opportunities of FinTech in the insurance industry Prof. Che Lin - - PowerPoint PPT Presentation

Along with AI: challenges and

• pportunities of FinTech

in the insurance industry

• Prof. Che Lin

National Tsing Hua University 國立清華大學電機系 林澤教授 Joint Regional Seminar 2018/7/25

• Research interests:
• Deep Learning, Data Science, FinTech, Signal Processing in Wireless Communications,

Optimization Theory, Systems Biology

• Education:
• Ph.D. in ECE, UIUC, 2008
• Advisor: Venugopal V. Veeravalli
• M.S. in Applied Mathematics, UIUC, 2008
• M.S. in ECE, UIUC, 2003
• Advisor: Weng Cho Chew
• BS in EE, National Taiwan University, 1999
• Honors and Awards:
• Young Scholar Innovation Award, Foundation for the Advancement of Outstanding

Scholarship, 2017.

• CIEE Outstanding Young Electrical Engineer Award, 2015.
• Best Paper Award for 2014 GIW-ISCB-ASIA conference.
• Best Poster Award, International Workshop on Mathematical Issues in Information

Sciences (MIIS), 2012, Xian, China.

• Master Thesis Award of the Taiwan Institute of Electrical and Electronic Engineering

(Advisor; 2011, 2014)

• University of Illinois: E. A. Reid Fellowship Award, Spring 2008.
• University of Illinois: Vodafone Fellowship, Fall 2006 - Spring 2008.

About Myself

https://goo.gl/images/7PzHKK

What is AI?

https://goo.gl/images/4CFvzo

Who is “Master”?

• Consecutive 60 victories online
• Defeat top GO players: 聶衛平、柯潔、陳耀燁

How come?

https://www.bnext.com.tw/article/42607/unknown-master-beats-top-go-players

中國圍棋網站最近出現名為「Master」的神秘棋士， 它連敗中、日、韓圍棋冠軍及多名好手， 「中國棋王」柯潔也在近日成為Master的手下敗將， 不料柯潔今天竟突然在個人微博發文，透露自己住院， 讓不少粉絲為之心疼，安慰他：「輸個棋而已，壓力別太大」。

A big shock!

http://news.ltn.com.tw/news/world/breakingnews/1937343

• - 自由時報(Jan. 4th , 2017)

No way to defeat Master?

周俊勳認為，即使「Master」下法是過去認為 不好的下法與位置，「但就是拿他沒辦法」。

• - 蘋果日報(Jan. 4th , 2017)

「Master」今（4日）早再度現身， 台灣圍棋高手「紅面棋王」周俊勳出馬迎戰， 周俊勳使用初手天元，之後完全仿照對手下子的「模仿棋」戰術， 但仍遭「Master」完美破解，在第118手認輸投降。

• - 自由時報(Jan. 4th , 2017)

http://www.appledaily.com.tw/realtimenews/article/new/20170104/1027615/ http://news.ltn.com.tw/news/life/breakingnews/1936664

http://www.storm.mg/article/99782

Behind AlphaGo

http://www.techapple.com/archives/4452

Can human defeat AlphaGO?

Rise of the Machines

https://www.youtube.com/watch?v=ebph4hbcZd4

• Sir Christopher Pissarides (Nobel Prize in Economics in

2010)

• Almost certainly disappear as jobs for humans:
• telemarketers (99%)
• loan officers (98%)
• cashiers (97%)
• legal assistants (94%)
• taxi drivers (89%)
• fast food chefs (81%)

Jobs that will be replaced by robots

https://goo.gl/images/umfQkB https://goo.gl/images/4VRMZh

FinTech related

• Fintech (financial technology): a broad category that refers to the

innovative use of technology in the design and delivery of financial services and products.

What is FinTech?

AI in FinTech

https://goo.gl/images/YrP9zX

Ris ise of

• f Beh

ehavior vioral al Big Data (BBD)

https://goo.gl/images/ZhZEHD

• Sir Christopher Pissarides (Nobel Prize in Economics in

2010)

• Almost certainly disappear as jobs for humans:
• telemarketers (99%)
• loan officers (98%)
• cashiers (97%)
• legal assistants (94%)
• taxi drivers (89%)
• fast food chefs (81%)

Jobs that will be replaced by robots

https://goo.gl/images/3fY58B

Traditional way of marketing

Pr Precision ecision ma marking ing bas ased ed on

• n BBD

https://goo.gl/images/ZXjBbe

Integrating deep learning, big data analytics, ChatBot, and customer relation management systems for customer-centric precision marketing

Deep learning in a nutshell

DNN

https://www.kdnuggets.com/2017/04/ai-machine-learning-black-boxes-transparency-accountability.html https://hackernoon.com/challenges-in-deep-learning-57bbf6e73bb

Why deep learning?

https://goo.gl/images/Lrz6ZS

Why deep learning?

