ai-fml international academy ai人機共學國際學院
TRANSCRIPT
AI-FML International Academy
AI人機共學國際學院Supported by
International Fuzzy Systems Association (IFSA)
IEEE CIS Task Force on Fuzzy Systems for Web Intelligence
IEEE CIS Task Force on Competitions
IEEE CIS Task Force on Fuzzy Systems Software
Chang-Shing Lee
National University of Tainan, Taiwan
Dec. 28, 2019
• Goal– Promote IEEE 1855-2016 Standard FML.
– Propose FML tools for real-world applications.
– Propose FML learning curriculums for high-
school students and learners.
– Promote high-school students around the world
to combine FML tools with machine learning for
real-world AI applications.
– Apply FML to real-world robotic applications.
– Apply FML to AIOT Applications
– AI-FML as a Service, AI-FaaS/ 541
AI-FML International Academy
Current candidate places for this event are as follows:– @ Taiwan: Tainan, Kaohsiung, …
– @ Japan: Tokyo, Osaka, …
– @ Italy: Napoli, …
– @ Canada: Edmonton, …
– @ Spain: Santiago de Compostela, Cordoba, Granada, …
– @ France: Paris, …
– @ UK: Nottingham, …
– …
/ 542
AI-FML International Academy
• Organizing Committee
– President: Hung-Duen Yang, Taiwan
– Vice Presidents for AI-FML Tools
– Vice Presidents for AI-FML Applications
/ 543
Chang-Shing Lee
TaiwanMarek Reformat
Canada
Giovanni Acampora
Italy
Jose M. Alonso
Spain
Toru Yamaguchi
Japan
Po-Hsun Cheng
Taiwan
Yusuke Nojima
Japan
Naoyuki Kubota
Japan
Marie-Jeanne Lesot
France
Amir Pourabdollah
UK
AI-FML International Academy
Jose Manuel Soto Hidalgo
Spain
/ 544
AI-FML International Academy
/ 545
AI-FML Applications
Co-Learning on Game of Go
Co-Learning on LanguageCo-Learning on Math
AI-FML International Academy
Chang-Shing Lee
National University of Tainan, Taiwan
WebDuino / Kebbi Air AI Robot
2019/12/28
AI-FML for AIoT Applications
AI-FML International Academy
Robotic Applications (Kebbi)
/ 547
http://www.uco.es/JFML/
This library offers a complete implementation of the four FLS enclosed in the W3C XML Schema definition (XSD) of the standard IEEE 1855 for FML:
Mamdani Takagi-Sugeno-Kang (TSK)
JFML: the first Java Library complying with IEEE 1855 Standard for FML
Tsukamoto AnYa
JFML makes use of the new Java API JAXB (Java Architecture for XML Binding) to bind W3C XML
schemas and Java representations, making it easy for Java developers to incorporate XML data and
processing functions in Java applications
JFML includes classes and methods in order to facilitate developers to import/export FLS defined with
• The standard IEC 61131-7
• PMML
• Matlab
JFML is now also accessible in Python 3.x through Py4JFML
Fuzzy-as-a-Service (FaaS)
Problem: Fuzzy Logic Systems are usually associated with dedicated hardware/software.
Stand-alone or simple-networked, desktop computing or embedded
Bottleneck: Complex FLS computations, particularly for ambient devices
Solution: Service-Oriented Architecture (SOA)
Abstraction of processing logic from presentation and data
Openness, reusability, elasticity and performance
FML as data exchange language
JFML as processing logic
Pourabdollah, A., Wagner, C., Acampora, G. and Lotfi, A., 2018, July. Fuzzy Logic As-a-
Service for Ambient Intelligence Environments. In 2018 IEEE International Conference
on Fuzzy Systems (FUZZ-IEEE) (pp. 1-7). IEEE.
/ 54 9
Human Monitoring System using FML/JFML
/ 54 10
Contacts
Nottingham Trent University, UK.
