Machine Learning for Modern Artiﬁcial Intelligence

(1)

Machine Learning for Modern Artificial Intelligence

Hsuan-Tien Lin National Taiwan University

June 25, 2019

other versions presented in Academia Sinica

& TWSIAM Annual Meeting

(2)

ML for (Modern) AI

Outline

ML for (Modern) AI

ML Research for Modern AI

ML for Future AI

H.-T. Lin (NTU) ML for Modern AI 1/38

(3)

ML for (Modern) AI

From Intelligence to Artificial Intelligence

intelligence: thinking and actingsmartly

• humanly

• rationally

artificialintelligence:computersthinking and actingsmartly

• humanly

• rationally

humanly≈smartly≈rationally

—are humans rational? :-)

(4)

ML for (Modern) AI

Traditional vs. Modern [My] Definition of AI

Traditional Definition

humanly ≈ intelligently ≈ rationally My Definition

intelligently ≈ easily

is your smart phone ‘smart’? :-)

modern artificial intelligence

=applicationintelligence

(5)

ML for (Modern) AI

Examples of Application Intelligence

Siri

By Bernard Goldbach [CC BY 2.0]

Amazon Recommendations

By Kelly Sims [CC BY 2.0]

iRobot

By Yuan-Chou Lo [CC BY-NC-ND 2.0]

Vivino

from nordic.businessinsider.com

(6)

ML for (Modern) AI

Machine Learning and AI

Easy-to-Use

Acting Humanly Acting Rationally

Machine Learning

machine learning: core behind modern (data-driven) AI

(7)

ML for (Modern) AI

ML Connects Big Data and AI

From Big Data to Artificial Intelligence

big data

ML

artificial intelligence

ingredient tools/steps dish

(Photos Licensed under CC BY 2.0 from Andrea Goh on Flickr)

“cooking” needs many possible tools & procedures

(8)

ML for (Modern) AI

Bigger Data Towards Better AI

best route by shortest path

best route by current traffic

best route by predicted travel time

big datacanmake machine look smarter

(9)

ML for (Modern) AI

ML for Modern AI

big data

ML AI

human learning/

analysis

domain knowledge

(HumanI)

method

model expert system

• human sometimesfaster learneroninitial (smaller) data

• industry: black plum is as sweet as white

often important to leverage human learning, especiallyin the beginning

(10)

ML for (Modern) AI

Application: Tropical Cyclone Intensity Estimation

meteorologists can ‘feel’ & estimate TC intensity from image

TC images

ML

_estimation^intensity

human learning/

analysis

domain knowledge

(HumanI)

CNN polar

rotation invariance

current weather

system

better than current system & ‘trial-ready’

(Chen et al., KDD 2018) (Chen et al., Weather & Forecasting 2019)

(11)

Outline

ML for (Modern) AI

ML for Future AI

(12)

Cost-Sensitive Multiclass Classification

(13)

What is the Status of the Patient?

?

H7N9-infected cold-infected healthy

• aclassificationproblem

—grouping ‘patients’ into different ‘status’

are all mis-prediction costs equal?

(14)

Patient Status Prediction

error measure = society cost XXXX

XXXXXX actual

predicted

H7N9 cold healthy

H7N9 0 1000 100000

cold 100 0 3000

healthy 100 30 0

• H7N9 mis-predicted as healthy:very high cost

• cold mis-predicted as healthy: high cost

• cold correctly predicted as cold: no cost

human doctors consider costs of decision;

how about computer-aided diagnosis?

(15)

Our Works

binary multiclass

regular well-studied well-studied

cost-sensitive known(Zadrozny et al., 2003) ongoing(our works, among others)

selected works of ours

• cost-sensitive SVM(Tu and Lin, ICML 2010)

• cost-sensitive one-versus-one(Lin, ACML 2014)

• cost-sensitive deep learning(Chung et al., IJCAI 2016)

why are peoplenot

using thosecool ML works for their AI? :-)

(16)

Issue 1: Where Do Costs Come From?

A Real Medical Application: Classifying Bacteria

• by human doctors: different treatments⇐⇒ serious costs

• cost matrix averaged from two doctors:

Ab Ecoli HI KP LM Nm Psa Spn Sa GBS

Ab 0 1 10 7 9 9 5 8 9 1

Ecoli 3 0 10 8 10 10 5 10 10 2

HI 10 10 0 3 2 2 10 1 2 10

KP 7 7 3 0 4 4 6 3 3 8

LM 8 8 2 4 0 5 8 2 1 8

Nm 3 10 9 8 6 0 8 3 6 7

Psa 7 8 10 9 9 7 0 8 9 5

Spn 6 10 7 7 4 4 9 0 4 7

Sa 7 10 6 5 1 3 9 2 0 7

GBS 2 5 10 9 8 6 5 6 8 0

issue 2: is cost-sensitive classification really useful?

