Classificanowfregressiaulinear
Probabilistic Interpretation whynot linear regression
logistic Regession memory Newton’s Method
Ansafut
Regression
Recally
LeastSquares
fori i n
en which X cIRD y eIR
Given Cx
De find 0 CRdt set
y
where holx
Assume y Propities of E we
nrgmw.IT y how 5 te
Tx
Whyy
Ex
Error unmobledeffects RANDOM Noise Cild Gaussian
want
IT’S UNBIASED
1 IECC
How noisy Vanaance ECCE T or
Turns out unique distribution PanamenrizeoBythis theGaussian
O
2 He Errors Are INDEPENDENT E C t ECE ECE
C n Nloo Pct 68.2
Therefore Ply x so exp lg z 5 parametrizesby0
yinfx jo NIE 04
Picks
exp I
ifj
o
Picking
A DISTRIBUTION
Likelihoods
Amongmanydistributions
Pick mostlikely given All dam iid assumption
LCE _ply Ix II ply
o
Ix Q
t.IT expl cyMzxfio5
log likelihood liar dog Lto
WE use
convenient
IE dog EeosoE
Doesn’tDepend on 0
tofind maximumlikelihood equivalently find
JLE mfs I Ea y x 05 D
o
cyciszzx.ci
zfIqCycisI Doaderudovo
Thus
7