LINE TECHPULSE 2019 - Keynote

Keynote by Marco Chen @ LINE TECHPULSE 2019 https://techpulse.line.me/

Transcript

2. > > Marco Chen > > Senior Technical Director, LINE Taiwan Limited LINE TAIWAN TECHPULSE 2019 Keynote
3. 2016 2019 2017 2018 10 Times Number of LINE Taiwan engineers has increased by 10 times since 2016 4th LINE TAIWAN TECHPULSE
4. More Technical Insight > Almost 30 LINE Taiwan engineers shared technical insight in 
 DevDay 2019, Japan
5. Life on LINE > > Connect with LINE Platform > > Natural Experience with AI Technology
6. LINE SPOT Finding restaurants nearby and make reservation Life on LINE Featuring services for every part of your day
7. TAIWAN JAPAN VIETNAM INDONESIA MENA Conomi ͠ͽͧ͡ NOW SHOPPING GO LINEϪξη GET IT LINE SHOPPING SWAY LIVE MENA
8. Connect with LINE Platform
9. Application publishing platform LINE MINI APP
10. Natural Experience with AI Technology
11. LINE is AI Company
12. What is AI?
13. LINE AI Beyond Clova and Chatbot
14. AI > Machine takes over human tasks > Learn, think, decide, improve
15. Fuzzy Logic, Neural Network Enble AI to Infer 
 Like Human Look Back on AI Development
17. LINE AI Application
18. LINEਮ๐盅ݣ
19. LINE MUSIC AI DJ
20. SMART CHANNEL
21. LINE TODAY > 糔֦ࡅ稭
22. LINEਮ๐ੜ䒻ಋ
23. LINE懱௳礚挨
24. 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
25. 22222 LINE Taiwan Service LINE AI Data NLU Ad Targeting OCR Text Recognition
26. v Next Step Face LINE AI Speech Voice
27. 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
28. 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…
29. 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
31. Privacy First
32. Data Governance Responsibility as a Platform Unshakable foundation for sustainable trustworthy of users
33. New records 1 Trillion Hive / Spark jobs, 
 30K Presto queries 70K (Compressed) 
 new data volume +390 TB Every Day
34. 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
35. Unified Self-Service Data Platform Data Science ML Engineering Data Governance Governed Self-Service Data Platform
36. 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)
37. 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)
38. Privacy AI Data Governance
39. Leading AI Technology