人工智慧在台灣: 產業轉型的契機與挑戰

陳昇瑋
台灣人工智慧學校執行長

台灣人工智慧學校執行長
玉山金融控股公司科技長
中央研究院資訊所研究員
人工智慧科技基金會董事長
科技生態發展公益基金會執行長

陳昇瑋 / 人工智慧民主化在台灣
博藍尼悖論 (Polanyi's Paradox)
18
我們懂的事情，
比我們能表達出來的更多。
哲學家博藍尼在 1964 年說明了這個現象：
博藍尼悖論不只限制我們能告訴另一個人的事情，一直以來，也為我們賦予機器
智慧的能力，設下根本的限制。長久以來，這限制了機器在經濟中能有效執行的
活動。

人工智慧發展簡史
19
第三波
2010 - 機器學習
1950 - 1970
第一波
邏輯推理
1980 - 1990
第二波
知識表達
第二次 AI 冬天
第一次 AI 冬天

陳昇瑋 / 台灣產業 AI 化 - 如何跨出第一步？
機器學習
24
Training a prediction machine by
showing examples instead of
programming it.
-Yann LeCun
(prediction machine: 可基於已知預測未知的數學模型)

機器學習的定義
25
讓電腦能從資料裡頭淬取出
規則的演算法。
Find the common patterns
from the left waveforms
It seems impossible to
write a program for
speech recognition
你好你好
你好你好
You quickly get lost in the
exceptions and special cases.
(Slide Credit: Hung-Yi Lee)

就放棄教電腦規則，讓它自己學吧！
你好
大家好
人帥真好
You said
“你好”
很多訓練資料
機器學習演算法
從訓練資料中
找到規則

機器學習學到的規則
跟你想的不太一樣
28

現代機器學習算法所學到的複雜規則
29

A Gaydar
30
Based on 35,000 facial images
Human judges: 61% for men, 54% for women
AI judges: 91% for men, 83% for women
A heat map of where the algorithm looks to detect signs of homosexuality (Kosinski and Wang)
https://osf.io/zn79k/

陳昇瑋 / 人工智慧民主化在台灣 33

Machine learning can be highly flexible
• Speech Recognition
• Handwritten Recognition
• Playing Go
• Dialogue System
( )=*f
( )=*f
( )=*f
( )=*f
“2”
“Morning”
“5-5”
“Hello”“Hi”
(what the user said) (system response)
(step)

多數 AI 應用是讓機器學會一個對應關係
40

Deep Learning, Machine Learning, and AI
44

Healthy Diseased
Hemorrhages
No DR Mild DR Moderate DR Severe DR Proliferative DR
1 2 3 4 5

Classical Machine Learning
Deep Learning
Rule-based System
Rule extraction

Using
Deep Learning

There is no free lunch
52
Classical
Machine Learning
Deep Learning

58(Slide Credit: McKinsey&Company)
機器學習在各行各業都已有許多充滿潛力的應用

RECENT ADVANCES IN
DEEP LEARNING

The ImageNet Benchmark in Object Classification:
Example Images in the Bird Class
67
(Classify 256x256 images into one of a thousand categories)

Auto Coloring
71
https://paintschainer.preferred.tech/index_zh.html
https://zhuanlan.zhihu.com/p/24712438

Colorful Image Colorization
72
Zhang, Richard, Phillip Isola, and Alexei A. Efros. "Colorful image colorization." European
Conference on Computer Vision. Springer International Publishing, 2016.

Colorful Image Colorization
73
http://richzhang.github.io/colorization/
A 313-class classification problem
Input: 224x224x1 (L)
Model output: 64x64x313
Pixel values: annealed mean of 313
colors

Colorizing Legacy Photos
74

GENERATIVE
MODELS
75

Truth vs. Generated Samples
78
https://metacademy.org/roadmaps/rgrosse/deep_learning

Source of images: https://zhuanlan.zhihu.com/p/24767059
DCGAN: https://github.com/carpedm20/DCGAN-tensorflow
Anime Girl Face Generation