Dr Amir Pourabdollah
Prof. Ahmad Lotfi
Bhavesh Pandya
/ 54 11
Web:Bit AI-FML Blocks/ AIoT
/ 5412
• By surfing on the web, you will find a
template of blocks below.
Web:Bit AI-FML Blocks/ AIoT
/ 5413
This is the ID of the robot Kebbi.
Define your topic name
based on MQTT protocol
Execute the actions that you want a robot Kebbi to do
using blocks provided by Web:Bit
Web:Bit AI-FML Blocks/ AIoT
• The followings are available blocks on the Web:Bit website
• FML-related blocks include logic (邏輯), math (數學), expansion
(擴充功能), and Kebbi (凱比機器人)
/ 5414
Web:Bit Logic Blocks/ AIoT
/ 5415
• Select block logic(邏輯)
If(如果)…Do(執行)
• Pull block “If …Do”
into the blocks
Web:Bit Logic Blocks/ AIoT
• Click on a gear icon.
Some blocks will be appeared.
• Select block elseif (否則如果) and move it to the
place below block “if (如果)”.
• A new block will be added.
/ 5416
Web:Bit Logic Blocks/ AIoT
• Take Tips Payment for an example. There
are 3 output fuzzy variables.
• Choose blocks elseif (否則如果) and if (否則)
and move them to suitable places.
/ 5417
低 高 中
elseif
else
Web:Bit Logic Blocks/ AIoT
• Select the second block from block logic (邏輯)
• Pull it into the block,
click on the drop-down menu, and
choose icon “<”
/ 5418
Web:Bit Expansion Blocks/ AIoT
• Select block expansion(擴充功能), broadcast(網路廣播), and then received broadcast(收到的廣播訊息)
• Pull it into block
/ 5419
Broadcast (網路廣播)
Web:Bit Math Blocks/ AIoT
• Select the first block from block math(數學)
• Pull it into block
/ 5420
Received broadcast message (收到的廣播訊息)
Web:Bit Math Blocks/ AIoT
• The first 「if」(如果) denotes “< 3.5” is 低(low).
• Fill 3.5 in block math (數學).
/ 5421
3.5
FML knowledge base
& rule base
Web:Bit Math Blocks/ AIoT
• The second block 「else if」(否則如果)
denotes that “<6.5” is 中(medium).
• Repeat previous action and fill 6.5 in block
math (數學).
/ 5422
6.5
FML knowledge base
& rule base
Web:Bit AI-FML Blocks/ AIoT
• We design FML Kebbi using blocks do, else if,
and else.
• Block expansion(擴充功能) for robot Kebbi
/ 5423
• In the first block “do”, we use specific Kebbi
blocks including blocks light (燈光), voice (聲音),
and animation (動畫表演), to control the robot
to action when the output fuzzy variable is
low(低).
/ 5424
Web:Bit AI-FML Blocks/ AIoT
• In the second block “do” , we use blocks light, voice
and animation to control the robot to action when
the output fuzzy variable is medium(中).
• In the second block “else” , we use blocks light,
voice and animation to control the robot to action
when the output fuzzy variable is high(高)
/ 5425
Web:Bit AI-FML Blocks/ AIoT
Web:Bit AI-FML Robot/ AIoT
/ 5426
AI-FML Robot
Web:Bit AI-FML Robot/ AIoT
/ 5427
AI-FML Robot
Appendix I
Real-World Applications
AI-FML International Academy
/ 5429
1. Diet/ Healthcare/ Travel• (1) Adaptive personalized diet linguistic recommendation mechanism based on type-2 fuzzy sets
and genetic fuzzy markup language, IEEE Trans. on Fuzzy Systems, vol. 23, no. 5, pp. 1777-1802,
2015.
• (2) Healthy diet assessment mechanism based on fuzzy markup language for Japanese food, Soft
Computing, vol. 20, no 1, pp 359-376, 2016.
