Concept单位球面映射 / L2归一化嵌入
Unit Sphere Embedding
A technique that applies L2 normalization to text embedding vectors so that they lie on the unit sphere, removing the influence of vector magnitude so that similarity can be measured directly by inner product or cosine distance, facilitating analysis of directional changes in semantic trajectories.
Timeline (last 90 days)
Oct 20
Unit sphere embedding maps reasoning segment vectors onto a sphere of radius 1 via L2 normalization, so that similarity can be measured directly by inner product or cosine distance.
Unverified50%