|
|
|
|
|
|
by pfdietz
743 days ago
|
|
|
Some dimensional reduction is always possible, due to the Johnson-Lindenstrauss lemma. For example, for 8 billion data points, reducing to 1400 dimensions enables preserving distances within +- 50% (that can probably be tightened a bit) regardless of what the uncompressed data is. |
|
|