• (3) A novel genetic fuzzy markup language and its application to healthy diet assessment,
International Journal of Uncertainty, Fuzziness, and Knowledge-Based Systems, vol. 20, no. 2, pp.
247-278, 2012.
• (4) Evaluating cardiac health through semantic soft computing techniques, Soft Computing, vol.16,
no. 7, pp. 1165-1181, 2012.
• (5) Diet assessment based on type-2 fuzzy ontology and fuzzy markup language, International
Journal of Intelligent System, vol. 25, no. 12, pp. 1187-1216, 2010.
• (6) Ontology-based multi-agents for intelligent healthcare applications, Journal of Ambient
Intelligence and Humanized Computing, vol. 1, no. 2, pp. 111-131, 2010.
AI-FML International Academy
/ 5430
2. E-Learning/ Education/ IRT/ Ontology Construction• (1) Performance Verification Mechanism for Adaptive Assessment e-Platform and e-Navigation
Application, International Journal of e-Navigation and Maritime Economy, vol. 2, pp. 47-62, 2015.
• (2) T2FS-based adaptive linguistic assessment system for semantic analysis and human
performance evaluation on game of Go, IEEE Trans. on Fuzzy Systems, vol. 23, no. 2, pp. 400-
420, 2015.
3. Game/ Go• (1) T2FS-based adaptive linguistic assessment system for semantic analysis and human
performance evaluation on game of Go, IEEE Trans. on Fuzzy Systems, vol. 23, no. 2, pp.
400-420, 2015.
• (2) Soft-Computing-based emotional expression mechanism for game of Computer Go,
Soft Computing, vol. 17, no. 7, pp. 1263-1282, 2013.
• (3) Genetic fuzzy markup language for game of NoGo, Knowledge-Based Systems, vol. 34,
pp. 64- 80, 2012.
• (4) An ontology-based fuzzy inference system for computer Go applications, International
Journal of Fuzzy Systems, vol. 12, no. 2, pp. 103-115, 2010.
AI-FML International Academy
/ 5431
4. Energy Management • (1) An optimization model for FML-based decision support system on energy management,
in Proceeding of 2014 IEEE International Conference on Fuzzy Systems, Beijing, China, 6-11, 2014, pp. 850-856.
• (2) FML-based decision support system for solar energy supply and demand analysis, 2013 IEEE International Conference on Fuzzy Systems, Hyderabad, India, 7-10, 2013.
5. Patent Evaluation• (1) Fuzzy markup language with genetic learning mechanism for invention patent quality
evaluation, in Proceeding of 2015 IEEE Congress on Evolutionary Computation, Sendai, Japan, 25-28, 2015, pp. 251-258.
6. Information Security• (1) IT2FS-based ontology with soft-computing mechanism for malware behavior analysis, Soft
Computing, vol. 18, no. 2, pp. 267-284, 2014.
7. University Assessment• (1) Apply fuzzy ontology and FML to knowledge extraction for university governance and
management, Journal of Ambient Intelligence and Humanized Computing, vol. 4, no. 4, pp. 493-513, 2013.
AI-FML International Academy
Appendix II
FML-based Machine Learning
Competition
AI-FML International Academy
Competition @ FUZZ-IEEE 2019FML-based Machine Learning Competition
for Human and Smart Machine Co-Learning
on Game of Go
Chang-Shing lee, Yusuke Nojima, Naoyuki Kubota
Giovanni Acampora, Marek Reformat, and Ryosuke Saga
Presenter: Chang-Shing Lee
July 16, 2019
Outline
• Organizers
• Scope and Topic
• Competition Data
• Metrics and Rules
• Evaluation
• Results
• Next Steps: AI-FML International Academy
34 / 54
Organizers
/ 5435
• Organizers and Poster
• Competition Websitehttp://oase.nutn.edu.tw/fuzz2019-fmlcompetition/
Outline
• Organizers
• Scope and Topic
• Competition Data
• Metrics and Rules
• Evaluation
• Results
36 / 54
Scope and Topic
• Use the FML tool to construct the knowledge base (KB) and
rule base (RB) of the fuzzy inference system.
