05/09/2015

IF4071 Pembelajaran Mesin:

Pendahuluan
Semester Ganjil 2015/2016

K1: Masayu Leylia Khodra

K2: Dessi Puji Lestari

Program Studi Sarjana

Teknik Informatika
ITB

PENGANTAR KULIAH

1
 05/09/2015

Administrasi
Kuliah Pilihan 3 sks
Prasyarat:
Probabilitas & Statistika
Struktur Data & Pemrograman
Inteligensi Buatan

Mailing-list: if4071@students.if.itb.ac.id

Pengajar: Masayu Leylia Khodra

Email: masayu@stei.itb.ac.id

Tatap muka:
Selasa, 10.00-11.40, 7610
Rabu, 9.00-9.50, 7610

Luaran

1. menjelaskan perbedaan dari ketiga jenis pembelajaran

(unsupervised, supervised, dan reinforcement)
2. mengimplementasikan algoritma sederhana untuk
ketiga jenis pembelajaran tersebut
3. memilih jenis pembelajaran yang tepat untuk kasus
persoalan/aplikasi tertentu
4. melakukan evaluasi terhadap kinerja suatu algoritma
pembelajaran pada kasus persoalan tertentu
5. menjelaskan persoalan overfitting, serta bagaimana
deteksi dan solusinya
4

2
 05/09/2015

Materi Kuliah
Konsep Pembelajaran Mesin
Pengantar Eksperimen:
konstruksi dataset
analisis karakteristik data
pengukuran kinerja
analisis hasil eksperimen
Supervised learning (overview konsep, variasi, isu) :
Decision tree learning
Artificial neural network
Instance-based learning
Bayesian learning
Support Vector Machines
Genetic algorithm/programming
Unsupervised Learning (overview konsep, variasi, isu)
Hierarchical vs partitional clustering
Graph-based clustering
Reinforcement learning (overview)
5

Penilaian
25% UTS
25% UAS
50% Tugas:
10% eksplorasi, latihan, dan baca makalah (individu)
15% implementasi algoritma (kelompok)
5% analisis problem (kelompok)
5% konstruksi dataset dan analisis karakteristik data
(kelompok)
15% eksperimen (kelompok)

3
 05/09/2015

Referensi
Mitchell, T., Machine Learning, 1997,
McGraw-Hill
Lecture slides for textbook Machine
Learning, T. Mitchell, McGraw Hill, 1997.
http://www.cs.cmu.edu/~tom/mlbook-
chapter-slides.html
Richard Duda, Peter Hart and David Stork,
Pattern Classification, 2nd ed. John Wiley
& Sons, 2001
Christopher Bishop, Pattern Recognition
and Machine Learning. Springer, 2006
Digital library /
http://scholar.google.com/
7

KONSEP PEMBELAJARAN MESIN

4
 05/09/2015

Apa itu Pembelajaran?

(KBBI)
Belajar (v)
berusaha memperoleh
kepandaian atau ilmu;
berlatih;
berubah tingkah laku atau
tanggapan yg disebabkan oleh
pengalaman;
Pembelajaran (n)
Proses, cara, perbuatan
menjadikan orang atau
makhluk hidup belajar; 9

Apa itu Pembelajaran/ Learning?

Learning is
improving with experience E at task T
with respect to performance measure P
(Mitchell, 1997)

We do not yet know how to make computers

learn nearly as well as people learn (Mitchell, 1997)
However, we know that some algorithms are effective
for certain types of learning tasks
10

5
 05/09/2015

Pembelajaran Mesin

Machine learning is
the science of getting computers to act
without being explicitly programmed.

(Coursera - Stanford - Machine Learning Andrew Ng,

https://www.coursera.org/course/ml)

11

Main Concerns
How to construct computer
programs that automatically improve
with experience

6
 05/09/2015

MENGAPA PERLU PEMBELAJARAN MESIN ?

13

Mengapa Pembelajaran Mesin ?

Information overload (online data), manual
processing is nearly impossible
Kategorisasi
Peringkasan
Ekstraksi informasi
Personalisasi (self customizing programs)
Expert system:
Automatic knowledge acquisition for solving knowledge.
acq. bottleneck
Improve decisions using historical data
Develop software applications we cant program
by hand
14

7
 05/09/2015

Some Problem Domains

Data mining problems where large databases may
contain valuable implicit regularities that can be
discovered automatically
(e.g., to analyze outcomes of medical treatments from patient databases or to
learn general rules for credit worthiness from financial databases);

Poorly understood domains where humans might

not have the knowledge needed to develop effective
algorithms
(e.g., human face recognition from images, speech recognition);
Domains where the program must dynamically
adapt to changing conditions
(e.g., controlling manufacturing processes under changing supply
stocks or adapting to the changing reading interests of individuals)

