Unverified50% confidenceFactExact time
The regularization term of EWC takes the form λ/2 × Σ Fᵢ(θᵢ - θ*ᵢ)², where θ*ᵢ is the weight value after training on the old task, Fᵢ is the Fisher information, and λ controls the constraint strength.
1
Sources
50%
Confidence
Long-term
Relevance
10/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDuring LoRA training, the pre-trained model weights W₀ are frozen and only the two matrices B and A are trained, with the forward propagation formula H = W₀X + (α/r)BAX.57% similarUnverifiedMFU的计算公式为MFU=(实际模型FLOPs)/(硬件理论峰值FLOPs×训练时间)，其只计算模型本身有效计算，不包含重计算等额外开销，区别于HFU56% similarUnverifiedThe partial derivative of the MSE loss function with respect to parameter W is ∂L/∂W = (2/n) * Xᵀ(XW - y), and the partial derivative with respect to b is ∂L/∂b = (2/n) * Σ(y_pred - y).56% similarUnverifiedL2算法相对于基向量系数位长β达到O(n⁴·β·(n+β))的位复杂度，在β=O(n)时相当于O(n⁶)，而经典LLL约为O(n⁹)56% similarUnverifiedOutlines通过将EBNF文法预编译为有限状态机（FSM）来高效实现约束采样56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/991783API
curl https://kongchang.com/api/v1/knowledge/claims/991783MCP
get_claim(id=991783)