(17)

Cost-Sensitive vs. Traditional on Bacteria Data

. . . . . .

Are cost-sensitive algorithms great?

RBF kernel

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

OVOSVM

csOSRSVM csOVOSVM csFTSVM

algorithms

cost

.

...Cost-sensitive algorithms perform better than regular algorithm

Jan et al. (Academic Sinica) Cost-Sensitive Classiﬁcation on SERS October 31, 2011 15 / 19

(Jan et al., BIBM 2011)

cost-sensitivebetter thantraditional;

but why are peoplestill not

using those cool ML works for their AI? :-)

(18)

Issue 3: Error Rate of Cost-Sensitive Classifiers

The Problem

0.1 0.15 0.2 0.25 0.3

0 0.05 0.1 0.15 0.2

Error (%)

Cost

• cost-sensitive classifier: low cost but high error rate

• traditional classifier: low error rate but high cost

• how can we get theblueclassifiers?: low error rate and low cost

cost-and-error-sensitive:

more suitable forreal-world medical needs

(19)

Improved Classifier for Both Cost and Error

(Jan et al., KDD 2012)

Cost

iris ≈

wine ≈

glass ≈

vehicle ≈

vowel

segment

dna

satimage ≈

usps

zoo

splice ≈

ecoli ≈

soybean ≈

Error

iris

wine

glass

vehicle

vowel

segment

dna

satimage

usps

zoo

splice

ecoli

soybean

now,are people using those cool ML works for their AI? :-)

(20)

Lessons Learned from

Research on Cost-Sensitive Multiclass Classification

? H7N9-infected cold-infected healthy

1 more realistic (generic) in academia

6=more realistic (feasible) in application e.g. the ‘cost’ ofinputting a cost matrix? :-)

2 cross-domain collaborationimportant

e.g. getting the ‘cost matrix’ fromdomain experts

3 not easy to winhuman trust

—humans are somewhatmulti-objective

(21)

Active Learning by Learning

(22)

Active Learning: Learning by ‘Asking’

labeling isexpensive:

active learning ‘question asking’

—query ynofchosenx_n

unknown target function f : X → Y

labeled training examples ( , +1), ( , +1), ( , +1)

( , -1), ( , -1), ( , -1)

learning algorithm

A

final hypothesis g≈f

+1

active: improve hypothesis with fewer labels (hopefully) by asking questionsstrategically

(23)

Pool-Based Active Learning Problem

Given

• labeled poolD_l =n

(featurexn ,label yn(e.g. IsApple?))oN n=1

• unlabeled pool D_u=n x˜_soS

s=1

Goal

design an algorithm that iteratively

1 strategically querysome˜xs to get associatedy˜s 2 move (x˜_s,y˜s)fromD_utoD_l

3 learnclassifier g^(t)fromD_l

and improvetest accuracy of g^(t) w.r.t#queries

how toquery strategically?

(24)

How to Query Strategically?

Strategy 1

askmost confused question

Strategy 2

askmost frequent question

Strategy 3

askmost debateful question

• choosingone single strategy isnon-trivial:

0 10 20 30 40 50 60

0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8

% of unlabelled data

Accuracy

RAND UNCERTAIN PSDS QUIRE

0 10 20 30 40 50 60

0.4 0.5 0.6 0.7 0.8 0.9

Accuracy

0 10 20 30 40 50 60

0.5 0.6 0.7 0.8 0.9 1

Accuracy

application intelligence: how to choose strategy smartly?

(25)

Idea: Trial-and-Reward Like Human

when do humanstrial-and-reward?

gambling

K strategies:

A₁, A2, · · · , AK

tryone strategy

“goodness” of strategy asreward

K bandit machines:

B₁, B2, · · · , BK

tryone bandit machine

“luckiness” of machine asreward

intelligent choice of strategy

=⇒intelligent choice ofbandit machine

(26)

Active Learning by Learning

(Hsu and Lin, AAAI 2015)

K strategies:

A₁, A₂, · · · , A_K

try one strategy

“goodness” of strategy as reward

Given: K existing active learning strategies for t = 1, 2, . . . , T

1 let some bandit modeldecide strategy A_k to try

2 query the ˜x_ssuggested by A_k, and compute g^(t)

3 evaluategoodness of g^(t) asrewardoftrialto update model

only remaining problem: what reward?

