SlideShare a Scribd company logo
Machine Learning Systems Engineering (MLSE):
Retrospective of Five-Year Activities in Japan
Hiroshi Maruyama*
Preferred Networks / Kao Corporation / U. Tokyo
Twitter; @maruyama
* Presentation is done by me but this is a collective effort by the MLSE members
Agenda
1. Backdrop
2. Launch of MLSE
3. Activities
4. Research Results
5. Retrospective
Googleʼs seminal paper on “technical debt” of ML systems (2015)
Workshop on “Towards real world implementation of ML systems”, in
conjunction with JSAI annual convention, 2016
● How to deploy ML systems for industry
applications
● Recent trends in ML businesses
● Intellectual property in ML systems
● Nikoniko-Deep Learning β
● Artificial life hackason, tried
● Evolution of communities, such as open
source
Business
Community
https://medium.com/@karpathy/software-2-0-a64152b37c35
Software 2.0: Find a program rather than write it by hand
Find a program
Set of programs that satisfy the spec
Search
algorithm
Write a program vs
x
x
🡺 Software 2.0 requires completely different set of skills
Concerns on the shortage of ML skills
Source: METI, IT人材の最新動向と将来推計に関
する調査結果 (Study results on the latest status
and the future trends of IT talents), 2016
Shortage of 48,000 talents on “big
data, IOT, and AI” in 2020
8
“Shortage of skills” -- doesn’t this sound familiar?
Who can write software ⇒ Software Crisis (1960’s)
🡺 Dawn of Software Engineering!
IBM System 360
Source: Wikipedia
System 360 Instruction set
Source:Quora
https://www.quora.com/How-did-you-learn-an-assembly-language-and-which-one
Keynote at APRES (Asia-Pacific Requirements Engineering
Symposium) by Maruyama (2016)
My email to Mikio Aoyama
“We have a number of projects going on
with customers. Sometimes I feel that we
are reinventing software engineering
practices in ML.
I wonder if the SE community can help us
to establish a new engineering discipline
in ML”
Mikio’s email to the organizer of SES2017, suggesting a
panel discussion on SE for ML
Annual SES (Software Engineering Symposium)
is the largest event in Japan dedicated to
Software Engineering, and Mikio suggested this
is the best place to draw the interests from the
SE community
Agenda
1. Backdrop
2. Launch of MLSE
3. Activities
4. Research Results
5. Retrospective
Panel discussion on machine learning engineering at SES 2017
“This panel discussion was the
highlight of this year’s SES”
Panelists
● Fuyuki Ishikawa (NII)
● Koichi Hamada (DeNA)
● Hiroshi Maruyama (PFN)
Moderator
● Mikio Aoyama INanzan U.)
https://www.facebook.com/bonotake/posts/1504893489556668
A couple of meetups among SE and ML engineers revealed new gaps
● ML engineer (ME): “my improved model now gives an erroneous output for a certain input
that was ok with my previous model”
● Software engineer (SE): “What did you do with your regression test? Don’t you have one?
● ME: “...” (ah, but what does regression test mean in ML? How can we do it?)
● ME: “My customer is concerned with the safety”
● SE: “What is the invariant in your code?”
● ME: “...” (Invariant? In an ML system?)
● SE: “You are concerned with the quality. Why don’t you use stronger-typed language than
Python?”
● ME: “...” (yes, I wish Python could statically check the shape of numpy ndarray)
Sorrow of ML Project: “Curse of infinite PoC”
Develop
model
Evaluate
Yes, we
achieved xx%
accuracy!
Can you
make a little
better?
ML engineer Customer
How can we have a reasonable level of customer expectation?
Looks good!
But not enough for
my customer
Michael Jordanʼs blog on the need for new engineering discipline
https://medium.com/@mijordan3/artificial-intelligence-the-revoluti
on-hasnt-happened-yet-5e1d5812e1e7
“... we do have a major challenge on our hands in
bringing together computers and humans in ways
that enhance human life. While this challenge is
viewed by some as subservient to the creation of
“artificial intelligence,” it can also be viewed more
prosaically — but with no less reverence — as
the creation of a new branch of engineering.
Much like civil engineering and chemical
engineering in decades past, this new discipline
aims to corral the power of a few key ideas, bringing
new resources and capabilities to people, and doing
so safely. “
The role of engineering -- my personal view
Theories * Safety Factor
Engineering as a form of agreement between engineers and the society
Civil Engineering Handbook, p999
Why do we trust bridges?
Because of the accumulated knowledge
called Civil Engineering
In Apr. 2018, the SIG on MLSE (Machine Learning Systems
Engineering, pronounced as “Mel-See”) is formed under JSSST
https://mlxse.connpass.com/
Agenda
1. Backdrop
2. Launch of MLSE
3. Activities
4. Research Results
5. Retrospective
MLSE kick-off meeting, Mar. 2018 (>500 participants)
Source: https://ledge.ai/mlse-symposium/
● Mikio Aoyama (Nanzan U), “Expectations to
MLSE”
● Takuya Kudo (Accenture), “Challenges in
software engineering and the new form of ML”
● Masashi Sugiyama (Riken AIP), “Current and
future of ML research”
● Akimichi Ariga (Cloudara), “ML starting from
business applications”
● Takahiro Kubo (TIS), “ML code design without
remorse”
● Shin Nakajima (NII), “Quality assurance of ML
software”
Common issues in ML-in-practice have surfaced
See https://leapmind.io/blog/2018/06/12/mlsekickoff/ for a report on the symposium
First things first: What are the challenges of MLSE?
Ishikawa, Fuyuki, and Nobukazu Yoshioka. "How do engineers perceive difficulties in engineering of machine-learning systems?-questionnaire survey." 2019 IEEE/ACM
Joint 7th International Workshop on Conducting Empirical Studies in Industry (CESI) and 6th International Workshop on Software Engineering Research and Industrial
Practice (SER&IP). IEEE, 2019.
Working groups
● ML operational infrastructures /
operations WG
