Locality sensitive hashing (LSH) |
- To support range and kNN queries
|
- Idea of randomized algorithms
|
|
- Algorithms that make random choices
|
|
- Monte Carlo: Probabilistic but bounded error results in bounded time accuracy improves with each run
|
|
- Las Vegas: Correct and deterministic result but varying time
|
- Goal in LSH: To find a hashing function that is approximately distance-preserving (within some tolerance)
|
- A hash function
is -sensitive if for any , ,
|
|
- If
, then ![](images/img269.png)
|
|
- If
, then ![](images/img271.png)
|