100 rounds

1000 rounds

5000 rounds

50,000 rounds

BEGAN (Boundary Equilibrium GAN)
Berthelot, David,Tom Schumm, and Luke Metz. "Began: Boundary equilibrium generative
adversarial networks." arXiv preprint arXiv:1703.10717(2017).
https://github.com/carpedm20/BEGAN-tensorflow

https://thispersondoesnotexist.com/
86

StyleGAN
87
https://www.technologyreview.com/s/612612/these-incredibly-real-fake-faces-show-
how-algorithms-can-now-mess-with-us/
https://www.youtube.com/watch?time_continue=362&v=kSLJriaOumA

StyleGAN for cats
88
https://buzzorange.com/techorange/2019/02/15/reve
nge-for-valentines-day/

StyleGAN failed cases
89
https://buzzorange.com/techorange/2019/02/
11/cat-style-gan/

https://buzzorange.com/techorange/2019/02/
15/revenge-for-valentines-day/

Sheng-Wei Chen / AI Now: A Data Science Perspective
AI 自動生成二次元妹子？
或將替代插畫師部分工作
93
http://bangqu.com/b4U76M.htmlhttp://make.girls.moe/#/

Interactive Image Translation with
pix2pix-tensorflow
https://affinelayer.com/pixsrv/

CycleGAN
97

Horse <-> Zibra
98

Deepfakes
100
https://www.inside.com.tw/2017/12/13/gal-gadot-fake-ai-porn
http://www.businessinsider.com/obama-deepfake-video-insulting-trump-2018-4

NATURAL LANGUAGE
PROCESSING (NLP)
102

Natural Language Processing (NLP): Some
Sentence Generation Examples by GTP-2
103
Bad news:
The technology can be used in generating fake news
Transformer based
(Slide credit: HT Kung)

More text generation samples
104https://openai.com/blog/better-language-models/

BERT and GPT-2
105
● Both BERT (from Google) and GPT-2 (OpenAI) are general purpose
pretrained NLP Feature Extractors based on the Transformer trained on
enormous amounts of text data.
● These models can be fine-tuned on small-data NLP tasks (like question
answering), resulting in substantial accuracy improvements compared to
training on these datasets from scratch
Pretrained BERT/GPT-2
Additional classifier for
your own tasks
The pretrained feature
extractor
Based on
Transformer

Advances of NLP Models in Recent Years
106
Deeper models, Larger Datasets
Better Features!
Performance Improvement: ~20%
Reuse of pre-training BERT/GPT-2: models.
Impact is similar to that of ImageNet in computer vision
Paper on theTransformer:Vaswani, A. et al, “Attention Is AllYou Need,” NIPS 2017
One-hot
Word
Embedding
ELMo BERT GPT-2
milestone
Transformer
based
BooksCorpus
(800M words) +
Wikipedia
(2,500M words)
scraped content from
the Internet of 8
million web pages
WMT 2011
(800M words)
WMT 2011
(800M words)
Large-scale dataset like ImageNet
RNN
based
Development of pretrained
‘feature extractor’ for NLP tasks:

REINFORCEMENT
LEARNING
111

Learn from interaction w/ environment to
achieve a goal
Reinforcement Learning
114
https://www.youtube.com/watch?v=IXuHxkpO5E8

Learning to play Go
Supervised v.s. Reinforcement
115(Slide Credit: Hung-Yi Lee)

https://www.youtube.com/watch?v=SHLuf2ZBQSw

NVidia Self Driving Car, 2016

Typical Applications of RL
Play games: Atari, poker, Go, ...
Explore worlds: 3D worlds, Labyrinth, ...
Control physical systems: manipulate, walk, swim, ...
Interact with users: recommend, optimize, personalize,
...
121
(Slide credit: David Silver)

More RL Applications
Flying Helicopter
Driving
Google Cuts Its Giant Electricity Bill With DeepMind-
Powered AI
Parameter tuning in manufacturing lines
Text generation
Hongyu Guo, “Generating Text with Deep Reinforcement
Learning”, NIPS, 2015
Marc'AurelioRanzato,SumitChopra,Michael Auli,Wojciech
Zaremba, “Sequence Level Training with Recurrent Neural
Networks”, ICLR, 2016
122(Slide Credit: Hung-Yi Lee)

