1
00:00:00,000 --> 00:00:09,340
Consequently, if we were to find some
other vector which is less than.

2
00:00:09,700 --> 00:00:18,524
400 or 350 bits away from.
X then, the chance of that happening

3
00:00:18,524 --> 00:00:26,772
purely at random is very small.
So if this vectors represent objects, it's

4
00:00:26,772 --> 00:00:33,060
highly likely that these two vectors
represent the same object.

5
00:00:34,180 --> 00:00:41,338
In many ways what we've done here is very
reminiscent, of locality sensitive

6
00:00:41,338 --> 00:00:44,823
hashing.
However, associative memories.

7
00:00:44,823 --> 00:00:50,526
Defer from.
Lsh or even indexing in that they don't

8
00:00:50,526 --> 00:01:00,444
really store the objects.
So, what they do instead is they choose.

9
00:01:00,444 --> 00:01:05,200
A set of random vectors.
A large number, M.

10
00:01:05,743 --> 00:01:11,831
Could be a million, but certainly not
anywhere nearly as large as the space

11
00:01:11,831 --> 00:01:18,891
itself.
And instances that are seen by this memory