● ML fairness WG
● ML system foundation WG
● Data quality engineering WG
● ML operations WG
● ML development process and case
studies WG
● ML security WG
● :
Active discussions on Discord, everybody is welcome
MLSE summer camps
● Main venue for community discussions
● 2-3 days, plenary / parallel sessions (workshops) + posters
● WG’s to report annual findings
● Every year new ideas are coming out
Online due to Covid-19
2019 venue in Hakone hot spa
Cooperative gatherings
Symposium on safety for AI/IoT systems
Symposium on ML and fairness w/ JSAI and
IBIS/ML, Jan., 2020
International activities -- 1/2
● iMLSE -- International Workshop on Machine Learning Systems Engineering, in conjunction with APSEC
1st iMLSE in Nara (2018) iMLSE 2020 (online)
Clark Barrett Jacomo Corbo
iMLSE 2021 (online)
Hironori Washizaki
● Shonan meeting, Nov. 2019
● MLSE International Symposium (2019)
● Sanjit Seshua, “Towards Verified Artificial
Intelligence”
● Akira Sakakibara, “Engineer's Responsibility
in Machine Learning Era”
● Foutse Khomh, “Towards Debugging and
Testing Deep Learning Systems”
● Lei Ma, “ Towards Testing and Analysis of
Deep Learning Systems”
● Amel Bennaceur, “Requirements for Machine
Learning Applications”
● Rüdiger Ehlers, “The Role of Verification in the
Engineering Process of Complex
Cyber-Physical Systems That Employ Machine
Learning”
International activities -- 2/2
Agenda
1. Backdrop
2. Launch of MLSE
3. Activities
4. Research Results
5. Retrospective
Req.
Req.
e.g., low risk in a specific situation?
e.g., good prediction performance for rare cases?
Reliable model
building with
small data
Controllable model update
for local improvement and
mitigation of degradation
Fine-Grained Requirements
for Dependability AI researchers and
SE researchers
Decrease oversight of
existing AI by 50% for
rare cases of cancers
Improve existing AI to
mitigate risks over
20+ fine-grained
safety metrics
Healthcare Automotive
“Engineerable AI” Project: Overview
“Engineerable AI” Project: Example of Techniques
Work with Fujitsu
[ Tokui+, NeuRecover: Regression-Controlled Repair of Deep Neural Networks with Training History,
SANER’22 ]
Target Neural Network
Analyze internal behavior regarding
occurrences of undesirable error pattern
e.g., misclassification of nearby pedestrian to rider
Identify and try to fix small part of neuron weight parameters
- that affected the occurrences of the error pattern
Also use hints by looking at past versions
- “past: success, now: fail” 🡪 try to fix relevant parameters
- “past: fail, now success” 🡪 not touch relevant parameters
Avoid side-effect of causing other errors or
“shuffling” of success/failure
caused by retraining or baseline method
Testing, Debugging, Analysis, Repairing Techniques and their Integration
into MLOps in a human-centered & Interactive Way
Continuous Quality Monitoring and Assurance of AI System &
AI System Trusthworty Technique Application across Diverse Domains
AI System Continuous Integration
& Continuous Delivery
Rule-based safeguard, with output space transformation
DNN
Policy
Filter
in Rn
Maruyama, Hiroshi. "Guaranteeing Deep Neural Network Outputs in a Feasible Region." Proceedings of the International Workshop on
Evidence-based Security and Privacy in the Wild and the 1st International Workshop on Machine Learning Systems Engineering. 2018.
Feasible Region
Non-feasible
solutions
in feasible region
Machine Learning Project Canvas
https://www.mitsubishichem-hd.co.jp/news_release/pdf/190718.pdf
Takeuchi, Hironori, et. al “Collecting Insights and Developing Patterns for Machine Learning Projects based on Project
Practices, 14th International Joint Conference on Knowledge-Based Software Engineering (JCKBSE)
Bad “smell” in
the project
37
ML system quality assurance guidelines in Japan
AIST, Machine Learning Quality Management Guideline, 2nd Edition,
https://www.digiarc.aist.go.jp/en/publication/aiqm/aiqm-guideline-en-2.1.1.0057-e26-
signed.pdf
Guideline for Quality Assurance of AI-based products and services
https://www.qa4ai.jp/download/
Guidelines on Assessment of AI Reliability in the Field of Plant Safety
https://www.meti.go.jp/english/press/2021/0330_001.html
The book “Machine Learning Engineering” by the community
1. What is machine learning systems engineering?
(Nakagawa, Ishikawa)
2. Project management of ML systems (Takeuchi)
3. Operation of ML systems (Horiuchi, Dobashi)
4. ML design patterns (Washizaki)
5. Quality assurance (Ishikawa)
6. Explainability of ML systems (Hara)
7. Ethics (Nakagawa)
8. Intellectual properties and contracts (Kakinuma)
9. Future of machine learning systems engineering
(Ishikawa)
ISBN-13 : 978-4065285862
Upcoming book on “Machine Learning Engineering for AI Project Managers”
1. Introduction to AI System Development (Yoshioka)
2. Requirement Engineering for AI Systems (Yoshioka)
3. Architecture and Design of ML Systems (Washizaki)
4. Project Management of AI Systems (Uchihira)
5. Cooperation with Stakeholders in AI Project (Takeuchi)
6. Future Vision of Machine Learning Engineering (Yoshioka)
Agenda
1. Backdrop
2. Launch of MLSE
3. Activities
4. Research Results
5. Retrospective
In five years, we …
● formed a very active community
○ From both SE and ML communities
○ From both industry and academia
○ Connecting people, sharing ideas, …
● produced research results
● helped the industry via many symposia and guideline documents, including the
quality assurance guidelines
● published a book (with more upcoming …)
Challenges ahead
● More academic activities as well as industry success stories
● Recognition as an engineering discipline by the general public
To me, perhaps this is the
biggest achievement
42
Thank you all
for making this movement possible
Twitter: @maruyama
https://sites.google.com/view/sig-mlse/en