人工智慧發展策略建議
AI
IN MEDICINE
127

0.95
F-score
Algorithm Ophthalmologist
(median)
0.91
“The study by Gulshan and colleagues truly
represents the brave new world in
medicine.”
“Google just published this paper in JAMA
(impact factor 44.405) [...] It actually lives
up to the hype.”
Dr. Andrew Beam, Dr. Isaac Kohane
Harvard Medical School
Dr. Luke Oakden-Rayner
University of Adelaide

Deep Learning for Detection of Diabetic
Eye Disease (2016)
132
Algorithm’s F1-score: 0.95
Median F1-score of 8 ophthalmologists : 0.91

134OCT: Optical CoherenceTomography (干涉光視網膜斷層掃描)

Top 5 Causes of Deaths
140
(Ref: Medical error—the third leading cause of death in the US |The BMJ)

arxiv.org/abs/1703.02442
Tumor localization score (FROC):
model: 0.89
pathologist: 0.73
(Slide Credit: Google Brain)

Deep Learning for Kidney Function Classification and
Prediction using Ultrasound-based Imaging
Chin-Chi Kuo1, Chun-Min Chang2, Kuan-Ting Liu2, Wei-Kai Lin2,
Chih-Wei Chung1, and Kuan-Ta Chen2
1Big Data Center, China Medical University Hospital, China Medical University, Taichung, Taiwan
2Institute of Information Science, Academia Sinica, Taiwan
eGFR
(腎功能指數)

Cardiologist-Level Arrhythmia Detection
with Convolutional Neural Networks
150
Goal: diagnose irregular heart rhythms, also known as
arrhythmias, from single-lead ECG signals better than a
cardiologist

Input and Output
Input: a time-series of raw ECG signal
The 30 second long ECG signal is sampled at 200 Hz
From 29,163 patients
Output: a sequence of rhythm classes
The model outputs a new prediction once every second
Total 14 rhythm classes are identified
151

Model
34 layers NN
16 residual blocks
2 conv layers per block
Filter length = 16 samples
# filter = 64*k, k start from 1 and is
incremented every 4-th residual block
Every residual block subsamples
its input by a factor of 2
153

Results – F1 score
154

Google Healthcare Focuses
156

Predictive tasks for healthcare
Given a large corpus of training data of de-identified medical records, can we
predict interesting aspects of the future for a patient not in the training set?
● will patient be readmitted to hospital in next N days?
● what is the likely length of hospital stay for patient checking in?
● what are the most likely diagnoses for the patient right now? and why?
● what medications should a doctor consider prescribing?
● what tests should be considered for this patient?
● which patients are at highest risk for X in next month?
Collaborating with several healthcare organizations, including UCSF, Stanford, and
Univ. of Chicago. Have early promising results (no public paper yet)

168
AI is bringing
new solutions
to
age-old problems.

AI in Finance Applications
169

A spectrum of learning tasks
• Low-dimensional data (e.g. less
than 100 dimensions)
• Lots of noise in the data
• There is not much structure in
the data, and what structure
there is, can be represented by a
fairly simple model.
• The main problem is
distinguishing true structure
from noise.
• High-dimensional data (e.g.
more than 1000+ dimensions)
• The noise is not sufficient to
obscure the structure in the data
if we process it right.
• There is a huge amount of
structure in the data, but the
structure is too complicated to
be represented by a simple
model.
• The main problem is figuring out
a way to represent complicated
structure that allows it to be
learned.
Statistics --------------------- Deep Learning

85%

Algorithmic Trading
176
Algorithmic finance is roughly a $1 trillion market (as of early 2018)
Growing at 10.3% CAGR [1]
90% volume in the public equities markets
50% volume in the futures markets
6 of the top 10 hedge funds based on performance are quant funds and most of
those are quant funds in commodities markets.
[1] source:Technavio’s 2016 report: Global AlgorithmicTrading Market 2016-2020

Algo. trading is somewhat lagged in Asia
177

Top-rank Hedge Funds using AI
Two Sigma (http://www.twosigma.com/)
PDT Partners (http://www.pdtpartners.com/)
DE Shaw (http://www.deshaw.com/)
Winton Capital Management (https://www.wintoncapital.com/)
Ketchum Trading, LLC (http://www.ketchumtrading.com/)
Citadel (http://www.citadel.com)
Voleon (http://voleon.com)
Vatic Labs (http://www.vaticlabs.com)
Cubist, aka Point72 (https://www.point72.com/careers/...)
Man AHL (https://www.ahl.com)
Renaissance Technologies
Rebellion Research
178