(27)

Design of Reward

ideal rewardafter updating classifier g^(t) by the query (x_n_t,ynt):

accuracy ofg^(t) ontest set {(x⁰_m,y_m⁰ )}^M_m=1

—test accuracyinfeasiblein practice because labelingexpensive more feasible reward: training accuracy on the fly

accuracy of g^(t) onlabeled pool {(xnτ,ynτ)}^t_{τ =1}

—butbiasedtowardseasierqueries

weighted training accuracyas a better reward:

acc. of g^(t) oninv.-prob. weightedlabeled pool n

(x_n_τ,y_n_τ,_p¹

τ) ot

τ =1

—‘bias correction’fromquerying probability within bandit model Active Learning by Learning (ALBL):

bandit +weighted training acc. as reward

(28)

Comparison with Single Strategies

UNCERTAINBest

5 10 15 20 25 30 35 40 45 50 55 60 0.55

0.6 0.65 0.7 0.75 0.8 0.85 0.9

Accuracy ALBL

vehicle

PSDSBest

5 10 15 20 25 30 35 40 45 50 55 60 0.5

0.55 0.6 0.65 0.7 0.75 0.8

Accuracy ALBL

sonar

QUIREBest

5 10 15 20 25 30 35 40 45 50 55 60 0.5

0.55 0.6 0.65 0.7 0.75

Accuracy ALBL

diabetes

• no single best strategyfor every data set

—choosing needed

• proposedALBLconsistentlymatches the best

—similar findings across other data sets

ALBL: effective inmaking intelligent choices

(29)

Discussion for Statisticians

weighted training accuracy ¹_t Pt τ =1

1

pτ qynτ =g^(t)(xnτ)y as reward

• is rewardunbiased estimatorof test performance?

no for learned g^(t) (yes for fixed g)

• is reward fixedbefore playing?

no because g^(t)learned from (xn_t,yn_t)

• is rewardindependent of each other?

no because past history all in reward

—ALBL: tools from statistics +wild/unintended usage

‘application intelligence’ outcome:

open-source toolreleased

(https://github.com/ntucllab/libact)

(30)

Lessons Learned from

Research on Active Learning by Learning

by DFID - UK Department for International Development;

licensed under CC BY-SA 2.0 via Wikimedia Commons

1 Is Statistics the same as ML or AI?

• does it really matter?

• Modern AI should embraceevery useful tool from other fields.

2 scalability bottleneckof ‘application intelligence’:

choiceof methods/models/parameter/. . .

3 think outside of themathbox:

‘unintended’ usage may begood enough

4 important to bebraveyetpatient

• idea: 2012

• paper:(Hsu and Lin, AAAI 2015); software:(Yang et al., 2017)

(31)

Tropical Cyclone Intensity Estimation

(32)

Experienced Meteorologists Can ‘Feel’ and Estimate Tropical Cyclone Intensity from Image

Can ML do the same/better?

• lack ofML-ready datasets

• lack ofmodel that properly utilizes domain knowledge issues addressed in our latest works

(Chen et al., KDD 2018) (Chen et al., Weather & Forecasting 2019)

(33)

Flow behind Our Proposed Model

TC images

ML

_estimation^intensity

human learning/

analysis

domain knowledge

(HI)

CNN polar

rotation invariance

current weather

system

is proposedCNN-TCbetter than current weather system?

(34)

Results

RMS Error

ADT 11.75

AMSU 14.40

SATCON 9.66 CNN-TC 9.03

CNN-TC much betterthan current weather system (SATCON)

why are peoplenot using thiscool ML model? :-)

(35)

Lessons Learned from

Research on Tropical Cyclone Intensity Estimation

1 again,cross-domain collaborationimportant e.g. even from ‘organizing data’ to be ML-ready

2 not easy to claimproduction ready

—can ML be used for ‘unseenly-strongTC’?

3 good AI system requiresboth human and machine learning

—still an ‘art’ to blend the two

(36)

ML for Future AI

Outline

ML for (Modern) AI

ML for Future AI

(37)

ML for Future AI

AI: Now and Next

2010–2015 AI becomes promising, e.g.

• initial success of deep learningon ImageNet

• mature tools for SVM (LIBSVM) and others

2016–2020 AI becomes competitive, e.g.

• super-human performance of alphaGoand others

• all big technology companies becomeAI-first

2021–

AI becomes necessary

• “You’ll not be replaced by AI, butby humans who know how to use AI”

(Sun, Chief AI Scientist of Appier, 2018)

(38)

ML for Future AI

Needs of ML for Future AI

more creative win humanrespect e.g. Appier’s 2018 work on

design matching clothes

(Shih et al., AAAI 2018)

more explainable win humantrust e.g. my students’

work on

automatic bridge bidding

(Yeh et al., IEE ToG 2018)

more interactive win humanheart e.g. my student’s work (w/ DeepQ) on efficient disease diagonsis

(Peng et al., NeurIPS 2018)

(39)

ML for Future AI

Summary

• ML for (Modern) AI:

tools + human knowledge ⇒easy-to-use application

• ML Research for Modern AI:

need to bemore open-minded

—in methodology, in collaboration, in KPI

• ML for Future AI:

crucial to be ‘human-centric’

Thank you! Questions?