f not do atrox hill 8 n no gs Roth h m n In general f Gilo - - PDF document

SLIDE 1

Maximum

Likelihood

Estimation

Statistics

about analyzing

understanding data

common approach

use parametric model of

data Bin hip

Poi

a

Expt

Nlp 62

How

to estimate params of

dish

from

samples

Approac

hi

Find MLE

most likely choice ofparameters

E

• utome Gn

coin tosses

X

in each x c HIT

assuming

indep

Berfp

whatisp

n

I

xn

in

• tTffxi

G

P

i

i

w

unwon

Prob qseergxi if param is

parameters intnis

ex

probJH's

discretecase

p.m.f

Cont

case

p.d.fi

Xi

Xin

h

H's

no

T's

h

moth

L

xn

xn

G

to

Find

that

maximizes

this

4

log

Likelihood

x

la

log Hx into

n logothologko

d

Ly

f

not

do

atrox

n

gs

hill 8

no

h

Roth

n

m

SLIDE 2

In general

Hx

into

f Gilo

LL

log fCxilo

it

And

max LLG tn

dd

gLL x19

fqLLCx48t

Osduefar

f qL4x't 1

i

samples from a

geometric

dish

wimpmmanam

O

ki ka

Ku

L

K

ka lo

II

I

e

• n

LL

ke

Knt

Effi

e log

to

HogG

T

L

Eddie EE4

• to

set this

to solve

yk E

If

i

ki 1

n

l 8

Eakin

n

SLIDE 3

X

xn

samples from

N O

02 deanPaniance

Lfxr.info 0a

f

xilQ 0a

range

C

I

It

1

i

L

densityof

N

Q

G at

xi

LLCElqoaj i.EE'zenfaroa

Giza

2

• 3
• f

IE

iEIEoa

E.EaEioa acxi

IaI

set

set

noisy

g

i

n

_this

xi 8,5

sample

men

sample

variance

E

E

the

man

SLIDE 4

gCx

a

An estimator

is called

unbiased

if

F 8

trueparameter

Ron If

Xinarefm

ht

xi 8,5

dishy

mean µ

2

Super

varia

is Efsamplevan

62

F Quik.nl

EftnEfxi

t.EC

i xP tnEfxi2 2xixtI7

th.Exi2 2Ih.ExixtI2

th.EE 252 txt

fix

I

EffnEEx

5

t.EE xi

n Kxi5

h

f

EH

EADY

L

a

pin

L

http

SLIDE 5

X Xn

xn

MP

Ex

it

fiTI I