Algorithmic Trading in Hedge Funds
179

But, AI is still far from perfect
180
http://www.eurekahedge.com/Indices/IndexView/Eurekahedge/683/Eurekahedge_AI
_Hedge_fund_Index

Correlation between AI funds and humans
has risen in past years
181
https://www.bloomberg.com/news/articles/2018-03-12/robot-takeover-
stalls-in-worst-slump-for-ai-funds-on-record

Algorithmic Trading in Stock Market
182
https://www.bloomberg.com/news/features/2017-12-05/how-ai-will-
invade-every-corner-of-wall-street

Common use of machine learning
Prediction of direction
Prediction of price
Prediction of risk
Prediction of portfolio performance
Detection of big players’ movements
Rebalancing
Liquidating
…
Fully automation of trading
183
https://www.quora.com/Is-there-any-hedge-fund-using-
machine-learning-based-algorithms-for-trading

Prediction for low-frequency trading
184

Alternative data and its applications
Individuals
Twitter sentiment data to trade the broad equity market (iSentium)
News sentiment data to trade bonds, equities and currencies (RavenPack)
Web search, personal email, app downloads, job listing, …
Business processes
Consumer transaction data to trade individual stocks (Eagle Alpha)
PoS, payments, invoices / receipts, real estates, cargo, retail prices, phone
calls, shipments via logistics, lodging, taxi, insurance, …
Sensors
Geolocation data to estimate retail activity and trade individual stocks
(Advan Research)
Satellite image data to estimate employee activity and trade individual
stocks (RS Metrics)
185

iSentium: Tweets
186
Construction of iSentium Daily Directional Indicator
The universe is limited to the 100 stocks which are most representative of the S&P 500,
filtered using tweet volume and realized volatility measures.
Tweets are assigned a sentiment score using a patented NLP algorithm.
By aggregating tweet scores, a sentiment level is produced per minute between 8:30 AM and
4:30 PM every day. Sentiment for the day is aggregated using an exponentially weighted
moving average over the past ten days.
S&P 500 returns are forecasted using a linear regression over the sentiment scores for the
past two days, with betas evolved via a Kalman filter.

Eagle-Alpha: Email Receipts
187
We have analyzed a dataset of email receipts for 97 companies. 36 of these were private
companies, and 61 public, 31 of which were S&P 500 constituents. Taking liquidity into
consideration, we decided to test trading signals for the S&P 500 companies only.

RSMetrics: Satellite images
188

Blurring between Quant and ML Engineer
189
https://www.indeed.com/jobs?q=Machine%20Learning
%20Quant

190
https://www.bloomberg.com/news/features/2017-12-
05/how-ai-will-invade-every-corner-of-wall-street

Mobile computing, inexpensive sensors collecting terabytes of data, and the
rise of machine learning that can use that data will fundamentally change the
way the global economy is organized.
- Fortune, “CEOs:The Revolution is Coming,” March 2016

孔祥重院士
• 美國卡內基美隆大學電腦教授
• 美國哈佛大學資訊科技與管理博士學程共同主席
• 行政院 SRB 會議海外專家與科技顧問
• 行政院科技顧問
• 國家級計畫重要推手
• 數位台灣計畫 (e-Taiwan)
• 行動台灣計畫 (M-Taiwan)
• 電信國家型計畫
• WiMAX 發展藍圖
• 網路通訊國家型計畫
H.T. Kung
• 中央研究院院士
• 美國哈佛大學電腦與電機系蓋茲講座教授
現任
經歷

2000: Mobile Device1980: PC 1990: Internet 2015: AI + IoT
Taiwan’s next opportunity?