More Related Content

Similar to 202212APSEC.pptx.pdf

Software Engineering Patterns for Machine Learning Applications
Software Engineering Patterns for Machine Learning ApplicationsSoftware Engineering Patterns for Machine Learning Applications
Software Engineering Patterns for Machine Learning Applications
Hironori Washizaki
 
Goal Dynamics_From System Dynamics to Implementation
Goal Dynamics_From System Dynamics to ImplementationGoal Dynamics_From System Dynamics to Implementation
Goal Dynamics_From System Dynamics to Implementation
Amjad Adib
 
Ph.D. Thesis: A Methodology for the Development of Autonomic and Cognitive In...
Ph.D. Thesis: A Methodology for the Development of Autonomic and Cognitive In...Ph.D. Thesis: A Methodology for the Development of Autonomic and Cognitive In...
Ph.D. Thesis: A Methodology for the Development of Autonomic and Cognitive In...
Universita della Calabria,
 
Strategy to build Beneficial Artificial General Intelligence inspired by the ...
Strategy to build Beneficial Artificial General Intelligence inspired by the ...Strategy to build Beneficial Artificial General Intelligence inspired by the ...
Strategy to build Beneficial Artificial General Intelligence inspired by the ...
The Whole Brain Architecture Initiative
 
Spohrer IESS 20230217 v9.pptx
Spohrer IESS 20230217 v9.pptxSpohrer IESS 20230217 v9.pptx
Spohrer IESS 20230217 v9.pptx
ISSIP
 
MK_MSc_Degree_Project_Report ver 5_updated
MK_MSc_Degree_Project_Report ver 5_updatedMK_MSc_Degree_Project_Report ver 5_updated
MK_MSc_Degree_Project_Report ver 5_updated
Mohammed Ali Khan
 
Ai open powermeetupmarch25th
Ai open powermeetupmarch25thAi open powermeetupmarch25th
Ai open powermeetupmarch25th
IBM
 
Ai open powermeetupmarch25th
Ai open powermeetupmarch25thAi open powermeetupmarch25th
Ai open powermeetupmarch25th
IBM
 
Ai open powermeetupmarch25th
Ai open powermeetupmarch25thAi open powermeetupmarch25th
Ai open powermeetupmarch25th
IBM
 
Data-X-Sparse-v2
Data-X-Sparse-v2Data-X-Sparse-v2
Data-X-Sparse-v2
Ikhlaq Sidhu
 
Personal Note On Software Engineering
Personal Note On Software EngineeringPersonal Note On Software Engineering
Personal Note On Software Engineering
Heidi Maestas
 
Machine Learning Software Engineering Patterns and Their Engineering
Machine Learning Software Engineering Patterns and Their EngineeringMachine Learning Software Engineering Patterns and Their Engineering
Machine Learning Software Engineering Patterns and Their Engineering
Hironori Washizaki
 
Stary2020_Chapter_TheInternet-of-BehaviorAsOrganRG.pdf
Stary2020_Chapter_TheInternet-of-BehaviorAsOrganRG.pdfStary2020_Chapter_TheInternet-of-BehaviorAsOrganRG.pdf
Stary2020_Chapter_TheInternet-of-BehaviorAsOrganRG.pdf
Hải Quân
 
ai_ml aicet internship report ppt 1.pptx
ai_ml aicet internship report ppt 1.pptxai_ml aicet internship report ppt 1.pptx
ai_ml aicet internship report ppt 1.pptx
SravyaSathi
 
Introduction to ICST 2017
Introduction to ICST 2017Introduction to ICST 2017
Introduction to ICST 2017
Keizo Tatsumi
 
Data-X-v3.1
Data-X-v3.1Data-X-v3.1
Data-X-v3.1
Ikhlaq Sidhu
 
MLSEV Virtual. ML Platformization and AutoML in the Enterprise
MLSEV Virtual. ML Platformization and AutoML in the EnterpriseMLSEV Virtual. ML Platformization and AutoML in the Enterprise
MLSEV Virtual. ML Platformization and AutoML in the Enterprise
BigML, Inc
 
A survey on Machine Learning and Artificial Neural Networks
A survey on Machine Learning and Artificial Neural NetworksA survey on Machine Learning and Artificial Neural Networks
A survey on Machine Learning and Artificial Neural Networks
IRJET Journal
 
Review on Algorithmic and Non Algorithmic Software Cost Estimation Techniques
Review on Algorithmic and Non Algorithmic Software Cost Estimation TechniquesReview on Algorithmic and Non Algorithmic Software Cost Estimation Techniques
Review on Algorithmic and Non Algorithmic Software Cost Estimation Techniques
ijtsrd
 
H1803044651
H1803044651H1803044651
H1803044651
IOSR Journals
 

Similar to 202212APSEC.pptx.pdf (20)

Software Engineering Patterns for Machine Learning Applications
Software Engineering Patterns for Machine Learning ApplicationsSoftware Engineering Patterns for Machine Learning Applications
Software Engineering Patterns for Machine Learning Applications
 
Goal Dynamics_From System Dynamics to Implementation
Goal Dynamics_From System Dynamics to ImplementationGoal Dynamics_From System Dynamics to Implementation
Goal Dynamics_From System Dynamics to Implementation
 
Ph.D. Thesis: A Methodology for the Development of Autonomic and Cognitive In...
Ph.D. Thesis: A Methodology for the Development of Autonomic and Cognitive In...Ph.D. Thesis: A Methodology for the Development of Autonomic and Cognitive In...
Ph.D. Thesis: A Methodology for the Development of Autonomic and Cognitive In...
 
