-
8/11/2019 ISOMAP.pdf
1/20
some r c e
e
COMP61021 Modelling and Vis
ure app ng
en
ualization of High Dimensional Data
-
8/11/2019 ISOMAP.pdf
2/20
n ro uc on
-
Exam les
Case Study: PCA vs. L
Relevant Issues
Conclusions
COMP61021 Modelling and Visu
A
lization of High Dimensional Data2
-
8/11/2019 ISOMAP.pdf
3/20
COMP61021 Modelling and Visu lization of High Dimensional Data3
-
8/11/2019 ISOMAP.pdf
4/20
A state-of-the-art non-linear
In ISOMAP, a geodesic dista
Useful for recovering a low d
COMP61021 Modelling and Visu
MDS method for modelling manifold
ce metric is employed.
imensional isometric embedding
lization of High Dimensional Data4
-
8/11/2019 ISOMAP.pdf
5/20
-
COMP61021 Modelling and Visu
lization of High Dimensional Data5
-
8/11/2019 ISOMAP.pdf
6/20
-
COMP61021 Modelling and Visu
lization of High Dimensional Data6
-
8/11/2019 ISOMAP.pdf
7/20
-
COMP61021 Modelling and Visu
lization of High Dimensional Data7
-
8/11/2019 ISOMAP.pdf
8/20
-
COMP61021 Modelling and Visu
lization of High Dimensional Data8
-
8/11/2019 ISOMAP.pdf
9/20
-
COMP61021 Modelling and Visu
lization of High Dimensional Data9
-
8/11/2019 ISOMAP.pdf
10/20
-
COMP61021 Modelling and Visu
lization of High Dimensional Data10
-
8/11/2019 ISOMAP.pdf
11/20
MDS slides
COMP61021 Modelling and Visu lization of High Dimensional Data11
-
8/11/2019 ISOMAP.pdf
12/20
COMP61021 Modelling and Visu lization of High Dimensional Data12
-
8/11/2019 ISOMAP.pdf
13/20
COMP61021 Modelling and Visu lization of High Dimensional Data13
-
8/11/2019 ISOMAP.pdf
14/20
COMP61021 Modelling and Visu
ijkstras lgorithm
lization of High Dimensional Data14
-
8/11/2019 ISOMAP.pdf
15/20
COMP61021 Modelling and Visu lization of High Dimensional Data15
-
8/11/2019 ISOMAP.pdf
16/20
Ex
COMP61021 Modelling and Visu
l
lization of High Dimensional Data16
-
8/11/2019 ISOMAP.pdf
17/20
Ex
COMP61021 Modelling and Visu
l
lization of High Dimensional Data17
-
8/11/2019 ISOMAP.pdf
18/20
Ex
COMP61021 Modelling and Visu
l
lization of High Dimensional Data18
-
8/11/2019 ISOMAP.pdf
19/20
COMP61021 Modelling and Visu
lization of High Dimensional Data19
-
8/11/2019 ISOMAP.pdf
20/20
in a low dimensional space via
shortest path and c) construct l
ISOMAP can fail in its two main
Geodesic Approximation Points need to be sampled uni
The intrinsic parameter space
MDS
There might not exist an isom
Suffer from a high computation
There are several extensions to
COMP61021 Modelling and Visu
inimum distortion embedding.
,w-dimensional embedding.
steps:
ormly (densely) from a noiseless manifold.
must be convex.
tric embedding (or anything close to one)
l cost and sensitive to noise.
overcome those problems.
lization of High Dimensional Data20