以人工智慧提升
台灣產業競爭力

March – November in 2017
232
台塑石化
長春石化
奇美實業
英業達
欣興電子
敬鵬工業
可成科技
致茂電子
永進機械
研華科技
聯發科技
台積電
紡織所
宏遠紡織
台元紡織
佳和紡織
強盛染整
農科院
龍鼎蘭花
經緯航太科技

產業 AI 化的挑戰
01 實戰人才的缺乏
02資料基礎建設不足
03 找對問題不簡單
04產學之間的鴻溝

Global Talent 2021 Report
237
(Source: https://www.oxfordeconomics.com/my-oxford/projects/128942)

242
產業共通
挑戰
預測性維護瑕疵檢測
原料組合最佳化自動流程控制

台灣人工智慧學校首屆開學典禮
產業共通挑戰 #1－瑕疵檢測
243

LCD Panel Defects
244(Slide Credit: IBM)

台灣人工智慧學校首屆開學典禮 245
Typical PCB defects

Typical defects after
SMT (Surface-Mount Technology) process
短路
空焊
極反
缺件
浮高
跪腳
撞件
錫球
墓碑
…
246
https://www.researchmfg.com/2011/02/soldering-defect-symptom/

More SMT/DIP Defect Examples
247

Convolution Neural Networks + Transfer Learning
Pre-trained using 14-million image dataset
ResNet with > 8-million parameters
Input
images
Model training /
inference
OK
OK
以深度學習進行自動瑕疵檢測

實際案例 – 視覺檢測速度比較
252
產線數量: 23 條
4 位目檢人員; 漏網率約 5%
AOI 設備每小時影像輸出量: 配合人力允許條件, 60 萬張/每日
(極限為每條產線 2 萬張/小時 = 1104 萬/日)
判定耗時: 30 萬張 / 人日 = 120 萬張/日
傳統
人力
目檢
深度
學習
系統
硬體設備: 中高階桌上型電腦 + NVIDIA GPU: 10 ~ 15 萬
軟體: 開源軟體 + 高度調校之深度學習模型
品質:模型漏網率控制在 0.01% 之下，目檢人員只需檢查原本總數之 5%
的圖片
判斷速度: 166.67 張 / sec  每日 1440 萬張影像

實際案例 – 視覺檢測效益評估
253
品質：根據複判初步統計，目檢人員漏網率至少為 12.9%
速度：目檢人員 8~10 位，每天約可檢查共約 3,000,000 張
傳統
人力
目檢
深度
學習
系統
硬體設備：中高階桌上型電腦 + NVIDIA GPU: 10 ~ 15 萬
軟體：開源軟體 + 高度調校之深度學習模型
品質：模型漏網率控制在 1% 之下，目檢人員只需檢查原本
總數之 10% 的圖片
速度：8,640,000 張 / 天 = 100 張 / 秒

產業共通挑戰 #2－自動流程控制
254

良品範圍
良率: 61%
QualityIndex
人為控制設備參數

256
採用深度學習控制設備參數
作業員良率: 61%
自動控制良率: 98%
QualityIndex

Especially important for equipment with high failure cost (such as
motors in machine tools)
Also important for expensive consumables (such as blades used in
precision cutting machines)
257
產業共通挑戰 #3－預測性維護

預測某段時間後的設備狀態
258

產業共通挑戰 #4－原料組合最佳化

染整業的打色問題

Pigment 1 Pigment 2 Pigment 3打色成功率: 70% to 95%

PROJECT Θ TEAM HAS
SOLVED
10+ PROBLEMS
FROM 10+ COMPANIES
WITHIN 6 MONTHS…
264

2017/06/26 Project θ Meeting
265

尋找一個把能量放大
的方法
1,000x

01 實戰人才的缺乏
02資料基礎建設不足
03 找對問題不簡單
04產學之間的鴻溝
產業 AI 化的挑戰
關鍵

陳昇瑋 / 人工智慧民主化在台灣 270http://bangqu.com/g394k4.html

台塑企業
奇美實業
英業達集團
義隆電子
聯發科技
友達光電
272
捐助人

273
http://aiacademy.tw/opening-video-180127/

http://aiacademy.tw/
01
02
03
產業 AI 化的軍校
領域專家 + 人工智慧
讓「找不到人才」不再是障礙

2018/01/27 台北一期開學
2018/04/12 台北一期結業
2018/05/12 台北二期開學
2018/07/21 新竹一期開學
2018/08/18 台中一期開學
2018/09/02 台北二期結業
2018/09/29 台北三期開學
2018/11/10 新竹一期結業
2018/12/08 新竹二期開學
2018/12/15 台中一期結業
2019/01/05 台中二期開學
2019/01/26 台北三期結業
2019/02/23 台北四期開學
2019/03/16 南部一期開學
2019/04/20 新竹二期結業
2019/05/04 新竹三期開學