Strategy to build Beneficial Artificial General Intelligence inspired by the ...
Strategy to build Beneficial Artificial General Intelligence inspired by the ...Strategy to build Beneficial Artificial General Intelligence inspired by the ...
Strategy to build Beneficial Artificial General Intelligence inspired by the ...
 
Spohrer IESS 20230217 v9.pptx
Spohrer IESS 20230217 v9.pptxSpohrer IESS 20230217 v9.pptx
Spohrer IESS 20230217 v9.pptx
 
MK_MSc_Degree_Project_Report ver 5_updated
MK_MSc_Degree_Project_Report ver 5_updatedMK_MSc_Degree_Project_Report ver 5_updated
MK_MSc_Degree_Project_Report ver 5_updated
 
Ai open powermeetupmarch25th
Ai open powermeetupmarch25thAi open powermeetupmarch25th
Ai open powermeetupmarch25th
 
Ai open powermeetupmarch25th
Ai open powermeetupmarch25thAi open powermeetupmarch25th
Ai open powermeetupmarch25th
 
Ai open powermeetupmarch25th
Ai open powermeetupmarch25thAi open powermeetupmarch25th
Ai open powermeetupmarch25th
 
Data-X-Sparse-v2
Data-X-Sparse-v2Data-X-Sparse-v2
Data-X-Sparse-v2
 
Personal Note On Software Engineering
Personal Note On Software EngineeringPersonal Note On Software Engineering
Personal Note On Software Engineering
 
Machine Learning Software Engineering Patterns and Their Engineering
Machine Learning Software Engineering Patterns and Their EngineeringMachine Learning Software Engineering Patterns and Their Engineering
Machine Learning Software Engineering Patterns and Their Engineering
 
Stary2020_Chapter_TheInternet-of-BehaviorAsOrganRG.pdf
Stary2020_Chapter_TheInternet-of-BehaviorAsOrganRG.pdfStary2020_Chapter_TheInternet-of-BehaviorAsOrganRG.pdf
Stary2020_Chapter_TheInternet-of-BehaviorAsOrganRG.pdf
 
ai_ml aicet internship report ppt 1.pptx
ai_ml aicet internship report ppt 1.pptxai_ml aicet internship report ppt 1.pptx
ai_ml aicet internship report ppt 1.pptx
 
Introduction to ICST 2017
Introduction to ICST 2017Introduction to ICST 2017
Introduction to ICST 2017
 
Data-X-v3.1
Data-X-v3.1Data-X-v3.1
Data-X-v3.1
 
MLSEV Virtual. ML Platformization and AutoML in the Enterprise
MLSEV Virtual. ML Platformization and AutoML in the EnterpriseMLSEV Virtual. ML Platformization and AutoML in the Enterprise
MLSEV Virtual. ML Platformization and AutoML in the Enterprise
 
A survey on Machine Learning and Artificial Neural Networks
A survey on Machine Learning and Artificial Neural NetworksA survey on Machine Learning and Artificial Neural Networks
A survey on Machine Learning and Artificial Neural Networks
 
Review on Algorithmic and Non Algorithmic Software Cost Estimation Techniques
Review on Algorithmic and Non Algorithmic Software Cost Estimation TechniquesReview on Algorithmic and Non Algorithmic Software Cost Estimation Techniques
Review on Algorithmic and Non Algorithmic Software Cost Estimation Techniques
 
H1803044651
H1803044651H1803044651
H1803044651
 

More from Hiroshi Maruyama

20230925プレジデント社60周年.pdf
20230925プレジデント社60周年.pdf20230925プレジデント社60周年.pdf
20230925プレジデント社60周年.pdf
Hiroshi Maruyama
 
20230912JSSST大会基調講演_丸山.pdf
20230912JSSST大会基調講演_丸山.pdf20230912JSSST大会基調講演_丸山.pdf
20230912JSSST大会基調講演_丸山.pdf
Hiroshi Maruyama
 
20230712Kuramae-Seminar.pdf
20230712Kuramae-Seminar.pdf20230712Kuramae-Seminar.pdf
20230712Kuramae-Seminar.pdf
Hiroshi Maruyama
 
20210731知財学会研究会
20210731知財学会研究会20210731知財学会研究会
20210731知財学会研究会
Hiroshi Maruyama
 
2021 06-17 ism-symposium
2021 06-17 ism-symposium2021 06-17 ism-symposium
2021 06-17 ism-symposium
Hiroshi Maruyama
 
Jsai
JsaiJsai
20181212 ibm aot
20181212 ibm aot20181212 ibm aot
20181212 ibm aot
Hiroshi Maruyama
 
20181205 sakigake
20181205 sakigake20181205 sakigake
20181205 sakigake
Hiroshi Maruyama
 
20181204i mlse discussions
20181204i mlse discussions20181204i mlse discussions
20181204i mlse discussions
Hiroshi Maruyama
 
20181204i mlse 1
20181204i mlse 120181204i mlse 1
20181204i mlse 1
Hiroshi Maruyama
 
20181120 ldp ai
20181120 ldp ai20181120 ldp ai
20181120 ldp ai
Hiroshi Maruyama
 
20181030 fun
20181030 fun20181030 fun
20181030 fun
Hiroshi Maruyama
 
20180719 cocn dist
20180719 cocn dist20180719 cocn dist
20180719 cocn dist
Hiroshi Maruyama
 
