DR+Clustering
From ResearchWiki
(Created page with "Goal of the project: Understand the impact of dimensionality reduction methods on clustering. Try to uncover relationship between a dimensionality reduction method and a clusteri...") |
m |
||
| Line 3: | Line 3: | ||
'''High dimensions are "weird"'''. | '''High dimensions are "weird"'''. | ||
| + | |||
A mathematician and his best friend, an engineer, attend a public lecture on geometry in thirteen-dimensional space. | A mathematician and his best friend, an engineer, attend a public lecture on geometry in thirteen-dimensional space. | ||
| + | |||
"How did you like it?" the mathematician wants to know after the talk. | "How did you like it?" the mathematician wants to know after the talk. | ||
| + | |||
"My head's spinning", the engineer confesses. "How can you develop any intuition for thirteen-dimensional space?" | "My head's spinning", the engineer confesses. "How can you develop any intuition for thirteen-dimensional space?" | ||
| + | |||
"Well, it's not even difficult. All I do is visualize the situation in arbitrary N-dimensional space and then set N = 13." | "Well, it's not even difficult. All I do is visualize the situation in arbitrary N-dimensional space and then set N = 13." | ||
| Line 11: | Line 15: | ||
And Clustering is "hard" (though Amit Daniely, Nati Linial, Michael Saks say its only hard when it does not matter!) | And Clustering is "hard" (though Amit Daniely, Nati Linial, Michael Saks say its only hard when it does not matter!) | ||
| - | Data | + | ==Data== |
MNIST Digits data: | MNIST Digits data: | ||
| - | + | ==Leader Board== | |
| - | == | + | |
{| border="1" style="width: 100%; text-align:left" class="content" | {| border="1" style="width: 100%; text-align:left" class="content" | ||
|- | |- | ||
Revision as of 06:24, 5 October 2012
Goal of the project: Understand the impact of dimensionality reduction methods on clustering. Try to uncover relationship between a dimensionality reduction method and a clustering technique of your choice (if there exists any).
High dimensions are "weird".
A mathematician and his best friend, an engineer, attend a public lecture on geometry in thirteen-dimensional space.
"How did you like it?" the mathematician wants to know after the talk.
"My head's spinning", the engineer confesses. "How can you develop any intuition for thirteen-dimensional space?"
"Well, it's not even difficult. All I do is visualize the situation in arbitrary N-dimensional space and then set N = 13."
And Clustering is "hard" (though Amit Daniely, Nati Linial, Michael Saks say its only hard when it does not matter!)
Data
MNIST Digits data:
Leader Board
| Data | # Data points | # Dimensions | # Target Dimensions | Dimensionality Reduction Method | Clustering Technique | Rand Index | NMI | Accuracy |
|---|---|---|---|---|---|---|---|---|
| MNIST | ||||||||
| Gisette | ||||||||
| Olivetti Faces |