• Use the data predicted by Facebook AI Research (FAIR)
Darkforest AI Bot as the training data.
• Use the data predicted by FAIR ELF OpenGo as the
Desired Output for the training data and the testing data.
• Optimize the FML KB and RB through the methodologies of
evolutionary computation and machine learning in order to
develop a regression model based on FML-based fuzzy
inference system.
/ 5437
1855-2016IEEE Standard for Fuzzy Markup Language
/ 5438
Outline
• Organizers
• Scope and Topic
• Competition Data
• Metrics and Rules
• Evaluation
• Results
39 / 54
Competition Data-Google AlphaGo Master 60 Games
• Google AlphaGo Master Series
– Google AlphaGo Master in Dec. 2016/ Jan. 2017
– 60 online fast time-control games with top
professional Go players.
– AlphaGo Master won 60 games.
– Google DeepMind Website:
https://deepmind.com/research/alphago/match-
archive/master/
/ 5440
Competition Data-Google AlphaGo Master 60 Games
/ 5441
Competition Data-FAIR ELF OpenGo AI bot Prediction
FAIR AI bot
Prediction Data
Game 1
/ 5442
1. DBSN, DWSN, DBWR, DWWR, DBTMR, and DWTMR
were predicted by FAIR Darkforest AI bot.
2. EBWR and EWWR were predicted by ELF OpenGo AI bot.
3. WR: Win Rate; SN: MCTS Simulation Number.
Competition Data
• Training Data: Game 1 to Game 45
– Input: MCTS Simulation Number (DBSN, DWSN), Win
Rate (DBWR, DWWR), and Top-Move Rate (DBTMR,
DWTMR) predicted by Darkforest (NUTN, Taiwan/OPU,
Japan).
– Desired Output: Win Rate (EBWR, EWWR) predicted by
ELF OpenGo (NUTN, Taiwan/OPU, Japan).
• Testing Data: Game 46 to Game 60
– Examine the generalization ability of the learned FML-
based fuzzy inference system.
/ 5443
Outline
• Organizers
• Scope and Topic
• Competition Data
• Metrics and Rules
• Evaluation
• Results
44 / 54
Metrics and Rules
• Mean Square Error (MSE)
/ 5445
Outline
• Organizers
• Scope and Topic
• Competition Data
• Metrics and Rules
• Evaluation
• Results
46 / 54
Evaluation
• FML KB and RB: 30%
• Learned KB and RB, Training Data Accuracy,
Testing Data Accuracy: 35%
• Evolutionary Computation & Machine
Learning PPT Slide: 35%
/ 5447
Outline
• Organizers
• Scope and Topic
• Competition Data
• Metrics and Rules
• Evaluation
• Results
48 / 54
Entries and Winners
• 14 entries registration
• 7 entries finish the competition
/ 5449
Video Presentation
• Video Presentation on Competition Website– Team CILAB_OPU (Osaka Prefecture University, Japan)
– Team OASEWIFI (National University of Tainan, Taiwan)
– Team TMU_Y&K (Tokyo Metropolitan University, Japan)
– Team WeLiveonTop (National Tainan First Senior High School/National Fengshan Senior High School, Taiwan)
– Team NKNU SE+ (National Kaohsiung Normal University, Taiwan)
– Team Milos (National Tainan First Senior High School, Taiwan)
– Team Taiwan Fish (Tsoying Senior High School, Taiwan)
/ 5450
/ 5451
/ 5452
A Series of FML-based Machine Learning
Competition Activities and Promotion in
Taiwan from 2018 to 2019
• Website https://youtu.be/f7BfQA-FMRQ
/ 5453
/ 5454
AI-FML International Academy