20180601 ai discussions
20180601 ai discussions20180601 ai discussions
20180601 ai discussions
Hiroshi Maruyama
 
構造改革徹底推進会合におけるプレゼン
構造改革徹底推進会合におけるプレゼン構造改革徹底推進会合におけるプレゼン
構造改革徹底推進会合におけるプレゼン
Hiroshi Maruyama
 
深層学習よもやま話
深層学習よもやま話深層学習よもやま話
深層学習よもやま話
Hiroshi Maruyama
 

More from Hiroshi Maruyama (16)

20230925プレジデント社60周年.pdf
20230925プレジデント社60周年.pdf20230925プレジデント社60周年.pdf
20230925プレジデント社60周年.pdf
 
20230912JSSST大会基調講演_丸山.pdf
20230912JSSST大会基調講演_丸山.pdf20230912JSSST大会基調講演_丸山.pdf
20230912JSSST大会基調講演_丸山.pdf
 
20230712Kuramae-Seminar.pdf
20230712Kuramae-Seminar.pdf20230712Kuramae-Seminar.pdf
20230712Kuramae-Seminar.pdf
 
20210731知財学会研究会
20210731知財学会研究会20210731知財学会研究会
20210731知財学会研究会
 
2021 06-17 ism-symposium
2021 06-17 ism-symposium2021 06-17 ism-symposium
2021 06-17 ism-symposium
 
Jsai
JsaiJsai
Jsai
 
20181212 ibm aot
20181212 ibm aot20181212 ibm aot
20181212 ibm aot
 
20181205 sakigake
20181205 sakigake20181205 sakigake
20181205 sakigake
 
20181204i mlse discussions
20181204i mlse discussions20181204i mlse discussions
20181204i mlse discussions
 
20181204i mlse 1
20181204i mlse 120181204i mlse 1
20181204i mlse 1
 
20181120 ldp ai
20181120 ldp ai20181120 ldp ai
20181120 ldp ai
 
20181030 fun
20181030 fun20181030 fun
20181030 fun
 
20180719 cocn dist
20180719 cocn dist20180719 cocn dist
20180719 cocn dist
 
20180601 ai discussions
20180601 ai discussions20180601 ai discussions
20180601 ai discussions
 
構造改革徹底推進会合におけるプレゼン
構造改革徹底推進会合におけるプレゼン構造改革徹底推進会合におけるプレゼン
構造改革徹底推進会合におけるプレゼン
 
深層学習よもやま話
深層学習よもやま話深層学習よもやま話
深層学習よもやま話
 

Recently uploaded

Skybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoptionSkybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoption
Tatiana Kojar
 
TrustArc Webinar - 2024 Global Privacy Survey
TrustArc Webinar - 2024 Global Privacy SurveyTrustArc Webinar - 2024 Global Privacy Survey
TrustArc Webinar - 2024 Global Privacy Survey
TrustArc
 
Driving Business Innovation: Latest Generative AI Advancements & Success Story
Driving Business Innovation: Latest Generative AI Advancements & Success StoryDriving Business Innovation: Latest Generative AI Advancements & Success Story
Driving Business Innovation: Latest Generative AI Advancements & Success Story
Safe Software
 
WeTestAthens: Postman's AI & Automation Techniques
WeTestAthens: Postman's AI & Automation TechniquesWeTestAthens: Postman's AI & Automation Techniques
WeTestAthens: Postman's AI & Automation Techniques
Postman
 
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
Alex Pruden
 
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
alexjohnson7307
 
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdfHow to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
Chart Kalyan
 
FREE A4 Cyber Security Awareness Posters-Social Engineering part 3
FREE A4 Cyber Security Awareness  Posters-Social Engineering part 3FREE A4 Cyber Security Awareness  Posters-Social Engineering part 3
FREE A4 Cyber Security Awareness Posters-Social Engineering part 3
Data Hops
 
A Comprehensive Guide to DeFi Development Services in 2024
A Comprehensive Guide to DeFi Development Services in 2024A Comprehensive Guide to DeFi Development Services in 2024
A Comprehensive Guide to DeFi Development Services in 2024
Intelisync
 
Taking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdfTaking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdf
ssuserfac0301
 
Azure API Management to expose backend services securely
Azure API Management to expose backend services securelyAzure API Management to expose backend services securely
Azure API Management to expose backend services securely
Dinusha Kumarasiri
 
GraphRAG for Life Science to increase LLM accuracy
GraphRAG for Life Science to increase LLM accuracyGraphRAG for Life Science to increase LLM accuracy
GraphRAG for Life Science to increase LLM accuracy
Tomaz Bratanic
 
Digital Marketing Trends in 2024 | Guide for Staying Ahead
Digital Marketing Trends in 2024 | Guide for Staying AheadDigital Marketing Trends in 2024 | Guide for Staying Ahead
Digital Marketing Trends in 2024 | Guide for Staying Ahead
Wask
 
Public CyberSecurity Awareness Presentation 2024.pptx
Public CyberSecurity Awareness Presentation 2024.pptxPublic CyberSecurity Awareness Presentation 2024.pptx
Public CyberSecurity Awareness Presentation 2024.pptx
marufrahmanstratejm
 
Astute Business Solutions | Oracle Cloud Partner |
Astute Business Solutions | Oracle Cloud Partner |Astute Business Solutions | Oracle Cloud Partner |
Astute Business Solutions | Oracle Cloud Partner |
AstuteBusiness
 
System Design Case Study: Building a Scalable E-Commerce Platform - Hiike
System Design Case Study: Building a Scalable E-Commerce Platform - HiikeSystem Design Case Study: Building a Scalable E-Commerce Platform - Hiike
System Design Case Study: Building a Scalable E-Commerce Platform - Hiike
Hiike
 