Source: Deep Learning, Y. Bengio, MIT

Deep learning vs traditional learning

https://www.kdnuggets.com/2017/04/ai-machine-learning-black-boxes-transparency-accountability.html

End-to-end training

Classification and regression problems

http://kindsonthegenius.blogspot.tw/2018/01/what-is-difference-between.html

• Define business/analytics goals and performance

evaluation metric

• 45,211 customers; 21 input features and 1 output

variable

• Demographic data and previous campaign records
• Age, job, marital, education
• Current/previous campaign records
• Social and economical context attributes
• Potential business goal
• Improve marketing effectiveness by targeting the right

customers

• Data mining goal
• Predict whether a certain customer will subscribe to a term

deposit or not

Bank marketing dataset (dataset 1)

DNN provides accurate predictions

50% improvement over traditional marketing

DNN better with increasing data

50% improvement over traditional marketing

• 30,000 customers; 23 input features and 1 output

variable

• Demographic data and credit card behavior (6

months)

• Age, income, education
• History/Amount of past payment; bill statement
• Potential business goal
• Prevent default payments by lowering risky customers’

credit amounts

• Prevent default payments by supervising risky

customers

• Corresponding analytics goal
• Predict whether a customer will default on next

payment

Credit card defaults dataset (dataset 2)

Recurrent neural network (RNN)

Handle time-series data

Source: Deep Learning, Y. Bengio, MIT

Improved RNN prediction with SVM

30% improvement over traditional default detection

Next step: deploy with ChatBot

http://knowledge.wharton.upenn.edu/article/rise-chatbots-time-embrace/ https://chatbotsmagazine.com/the-complete-beginner-s-guide-to-chatbots-8280b7b906ca

Hierarchical NLG w/ Linguistic Patterns

ENCODER

name[Midsummer House], food[Italian], priceRange[moderate], near[All Bar One]

All Bar One place it Midsummer House All Bar One is moderately priced Italian place it is called Midsummer House Near All Bar One is a moderately priced Italian place it is called Midsummer House DECODING LAYER1 DECODING LAYER2 DECODING LAYER3 DECODING LAYER4

Hierarchical Decoder

• 2. VERB
• 3. ADJ + ADV
• 4. Others

[ … 1, 0, 0, 1, 0, …]

Bidirectional GRU Encoder GRU Decoder

Italian priceRange name

…

All Bar One is a is a moderately All Bar One is moderately

… … … …

…

…

• 1. Repeat-input
• 2. Inner-Layer Teacher Forcing
• 3. Inter-Layer Teacher Forcing
• 4. Curriculum Learning

…

All Bar One is priced place it is called Midsummer House

• 1. NOUN + PROPN + PRON

Input Semantics Semantic 1-hot Representation

ChatBot talking to you

NLG Model BLEU ROUGE-1 ROUGE-2 ROUGE-L (a) Seq2Seq 44.7 51.6 19.5 40.6 (b) + Hierarchical Decoder 41.1 60.2 31.4 46.2 (c) + Hierarchical Decoder, Repeat-Input 41.2 60.5 33.8 48.6 (d) + Hierarchical Decoder, Curriculum Learning 40.9 62.9 34.5 50.1 (e) + All 44.1 67.3 38.0 53.8 (f) (e) w/ High Inner-Layer Teacher-Forcing Prob. 36.9 58.5 31.3 45.9 (g) (e) w/ High Inter-Layer Teacher-Forcing Prob. 42.5 67.3 38.7 53.3 (h) (e) w/ High Inner- and Inter-Layer Teacher-Forcing Prob. 41.7 64.5 36.6 52.0

• Sir Christopher Pissarides (Nobel Prize in Economics in

2010)

• Almost certainly disappear as jobs for humans:
• telemarketers (99%)
• loan officers (98%)
• cashiers (97%)
• legal assistants (94%)
• taxi drivers (89%)
• fast food chefs (81%)

Jobs that will be replaced by robots

https://goo.gl/images/umfQkB https://goo.gl/images/4VRMZh

Actuaries (?%)

Intelligent actuary in the age of AI

https://goo.gl/images/djX1qk

The growth of InsurTech

https://goo.gl/images/Cq9DXN https://goo.gl/images/ac6sGy

Usage-based insurance policy

https://goo.gl/images/teYA8e

A single photograph to underwrite policies

https://smile.lapetussolutions.com/upload

Chatbots for insurance advice

https://goo.gl/images/ha558f

AI vs Human

https://goo.gl/images/s8ij8z

AI will liberate human beings

https://goo.gl/images/QTnUMG

Along with AI

https://goo.gl/images/beV5dv