台中一期經理人班畢業照
281

台北三期經理人班畢業照
283

台灣人工智慧學校校區
285
台北總校新竹分校
台中分校
南部分校
(1000 坪, 三重)
(1200 坪, 竹科)
(900 坪, 台灣大道)
(1000 坪, 台南仁德)
(經理人班, 中山大學)

校區一覽
286
新竹分校
台中分校
台北總校
南部分校

台北總校 - 群光電子大樓
287

台北總校 - 群光電子大樓
288

新竹分校－
新竹市力行路二十一號
289

新竹分校－
新竹市力行路二十一號
290
上課教室

292
台中分校本部 & 技術班上課地點－
民權路 239 號國泰人壽大樓 15-17F

台中分校經理人班上課地點－
東海大學人文大樓
294

296
南部分校經理人班 (中山大學國研大樓)

技術領袖培訓班入學考試
程式設計
統計
機率
線性代數微積分

經理人班學員職位分布

空氣盒子計畫
中國信託商業銀行
中國鋼鐵
中國醫學大學醫院
中研院化學所
中研院生多中心
宜蘭食品工業
中華電信研究院
天下雜誌
商周集團
嘉實資訊
智邦科技
臺灣永光化工
天氣風險
宏遠興業
東森得易購
痞客邦 PIXNET
長春石化
雲象科技
敬鵬工業
305
學術夥伴
台塑集團
奇美實業
友達光電
義隆電子
聯發科技
英業達集團
裕隆紡織
元大投信
遠見雜誌

豐碩的技術班專題實作成果
台股指數漲跌預測
股市收盤價預測
全球股債漲跌預測
實價登錄房價預測
腸病毒感染爆發預測
癌症病人生存預測
速訊寫稿機器人
影像敍述自動生成
文章自動標籤生成
文章標題生成
染整業智慧打色
PCB 瑕疵偵測
銅箔缺陷多元分類
植物品種辨識
3D細胞影像分割
人物與人臉辨識
疲勞駕駛偵測
人與物互動辨識
動物聲音辨識
自走車學步
307

更多製造業專題
瑕疵偵測
故障預測
產線參數優化
台電契約容量優化
流程優化
訂單預測
…
308

問題探討Ｑ & A建模參考文獻資料前處理實戰成果

● 原始週報範例 ● Hybrid 週報結果
速訊寫稿機器人
I
II
III

自動挑出壞的咖啡豆
311

跟著狗躲著貓的自走車
312

Playing your own covers – 即時伴奏 Demo
Free Tempo

AI 為盲胞 “唸” 影片
316

校友傑出表現勇奪各項創新競賽
318「開源大補帖」冠軍 300萬元台幣獎金
2018 第三屆京台青年創新創業大賽
獲選優勝前進北京台法太空創新黑客松冠軍
NCTS 2018 年醫療黑客松第二名

校友傑出表現勇奪各項創新競賽
319
新北創力坊第八屆最佳技術升級獎
智在家鄉-數位社會創新競賽
AI神農氏獲選特別獎
工研院 Aidea 競賽
AOI 組冠軍經濟部工業局 Open Data 優勝
一期、二期同學親愛精誠聯手出
擊，獲得佳作x2 優勝 x 2 銅牌x1

334
台灣人工
智慧學校
顧問諮詢產學連結
技術推廣社群交流
新創輔導職涯發展
台灣人工智慧學校將不只是學校

台灣人工智慧學校招生中…
335
台北第五期
截止: 6/4
開學: 7/20
新竹第四期
截止: 7/23
開學: 9/7
台中第三期
截止: 4/16
開學: 6/1
南部第二期
截止: 6/28
開學: 8/3
400 學員 x 每年 3 期 x 4 校區
= 每年培養 4800 AI 工程師 / 經理人