Skybuffer AI: Advanced Conversational and Generative AI Solution on SAP Busin...
Skybuffer AI: Advanced Conversational and Generative AI Solution on SAP Busin...Skybuffer AI: Advanced Conversational and Generative AI Solution on SAP Busin...
Skybuffer AI: Advanced Conversational and Generative AI Solution on SAP Busin...
Tatiana Kojar
 
Monitoring and Managing Anomaly Detection on OpenShift.pdf
Monitoring and Managing Anomaly Detection on OpenShift.pdfMonitoring and Managing Anomaly Detection on OpenShift.pdf
Monitoring and Managing Anomaly Detection on OpenShift.pdf
Tosin Akinosho
 
5th LF Energy Power Grid Model Meet-up Slides
5th LF Energy Power Grid Model Meet-up Slides5th LF Energy Power Grid Model Meet-up Slides
5th LF Energy Power Grid Model Meet-up Slides
DanBrown980551
 
Fueling AI with Great Data with Airbyte Webinar
Fueling AI with Great Data with Airbyte WebinarFueling AI with Great Data with Airbyte Webinar
Fueling AI with Great Data with Airbyte Webinar
Zilliz
 

Recently uploaded (20)

Skybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoptionSkybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoption
 
TrustArc Webinar - 2024 Global Privacy Survey
TrustArc Webinar - 2024 Global Privacy SurveyTrustArc Webinar - 2024 Global Privacy Survey
TrustArc Webinar - 2024 Global Privacy Survey
 
Driving Business Innovation: Latest Generative AI Advancements & Success Story
Driving Business Innovation: Latest Generative AI Advancements & Success StoryDriving Business Innovation: Latest Generative AI Advancements & Success Story
Driving Business Innovation: Latest Generative AI Advancements & Success Story
 
WeTestAthens: Postman's AI & Automation Techniques
WeTestAthens: Postman's AI & Automation TechniquesWeTestAthens: Postman's AI & Automation Techniques
WeTestAthens: Postman's AI & Automation Techniques
 
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
 
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
 
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdfHow to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
 
FREE A4 Cyber Security Awareness Posters-Social Engineering part 3
FREE A4 Cyber Security Awareness  Posters-Social Engineering part 3FREE A4 Cyber Security Awareness  Posters-Social Engineering part 3
FREE A4 Cyber Security Awareness Posters-Social Engineering part 3
 
A Comprehensive Guide to DeFi Development Services in 2024
A Comprehensive Guide to DeFi Development Services in 2024A Comprehensive Guide to DeFi Development Services in 2024
A Comprehensive Guide to DeFi Development Services in 2024
 
Taking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdfTaking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdf
 
Azure API Management to expose backend services securely
Azure API Management to expose backend services securelyAzure API Management to expose backend services securely
Azure API Management to expose backend services securely
 
GraphRAG for Life Science to increase LLM accuracy
GraphRAG for Life Science to increase LLM accuracyGraphRAG for Life Science to increase LLM accuracy
GraphRAG for Life Science to increase LLM accuracy
 
Digital Marketing Trends in 2024 | Guide for Staying Ahead
Digital Marketing Trends in 2024 | Guide for Staying AheadDigital Marketing Trends in 2024 | Guide for Staying Ahead
Digital Marketing Trends in 2024 | Guide for Staying Ahead
 
Public CyberSecurity Awareness Presentation 2024.pptx
Public CyberSecurity Awareness Presentation 2024.pptxPublic CyberSecurity Awareness Presentation 2024.pptx
Public CyberSecurity Awareness Presentation 2024.pptx
 
Astute Business Solutions | Oracle Cloud Partner |
Astute Business Solutions | Oracle Cloud Partner |Astute Business Solutions | Oracle Cloud Partner |
Astute Business Solutions | Oracle Cloud Partner |
 
System Design Case Study: Building a Scalable E-Commerce Platform - Hiike
System Design Case Study: Building a Scalable E-Commerce Platform - HiikeSystem Design Case Study: Building a Scalable E-Commerce Platform - Hiike
System Design Case Study: Building a Scalable E-Commerce Platform - Hiike
 
Skybuffer AI: Advanced Conversational and Generative AI Solution on SAP Busin...
Skybuffer AI: Advanced Conversational and Generative AI Solution on SAP Busin...Skybuffer AI: Advanced Conversational and Generative AI Solution on SAP Busin...
Skybuffer AI: Advanced Conversational and Generative AI Solution on SAP Busin...
 
Monitoring and Managing Anomaly Detection on OpenShift.pdf
Monitoring and Managing Anomaly Detection on OpenShift.pdfMonitoring and Managing Anomaly Detection on OpenShift.pdf
Monitoring and Managing Anomaly Detection on OpenShift.pdf
 
5th LF Energy Power Grid Model Meet-up Slides
5th LF Energy Power Grid Model Meet-up Slides5th LF Energy Power Grid Model Meet-up Slides
5th LF Energy Power Grid Model Meet-up Slides
 
Fueling AI with Great Data with Airbyte Webinar
Fueling AI with Great Data with Airbyte WebinarFueling AI with Great Data with Airbyte Webinar
Fueling AI with Great Data with Airbyte Webinar
 

