horizons. IEEE Transactions on Software Engineer-

ing, X(X):1–37.

Zhang, T., Gao, C., Ma, L., Lyu, M., and Kim, M. (2019a).

An Empirical Study of Common Challenges in Devel-

oping Deep Learning Applications. 30th International

Symposium on Software Reliability Engineering, IS-

SRE, 2019-Octob:104–115.

Zhang, X., Yang, Y., Feng, Y., and Chen, Z. (2019b).

Software Engineering Practice in the Development of

Deep Learning Applications. arXiv.

DATA 2022 - 11th International Conference on Data Science, Technology and Applications

210