seamless 원활한
it'll be very seamless for you to work on this assignment
generously 관대하게
google generously sopported the cloud
whatnot 이것저것
when you want to use larger machine or whatnot
administrative 행정의
those are administrative issues today
grid 격자(바둑판 같은거)
what computer see is gigantic gird of number
semantic 의미론적인
there's huge gap between the semantic of a cat and
these pixel valuse that the computer is seeing.
robust 강력한, 건강한
And our algorithms need to be robust to this.
deform 변형하다.
objects can also deform
assume 자세를 취하다
cats can really assume a lot of different poses
occlusion 닫힘
clutter 어수선함, 어지렵혀있음
brittle 다루기 힘든
There is some reasons. one is supper brittle.
scalable 확장 가능한
If ~, you have to do again, so this is not a scalable approach.
ingest 섭취하다, 흡수하다
ml classifier that is gonna ingest all of data
spit out 뱉다, 결과를 내다
ml classifier ~, ~, and then spit out the model
with respect to ~에 관하여
It has a lot of nice property with repect to data-driveness and whatnot
left most 가장 왼쪽
for the left most column
backward 거꾸로, 뒤로
this is actually somewhat backwards
carve 조각하다
is just sort of carving up the space
generalization 일반화 (더 큰 범주로 만드는것)
a slight generalization
flip back and forth 뒤집다(반대로 생각해보다)
it is important to flip back and forth several different view point
retrieve 검색하다
metric 미터법, underlying 숨겨져있는, 함축된, topology 위상
different distance metrics make different assumptions
about underlying geometry or topology
underlying 숨겨져있는, 함축된
it should be drawn from the same underlying probability distribution
generic 일반적인
if it's just a generic vector in some space
way = much (비격식)
that was way more trouble than it was worth
ahead of time 미리
choices that you make ahead of time
alleviate 덜다, 해결하다
then that's how we try to alleviate this problem
plot 도면, 그래프
whenever you make ml model, you end up making plots like this.
underlying manifold : 작은 차원의 데이터 공간
k-nearest algorithm doesn't really make any assumptions about these underlying manifold
offset: 보정값
I just sort of carefully constructed these translations and these offsets to match exactly
modular : 모듈형( 레고블록처럼 완성된 부품으로 전체를 만드는것)
nature 특성
another example of kind of this modular nature of nn
tune: 일치시키다
parametric model : parameter(weight)를 튜닝하는 모델
quadrant : 사분면
even number will be the opposite two quadrants