202212APSEC.pptx.pdf

  • 1. Machine Learning Systems Engineering (MLSE): Retrospective of Five-Year Activities in Japan Hiroshi Maruyama* Preferred Networks / Kao Corporation / U. Tokyo Twitter; @maruyama * Presentation is done by me but this is a collective effort by the MLSE members
  • 2. Agenda 1. Backdrop 2. Launch of MLSE 3. Activities 4. Research Results 5. Retrospective
  • 3. Googleʼs seminal paper on “technical debt” of ML systems (2015)
  • 4. Workshop on “Towards real world implementation of ML systems”, in conjunction with JSAI annual convention, 2016 ● How to deploy ML systems for industry applications ● Recent trends in ML businesses ● Intellectual property in ML systems ● Nikoniko-Deep Learning β ● Artificial life hackason, tried ● Evolution of communities, such as open source Business Community
  • 6. Software 2.0: Find a program rather than write it by hand Find a program Set of programs that satisfy the spec Search algorithm Write a program vs x x 🡺 Software 2.0 requires completely different set of skills
  • 7. Concerns on the shortage of ML skills Source: METI, IT人材の最新動向と将来推計に関 する調査結果 (Study results on the latest status and the future trends of IT talents), 2016 Shortage of 48,000 talents on “big data, IOT, and AI” in 2020
  • 8. 8 “Shortage of skills” -- doesn’t this sound familiar? Who can write software ⇒ Software Crisis (1960’s) 🡺 Dawn of Software Engineering! IBM System 360 Source: Wikipedia System 360 Instruction set Source:Quora https://www.quora.com/How-did-you-learn-an-assembly-language-and-which-one
  • 9. Keynote at APRES (Asia-Pacific Requirements Engineering Symposium) by Maruyama (2016)
  • 10. My email to Mikio Aoyama “We have a number of projects going on with customers. Sometimes I feel that we are reinventing software engineering practices in ML. I wonder if the SE community can help us to establish a new engineering discipline in ML”
  • 11. Mikio’s email to the organizer of SES2017, suggesting a panel discussion on SE for ML Annual SES (Software Engineering Symposium) is the largest event in Japan dedicated to Software Engineering, and Mikio suggested this is the best place to draw the interests from the SE community
  • 12. Agenda 1. Backdrop 2. Launch of MLSE 3. Activities 4. Research Results 5. Retrospective
  • 13. Panel discussion on machine learning engineering at SES 2017 “This panel discussion was the highlight of this year’s SES” Panelists ● Fuyuki Ishikawa (NII) ● Koichi Hamada (DeNA) ● Hiroshi Maruyama (PFN) Moderator ● Mikio Aoyama INanzan U.) https://www.facebook.com/bonotake/posts/1504893489556668
  • 14. A couple of meetups among SE and ML engineers revealed new gaps ● ML engineer (ME): “my improved model now gives an erroneous output for a certain input that was ok with my previous model” ● Software engineer (SE): “What did you do with your regression test? Don’t you have one? ● ME: “...” (ah, but what does regression test mean in ML? How can we do it?) ● ME: “My customer is concerned with the safety” ● SE: “What is the invariant in your code?” ● ME: “...” (Invariant? In an ML system?) ● SE: “You are concerned with the quality. Why don’t you use stronger-typed language than Python?” ● ME: “...” (yes, I wish Python could statically check the shape of numpy ndarray)
  • 15. Sorrow of ML Project: “Curse of infinite PoC” Develop model Evaluate Yes, we achieved xx% accuracy! Can you make a little better? ML engineer Customer How can we have a reasonable level of customer expectation? Looks good! But not enough for my customer
  • 16. Michael Jordanʼs blog on the need for new engineering discipline https://medium.com/@mijordan3/artificial-intelligence-the-revoluti on-hasnt-happened-yet-5e1d5812e1e7 “... we do have a major challenge on our hands in bringing together computers and humans in ways that enhance human life. While this challenge is viewed by some as subservient to the creation of “artificial intelligence,” it can also be viewed more prosaically — but with no less reverence — as the creation of a new branch of engineering. Much like civil engineering and chemical engineering in decades past, this new discipline aims to corral the power of a few key ideas, bringing new resources and capabilities to people, and doing so safely. “
  • 17. The role of engineering -- my personal view Theories * Safety Factor Engineering as a form of agreement between engineers and the society Civil Engineering Handbook, p999 Why do we trust bridges? Because of the accumulated knowledge called Civil Engineering
  • 18. In Apr. 2018, the SIG on MLSE (Machine Learning Systems Engineering, pronounced as “Mel-See”) is formed under JSSST https://mlxse.connpass.com/
  • 19. Agenda 1. Backdrop 2. Launch of MLSE 3. Activities 4. Research Results 5. Retrospective
  • 20. MLSE kick-off meeting, Mar. 2018 (>500 participants) Source: https://ledge.ai/mlse-symposium/ ● Mikio Aoyama (Nanzan U), “Expectations to MLSE” ● Takuya Kudo (Accenture), “Challenges in software engineering and the new form of ML” ● Masashi Sugiyama (Riken AIP), “Current and future of ML research” ● Akimichi Ariga (Cloudara), “ML starting from business applications” ● Takahiro Kubo (TIS), “ML code design without remorse” ● Shin Nakajima (NII), “Quality assurance of ML software” Common issues in ML-in-practice have surfaced See https://leapmind.io/blog/2018/06/12/mlsekickoff/ for a report on the symposium
  • 21. First things first: What are the challenges of MLSE? Ishikawa, Fuyuki, and Nobukazu Yoshioka. "How do engineers perceive difficulties in engineering of machine-learning systems?-questionnaire survey." 2019 IEEE/ACM Joint 7th International Workshop on Conducting Empirical Studies in Industry (CESI) and 6th International Workshop on Software Engineering Research and Industrial Practice (SER&IP). IEEE, 2019.
