4 ms·
lsh: similarity(x, y) < thresh => E[P(d(lsh(x), lsh(y)) < 1)] > 1-eps for some eps For a model that learns a good representation, and some suitable distance fu
by pizza 3y ago
lsh: similarity(x, y) < thresh => E[P(d(lsh(x), lsh(y)) < 1)] > 1-eps for some eps
For a model that learns a good representation, and some suitable distance function to measure distances between embeddings (e.g. Euclidean)
model: similarity(x, y) < thresh => E[P(d(emb(x), emb(y)) < r] > 1-eps
for some cutoff distance r in the vector space
- charcircuit 3y agoThe difference in definitions is that my definition does not imply that the mapping maps similar objects to similar locations in space. Embeddings that do can be useful which is why you see people using machine learning to create them, but it is not a necessary property of an embedding.
- pizza 3y agoTrue, it’s not a necessity just based on the types of things, but in practice it’s the intended use