Applications
Data-mining programs that learn to detect
fraudulent credit card transactions
Information-filtering systems that learn users'
reading preferences
Autonomous vehicles that learn to drive on
public highways
etc

8
 05/09/2015

(Some) Successful Applications

Typical Data Mining Task

18

9
 05/09/2015

Bidang Ilmu Multi Disiplin

Inteligensi Buatan
Sistem berbasis Pengetahuan
Logika
Teori Bayesian
Teori informasi
Probabilitas dan Statistik
Teori kompleksitas komputasional
Psikologi, neurobiologi, psikolinguistik,
antropolog linguistik, dst

19

Definition
A computer program is said to learn from
experience E with respect to some class
of tasks T and performance measure P,
if
its performance at tasks in T, as measured
by P, improves with experience E.

10
 05/09/2015

Well Defined Learning Problems (1)

A checkers learning problem:
Task T: playing checkers
Performance measure P: percent of games won
against opponents
Training experience E: playing practice games
against itself

Well Defined Learning Problems (2)

A handwriting recognition learning problem:
Task T: recognizing and classifying handwritten
words within images
Performance measure P: percent of words
correctly classified
Training experience E: a database of handwritten
words with given classifications

11
 05/09/2015

Well Defined Learning Problems (3)

Robot driving learning
problem:
Task T: driving on public
four-lane highways using
vision sensors
Performance measure P:
average distance
traveled before an error
(as judged by human
overseer)
Training experience E: a
sequence of images and
steering commands
recorded while
observing a human
driver

Well Defined Learning Problems (4)

Speech Recognition learning problem:
Task T: recognizing and classifying spoken words
within speech signals
Performance measure P: percent of words
correctly classified
Training experience E: a database of spoken words
with given classifications

12
 05/09/2015

Desain Learning System

1. Pemilihan training experiences
2. Pemilihan fungsi target
3. Pemilihan representasi fungsi target
4. Pemilihan algoritma pembelajaran
5. Finalisasi desain

25

1. Pemilihan training experiences

Type learning experiences (feedback)
Direct
Individual checkers board states and correct move for
each
<board state, correct move>*
Indirect
move sequences and final outcomes of various games
played
<move sequences, final outcome>*
Required credit assignment
Lebih sulit dibanding direct training

26

13
 05/09/2015

1. Pemilihan training experiences (lanj)

Tingkat kontrol
Direct:
Bergantung pada annotator (teacher):
pemilihan data (contohnya informative board state)
anotasi (correct move for each board state)
Siapkan board state (random, heuristics, confusing),
lalu anotasi
Indirect:
Complete control
Memerlukan skenario konstruksi data

27

1. Pemilihan training experiences (lanj)

Representasi distribusi data terhadap ukuran

kinerja
Ideal:
distribusi data antara data training dan testing data di
dunia nyata sama
Training:
experience against itself vs testing in world tournament

28

14
 05/09/2015

2. Pemilihan Fungsi Target

This learning task is representative of:
a large class of tasks for which the legal moves that
define some large search space are known a priori
but for which the best search strategy is not known.
Thus, the program needs to learn how to choose
the best move for any given board state
Fungsi Target : ChooseMove: Board
Move
Board: set of legal board states B
Move: set of legal moves

29

2. Pemilihan Fungsi Target (Lanj.)

Feedback yang dipilih: indirect (playing against
itself)
+ : no external trainer
data sebanyak mungkin
Alternatif: memberi skor untuk setiap board
V: Board
Real {higher score for better board}
Move: max(V(legal successor board state))

30

15
 05/09/2015

Fungsi Target V: Board
Real

31

3. Pemilihan Representasi Fungsi

Target
Collection of rules
Neural Net
Polynomial Function of board features

32

16
 05/09/2015

4. Pemilihan Algoritma Pembelajaran

Estimate training values

Adjust the weights: Least Minimum Square

33

5. Desain Final
Sistem: <move sequences, final outcome>
Critic: <board state, Vtrain>
Generalizer: w0..w6
Experiment generator

34

17
 05/09/2015

Design
Choices

35

Issues in Machine Learning

What algorithms can approximate function well and
when?
How much training data is sufficient?
How does complexity of hypothesis representation
impact accuracy?
How does noisy data influence accuracy
What are the theoretical limit of learnability?
How can prior knowledge of learner can help?
What clues can we get from biological learning system?
How can system alter their own representation?

36

18
 05/09/2015

Referensi
Mitchell, T., Machine Learning, 1997,
McGraw-Hill, Chapter 1.
KBBI

37

19