  • 22. Working groups ● ML operational infrastructures / operations WG ● ML fairness WG ● ML system foundation WG ● Data quality engineering WG ● ML operations WG ● ML development process and case studies WG ● ML security WG ● : Active discussions on Discord, everybody is welcome
  • 23. MLSE summer camps ● Main venue for community discussions ● 2-3 days, plenary / parallel sessions (workshops) + posters ● WG’s to report annual findings ● Every year new ideas are coming out Online due to Covid-19 2019 venue in Hakone hot spa
  • 24. Cooperative gatherings Symposium on safety for AI/IoT systems Symposium on ML and fairness w/ JSAI and IBIS/ML, Jan., 2020
  • 25. International activities -- 1/2 ● iMLSE -- International Workshop on Machine Learning Systems Engineering, in conjunction with APSEC 1st iMLSE in Nara (2018) iMLSE 2020 (online) Clark Barrett Jacomo Corbo iMLSE 2021 (online) Hironori Washizaki ● Shonan meeting, Nov. 2019
  • 26. ● MLSE International Symposium (2019) ● Sanjit Seshua, “Towards Verified Artificial Intelligence” ● Akira Sakakibara, “Engineer's Responsibility in Machine Learning Era” ● Foutse Khomh, “Towards Debugging and Testing Deep Learning Systems” ● Lei Ma, “ Towards Testing and Analysis of Deep Learning Systems” ● Amel Bennaceur, “Requirements for Machine Learning Applications” ● Rüdiger Ehlers, “The Role of Verification in the Engineering Process of Complex Cyber-Physical Systems That Employ Machine Learning” International activities -- 2/2
  • 27. Agenda 1. Backdrop 2. Launch of MLSE 3. Activities 4. Research Results 5. Retrospective
  • 28. Req. Req. e.g., low risk in a specific situation? e.g., good prediction performance for rare cases? Reliable model building with small data Controllable model update for local improvement and mitigation of degradation Fine-Grained Requirements for Dependability AI researchers and SE researchers Decrease oversight of existing AI by 50% for rare cases of cancers Improve existing AI to mitigate risks over 20+ fine-grained safety metrics Healthcare Automotive “Engineerable AI” Project: Overview
  • 29. “Engineerable AI” Project: Example of Techniques Work with Fujitsu [ Tokui+, NeuRecover: Regression-Controlled Repair of Deep Neural Networks with Training History, SANER’22 ] Target Neural Network Analyze internal behavior regarding occurrences of undesirable error pattern e.g., misclassification of nearby pedestrian to rider Identify and try to fix small part of neuron weight parameters - that affected the occurrences of the error pattern Also use hints by looking at past versions - “past: success, now: fail” 🡪 try to fix relevant parameters - “past: fail, now success” 🡪 not touch relevant parameters Avoid side-effect of causing other errors or “shuffling” of success/failure caused by retraining or baseline method
  • 30.
  • 31. Testing, Debugging, Analysis, Repairing Techniques and their Integration into MLOps in a human-centered & Interactive Way
  • 32. Continuous Quality Monitoring and Assurance of AI System & AI System Trusthworty Technique Application across Diverse Domains AI System Continuous Integration & Continuous Delivery
  • 33.
  • 34. Rule-based safeguard, with output space transformation DNN Policy Filter in Rn Maruyama, Hiroshi. "Guaranteeing Deep Neural Network Outputs in a Feasible Region." Proceedings of the International Workshop on Evidence-based Security and Privacy in the Wild and the 1st International Workshop on Machine Learning Systems Engineering. 2018. Feasible Region Non-feasible solutions in feasible region
  • 35. Machine Learning Project Canvas https://www.mitsubishichem-hd.co.jp/news_release/pdf/190718.pdf
  • 36. Takeuchi, Hironori, et. al “Collecting Insights and Developing Patterns for Machine Learning Projects based on Project Practices, 14th International Joint Conference on Knowledge-Based Software Engineering (JCKBSE) Bad “smell” in the project
  • 37. 37 ML system quality assurance guidelines in Japan AIST, Machine Learning Quality Management Guideline, 2nd Edition, https://www.digiarc.aist.go.jp/en/publication/aiqm/aiqm-guideline-en-2.1.1.0057-e26- signed.pdf Guideline for Quality Assurance of AI-based products and services https://www.qa4ai.jp/download/ Guidelines on Assessment of AI Reliability in the Field of Plant Safety https://www.meti.go.jp/english/press/2021/0330_001.html
  • 38. The book “Machine Learning Engineering” by the community 1. What is machine learning systems engineering? (Nakagawa, Ishikawa) 2. Project management of ML systems (Takeuchi) 3. Operation of ML systems (Horiuchi, Dobashi) 4. ML design patterns (Washizaki) 5. Quality assurance (Ishikawa) 6. Explainability of ML systems (Hara) 7. Ethics (Nakagawa) 8. Intellectual properties and contracts (Kakinuma) 9. Future of machine learning systems engineering (Ishikawa) ISBN-13 : 978-4065285862
  • 39. Upcoming book on “Machine Learning Engineering for AI Project Managers” 1. Introduction to AI System Development (Yoshioka) 2. Requirement Engineering for AI Systems (Yoshioka) 3. Architecture and Design of ML Systems (Washizaki) 4. Project Management of AI Systems (Uchihira) 5. Cooperation with Stakeholders in AI Project (Takeuchi) 6. Future Vision of Machine Learning Engineering (Yoshioka)
  • 40. Agenda 1. Backdrop 2. Launch of MLSE 3. Activities 4. Research Results 5. Retrospective
  • 41. In five years, we … ● formed a very active community ○ From both SE and ML communities ○ From both industry and academia ○ Connecting people, sharing ideas, … ● produced research results ● helped the industry via many symposia and guideline documents, including the quality assurance guidelines ● published a book (with more upcoming …) Challenges ahead ● More academic activities as well as industry success stories ● Recognition as an engineering discipline by the general public To me, perhaps this is the biggest achievement
  • 42. 42 Thank you all for making this movement possible Twitter: @maruyama https://sites.google.com/view/sig-mlse/en