Unverified50% confidenceFactExact time
Low-rank decomposition is typically implemented using Singular Value Decomposition (SVD), reducing the parameter count from m×n to r×(m+n).
1
Sources
50%
Confidence
Long-term
Relevance
10/8/2026
First Seen
Sources
Related Entities
Related Claims
Unverified完整雅可比矩阵的计算代价是O(m×n)次前向传播，实际应用中研究者通常采用随机化SVD或Lanczos迭代等算法近似计算其低秩分解69% similarUnverifiedLoRA leverages the low-rank approximation principle of SVD to decompose the incremental updates of large weight matrices into the product of two low-rank matrices, thereby compressing the number of trainable parameters from billions to the millions range.66% similarUnverifiedIn practice, Muon uses approximately 5 steps of Newton-Schulz iteration to approximate the matrix sign function, reducing the computational overhead of exact SVD.64% similarUnverified强行取极值样本会引入离群点，迫使SVM构建更复杂的决策边界，高阶poly核在这种数据上收敛更慢63% similarUnverifiedWhen ε ≈ 1, the gradient second-moment term in the denominator is diluted, adaptive scaling has almost no effect, and the optimizer behaves close to pure SGD.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/986700API
curl https://kongchang.com/api/v1/knowledge/claims/986700MCP
get_claim(id=986700)