ADVICES FOR
AI DEPLOYMENT
339

人工智慧導入進程
340
成功
案例
開放
文化
資料
生態
技術
工具
流程
整合
先以成功案例
來創造價值及
營造信心
打破部門間的
隔閡，讓資料
可以分享及集
中處理
選擇正確的人
工智慧技術工
具，建立團隊
或尋找技術伙
伴
讓人工智慧技
術成為工作流
程的一部分
人與人工智慧
協同增進生產
力

Execute pilot projects to gain momentum
Build an in-house AI team
Provide broad AI training
Develop an AI strategy
Develop internal and external communications
341

State of AI In The Enterprise, 2018
347
Deloitte interviewed 1,100 IT and line-of-business executives
from US-based companies in the 3rd quarter of 2018.
82% of enterprise AI early adopters are seeing a positive ROI from
their production-level projects this year.
69% of enterprises are facing a “moderate, major or extreme”
skills gap in finding skilled associates to staff their new AI-driven
business models and projects.
63% of enterprises have adopted machine learning, making this
category the most popular of all AI technologies in 2018.

348

349

Source : Gartner
變革文化/決心不足
資料問題
技術人才缺乏
目標/效果不明確
資源不足

陳昇瑋 / 資料科學人才的養成
Major Roles in an AI Team
353
Product Manager Data Scientist
Machine Learning
Engineer
Data Engineer

陳昇瑋 / 資料科學人才的養成 354
IT / Data
Infrastructure
ML / DL
Skills
Data Engineer
Machine Learning
Engineer
Data Scientist
Domain
Expertise
連結資料與商業價值
蒐集及處理資料以資料開發模型，
讓人工智慧決策更精準

陳昇瑋 / 從大數據走向人工智慧

經理人須具備的 10 個認知 (1/2)
沒有資料就不會有人工智慧
也不是有資料，就一定能產出人工智慧：資料
品質與模型建立方法也是關鍵
你從來沒有答案的事情，人工智慧也不會有
沒有變異的資料，等於沒有資訊，同樣不會有
答案
不要忽略資料的成本
362

經理人須具備的 10 個認知 (2/2)
統計圖表與機器學習模型各有所長，不能盡信任一種
人工智慧導入之後必然造成企業工作方式或流程的改
變
身為推動人工智慧導入工程的主管，本身一定要相信
資料和模型
不要把人工智慧的導入及建置全部丟給資訊人員，業
務主管一定要親自參與
人工智慧跟人一樣，需要長期觀察，並需要不斷翻新
363

Perception:
ML products are mostly about ML
364

Reality: ML requires DevOps, lots of it
365

持續的團隊支援
366
A common data platform and workflow is
crucial for enterprise success.
Data Engineer ML Engineer Biz Analyst DevOps DevOps +
ML Engineer
App
Developer
(Credit: IBM Systems Lab Services)
(all under the supervision of Data Scientist)

AI vs. BI
AI systems suggest decisions for users by making
predictions
BI systems support users make decisions based on data
visualization
Key difference
AI systems are based on generalizable models
BI systems require humans to generalize
Best practice
AI: fast, massive, error-tolerant, ML-capable problems
BI: otherwise
AI+BI: making sense of AI decisions
369

LIMITATIONS OF AI
383

What we can and cannot today
What we can have
Safer car, autonomous car
Better medical image
analysis
Personalized medicine
Adequate language
translation
Useful but stupid chatbots
Information search,
retrieval, filtering
Numerous applications in
energy, finance,
manufacturing,
commerce, law, …
What we cannot have
(yet)
Machine with common
sense
Intelligent personal
assistants
“Smart” chatbots
Household robots
Agile and dexterous robots
Artificial General
Intelligence (AGI)
392(Credit:Yann LeCun)

Strong AI Weak AI
Can think
Own conscious
Act as it can think
Consciousless
(1980)

陳昇瑋 / 人工智慧民主化在台灣 396https://www.youtube.com/watch?time_continue=5&v=RSBWgWThNHo

Mistakes Humans Would Never Make
400

Generating adversarial patches against
YOLOv2
402
https://www.youtube.com/watch?feature=youtu.be
&v=MIbFvK2S9g8&app=desktop

