Slide 1
Slide 1 text
No content
Slide 2
Slide 2 text
> > Marco Chen
> > Senior Technical Director, LINE Taiwan Limited
LINE TAIWAN
TECHPULSE 2019 Keynote
Slide 3
Slide 3 text
2016 2019
2017 2018
10 Times
Number of LINE Taiwan engineers has increased by 10 times since 2016
4th LINE TAIWAN TECHPULSE
Slide 4
Slide 4 text
More Technical
Insight
> Almost 30 LINE Taiwan engineers
shared technical insight in
DevDay 2019, Japan
Slide 5
Slide 5 text
Life on LINE
> > Connect with LINE Platform
> > Natural Experience with AI Technology
Slide 6
Slide 6 text
LINE SPOT
Finding restaurants
nearby and make
reservation
Life on LINE
Featuring services for every part of your day
Slide 7
Slide 7 text
TAIWAN
JAPAN
VIETNAM
INDONESIA
MENA
Conomi
͠ͽͧ͡ NOW
SHOPPING GO LINEϪξη
GET IT
LINE SHOPPING
SWAY
LIVE MENA
Slide 8
Slide 8 text
Connect with LINE Platform
Slide 9
Slide 9 text
Application publishing platform
LINE MINI APP
Slide 10
Slide 10 text
Natural Experience with
AI Technology
Slide 11
Slide 11 text
LINE is AI Company
Slide 12
Slide 12 text
What is AI?
Slide 13
Slide 13 text
LINE AI
Beyond Clova and Chatbot
Slide 14
Slide 14 text
AI
> Machine takes over human tasks
> Learn, think, decide, improve
Slide 15
Slide 15 text
Fuzzy Logic,
Neural Network
Enble
AI to Infer
Like Human
Look Back on AI Development
Slide 16
Slide 16 text
No content
Slide 17
Slide 17 text
LINE AI Application
Slide 18
Slide 18 text
LINEਮ๐盅ݣ
Slide 19
Slide 19 text
LINE MUSIC
AI DJ
Slide 20
Slide 20 text
SMART CHANNEL
Slide 21
Slide 21 text
LINE TODAY
> 糔֦ࡅ稭
Slide 22
Slide 22 text
LINEਮ๐ੜ䒻ಋ
Slide 23
Slide 23 text
LINE懱௳礚挨
Slide 24
Slide 24 text
Training Reimburse Data
Model
Actual Reimburse Data
Text Classifier
Training
Evaluation
Data
Model Creation Process Usage Process
Automatically generate all documents
for reimburse application
LINE
OCR API
Mac App
iOS App
LINE AI Hackathon - ᛔ㵕䁭癱羬翄
ML
ML
Slide 25
Slide 25 text
22222
LINE Taiwan Service
LINE AI
Data
NLU
Ad Targeting
OCR
Text Recognition
Slide 26
Slide 26 text
v
Next Step
Face
LINE AI
Speech
Voice
Slide 27
Slide 27 text
Collaboration With
LINE Clova AI Team
LINE Clova AI team
instructed on leveraging
LINE AI technology
Join Global Team
Share platform and insight
with global team to evolve
LINE services
Machine Learning
System From Scratch
Created local
machine learning
system to build chatbot
Collaboration With
LINE BRAIN Team
Advance LINE AI
technology with Chinese
language support
LINE Taiwan Data Engineering Task Force
Slide 28
Slide 28 text
Provide for 3rd Parties
Data
New Services / Functions
UI / UX Improvement
Algorithms
Users /Services ML Engineers
Search, Recommendation, Ad Platform, Monitoring,
Chatbot, Voice, Speech, Vision, OCR, Face, Video…
Slide 29
Slide 29 text
LINE Taiwan Data Engineering Task Force
and LINE Brain
Advance LINE AI technology with Chinese language support
Evolve LINE Taiwan services with global team
LINE BRAIN
CHATBOT
LINE BRAIN
TEXT ANALYTICS
LINE BRAIN
SPEECH TO TEXT
LINE BRAIN
TEXT TO SPEECH
LINE BRAIN
OCR
LINE BRAIN
VISION
LINE BRAIN
VIDEO ANALYTICS
Text Speech Vision
Slide 30
Slide 30 text
No content
Slide 31
Slide 31 text
Privacy First
Slide 32
Slide 32 text
Data
Governance
Responsibility as a Platform
Unshakable foundation for sustainable trustworthy of users
Slide 33
Slide 33 text
New records
1 Trillion
Hive / Spark jobs,
30K Presto queries
70K
(Compressed)
new data volume
+390 TB
Every Day
Slide 34
Slide 34 text
Implementing a scalable, on-demand analytics environment
Multi-Tenancy
Data Quality
Automated data validation & meta data collection
Accessibility
Seamless unification of Hadoop clusters into a single big cluster
Key Challenges
Slide 35
Slide 35 text
Unified Self-Service
Data Platform
Data Science
ML Engineering
Data
Governance
Governed Self-Service Data Platform
Slide 36
Slide 36 text
Feature as a Service
Standardization Democratization
Feature
Engineering
Y - features
Obfuscated
User Features
Company-wide
Components
Service-specific
Components
Machine Learning Team
Service-specific
Components
etc.
Various Teams
ML Solutions
Centralized Features (Managed by ML Team)
Z - features
User Features
C - features
Content
Features
Feature
Engineering
E
T
L
Log
E
T
L
Contents
Policy
Service(s)
Slide 37
Slide 37 text
Feature Extraction
List of package ids (order by timestamp)
mid: [10, 4465, 1025960, 5413, 1456, 1299646, ...]
⁃ 10: [0.1, 0.5, 0.2, 0.2, 0.35, ...]
⁃ 4465: [0.9, 0.8, 0.4, 0.1, 0.2, ...]
⁃ 1025960: [0.45, 0.2, 0.2, 0.6, 0.9, ...]
⁃ 5413: [0.8, 0.8,0.1, 0.2, 0.5, ...]
⁃ 1456: [0.7, 0.3, 0.7, 0.3, 0.2, ...]
⁃ 10: [0.0, 0.0, 0.0, ...., 0.3, 0.1, 0.9, .., 0.0, 0.0, 0.0]
⁃ 4465: [0.3, 0.6, 0.1, …., 0.0, 0.0, 0.0, ..., 0.0, 0.0, 0.
⁃ 1025960: [0.0, 0.0, 0.0, ..., 0.0, 0.0, 0.0, ..., 0.7, 0.3, 0.1]
⁃ 5413: [0.9,0.1,0.1,....,0.0, 0.0,0.0, ..., 0.0, 0.0, 0.0]
⁃ 1456: [0.0, 0.0, 0.0, ..., 0.0, 0.0, 0.0, ..., 0.2, 0.2, 0.6
mid: [0.1, 0.2, 0.1, … 0.3, 0.1, 0.8, …, 0.1, 0.6, 0.4]
mid: [0.5, 0.3, 0.2, 0.9, 0.8]
fastText
GMM
Accumulate
dimensionality reduction (PCA with matrix sketching)
y-sparse-features
(typically 6,000 dimension)
y-dense-features
(typically 400 dimension)
Slide 38
Slide 38 text
Privacy
AI
Data
Governance
Slide 39
Slide 39 text
Leading AI Technology