Unverified50% confidenceFactExact time
Embedding models are deep neural networks that map text into a fixed-dimensional dense vector space (such as 768 dimensions or 1536 dimensions); cosine similarity is commonly used to measure semantic closeness, with a value range from -1 to 1.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
Unverified当需要表征的特征数量远超可用神经元数量时，网络会利用高维空间中近似正交向量数量远超维度数量的特性（Johnson-Lindenstrauss引理）将多个特征叠加压缩编码到同一组神经元上66% similarVerifiedLatent space refers to the high-dimensional continuous vector space to which input data is mapped after being processed by the encoding layers of a neural network in deep learning, where semantically similar concepts are geometrically closer in that space.64% similarUnverifiedLocal embedding computation typically requires running a small neural network model (such as all-MiniLM-L6-v2, approximately 80 MB), and vector indexing commonly uses approximate nearest neighbor search libraries such as FAISS and Hnswlib.63% similarUnverifiedSPLADE and DeepImpact are representatives of learned sparse retrieval, producing sparse vectors whose per-term weights are learned by a neural network and enabling term expansion.63% similarUnverifiedThe tutorial uses Zhipu AI as an example to demonstrate EmbeddingModel, and DeepSeek as an example to demonstrate the most basic dialogue use case.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1015384API
curl https://kongchang.com/api/v1/knowledge/claims/1015384MCP
get_claim(id=1015384)