AI Don’t Know What They are Talking About
404
https://www.facebook.com/playgroundenglish/videos/629372370729430/?hc_ref=ARQ
HCaS2GZ9jUgZermEupF5yerADq2X9F9P40OR3n70poUiCy7R0X3oHrGxyLSrWVdI

Change is the only constant.
- Heraclitus (535 BC - 475 BC)
(Slide Credit: Albert Chen)

https://www.israel21c.org/food-expiration-dates-are-about-to-undergo-a-revolution/

Precision Medicine
416

陳昇瑋 / 人工智慧民主化在台灣 418(Credit: 李開復, 《AI 新世界》)

陳昇瑋 / 人工智慧民主化在台灣 420(Credit: 李開復, 《AI 新世界》)

(Source: Future of Jobs Report 2018, World Economic Forum)

AI outperformed 20 corporate lawyers at
legal work
422
Challenge: review risks contained in five non-disclosure agreements (NDAs).
AI vs. associates and in-house lawyers from global firms such as Goldman Sachs, Cisco
and Alston & Bird, as well as general counsel and sole practitioners.
AI matched the top-performing lawyer for accuracy – both achieved 94%. Collectively,
the lawyers managed an average of 85%, with the worst performer recording 67%.
AI: 26 seconds; lawyers’ average: 92 minutes, where the speediest lawyer took 51
minutes
https://www.weforum.org/agenda/2018/11/this-ai-outperformed-20-corporate-
lawyers-at-legal-work/

The Jobs Landscape in 2022
423(Source: Future of Jobs Report 2018,World Economic Forum)
Note: Roles marked with * appear across multiple columns. This reflects the fact that they might be seeing stable or declining demand across one
industry but be in demand in another.
*
*

人機協作
Camelyon Grand Challenge in 2016
根據切片檢查偵測轉移性乳癌
The Winning Team
深度學習演算法: 92.5％
病理科醫師: 96.6％
兩者合作: 99.5％
人類與機器擅長不同的預測層面
人類與機器犯不同類型的錯。
確認這兩種不同的能力，結合人類與機器的預測來克服這
些弱點，這樣的組合可以大幅減少錯誤率。
428

machine-learning model from 30,000+ deals from the last decade that draws from
many sources, including Crunchbase, Mattermark, and PitchBook Data. For each deal,
we looked at whether a team made it to a series-A round by exploring 400 features and
identified 20 features as most predictive of future success.
One of the insights we uncovered is that start-ups that failed to advance to series A had
an average seed investment of $0.5 million, and the average investment for start-ups
that advanced to series A was $1.5 million.
Another example insight came from analyzing the background of founders, which
suggests that a deal with two founders from different universities is twice as likely to
succeed as those with founders from the same university.
from the 2015 cohort of seed-stage companies, 16 percent of all seed-stage companies
backed by VCs went on to raise series-A funding within 15 months. By comparison, 40
percent of recommended by ML (2.5 times improvement)
Human + AI would yield the best performance: 3.5 times the industry average
429
https://www.mckinsey.com/industries/high-tech/our-insights/a-machine-learning-
approach-to-venture-capital

人才媒合
http://jobs.aiacademy.tw/

441
專為人工智慧人才 / 職缺設計的媒合網站。
校友刊登的職缺需求將定期發送給所有校友。
http://jobs.aiacademy.tw/

Take-Home Messages
今天的 AI 如同 1994 年的 World Wide Web
今天學 ML 可能如同 1994 年學寫 HTML and CGI
技術快速進展及堆疊，這是最值得投資技術的時代
AI 在許多領域有殺手級應用，是技術人投入重點領域的黃
金時代
AI 不會在這裡停住，AI 技術才剛開始發展
442

AI 所帶來的改變才正開始…
443
2001 2018

陳昇瑋
台灣人工智慧學校

人工智慧在台灣: 產業轉型的契機與挑戰

Recomendados

Recomendados

Mais conteúdo relacionado

Mais procurados

Mais procurados (20)

Semelhante a 人工智慧在台灣: 產業轉型的契機與挑戰

Semelhante a 人工智慧在台灣: 產業轉型的契機與挑戰 (20)

Mais de Sheng-Wei (Kuan-Ta) Chen

Mais de Sheng-Wei (Kuan-Ta) Chen (17)

人工智慧在台灣: 產業轉型的契機與挑戰