:sti on s :

_1 pot ar-
. Ne ura l Ne two rks (8]
3tiC 1:U
p ·d d ina .tbe Jll ·t's
(a) roV1 e an ove rvi ew of 3 1 £ h fig ure a.n
aye r eed for wa rd neu ral net wo rk wit bil e l
tio ns .
. . bouDda.rieS ~ [4]
(b) EArplain how m lt· 1 [4]
ild in u aye r neu ral net wo rks can mo del non -lin e~r dec1s10n
bas ic bu 1-
g blo ck Per cep tro n(/ neu ron ) is jus t
a lin ear classifier.
.
(c) Bri efl y exp lai n how 1ear mn . pef,orm ed in neu ral net wo rks .
. g 1s
[3+21
!. CN N 1ut ion
. of lD Co nvo . al da ta
. . . . mu la . h cfu nen s1o n ]
p n of Co nvo luti on ope rat ion wit h for [ 4
(a) roV1de bn ef des cnp tio . · · g for big .
• upo n neu ral net wo rks wh ile trrunID
(b) E A'1)lam how CN N's imp rov ed I/ filt er of siz
h
e
suc h as ima ges . .th the ker ne
29 X 29. It is con vol ute ~ Wl f im age . Co mp ute t e
to the CN N is of size [2]
(c) Th e inp ut ima ge g of 1 pix els on eac h sid e
O
an
I and zer o pad din
5 X 5 wit h a stri de of [3]
tur e em bed din g.
dim ens ion of out put fea
of vanishing gradients in dee p lea rni ng.
(d) Bri efly exp lain the con cep t

3. En sem ble Lea rni ng

[3]
sed by ens em ble lea rni ng [4]
(a) Briefly exp lain 3 key pro ble ms add res
ensemble lea rni ng tec hni que s.
(b) Giv e sho rt overview of at lea st two [P. T. 0. ]

1
 g
4. D a ta C lu st er in [4]
io n of a .
(a) Give v is u al co n st ru ct
1o m er ati~ e cl us te ring technique. [4]
code o f gg
(b) P ro v id e p su ed o K means clustermg al
gorithm . [5]
I
(c) L.ist a t ea st 5 li m it at io f
°
K m ea ns cl us te ring algorithm .
ns
els
5. Time-series Mod .
st at e tr ansition di ag ra m]
.
ple st t t rans1t. 1on probabilty m at ri x an d d ra w as so ci at ed [3 +3
(a ) P ro v id e a sain a e
(
t 3 st at es in th e Markov process). e n o tati ons for in p u t,
assu m e a t le as h ap p ro p ri at [4]
at io n o f R N N un folding with time wit
visualiz
(b) D ra w g ra p hpicaralam et er s.
o u p u t an d
N Classifier
6 · D im en si o n al it y R ed u ct io n & KN n th e in p u t d-dimensional
io n gi ve
ti n g P C A ba se d di mensionality reduct (6 ]
s o f co m p u
(a ) P ro v id e key st ep sisting o f n samples. fier . [31
d a ta m at ri x X co n K N N cl as si
m il ar it y m ea su re s th a t ca n b e used in
la o f th re e si
(b) P ro v id e formu
ea r
& K er n el T o ck h 1 in so lv in g n o n -l in l 1
7. SV M li .
zat lo n ca n e P 6
n o f example o f ho w ke rn e l51
al v is u al iz at io
(a ) Give g ra h p ic
ms.
classification proble
rmulation o f SVM.
(b ) P ro v id e d u al fo

Good Luck!