Unverified50% confidenceFactExact time
MSE损失函数对参数W的偏导为 ∂L/∂W = (2/n) * Xᵀ(XW - y)，对b的偏导为 ∂L/∂b = (2/n) * Σ(y_pred - y)
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified低秩参数化将完整协方差矩阵分解为Σ = WW^T + D的形式，使参数量从O(d²)降至O(dr)69% similarUnverified均方误差（MSE）是损失函数的最经典形式，其数学表达为所有样本预测误差平方的均值68% similarUnverified逻辑回归的前向传播流程为：线性组合 z = w·x + b，通过 sigmoid 激活 ŷ = 1/(1+e^(-z))，再计算交叉熵损失61% similarUnverified单变量线性回归模型可表示为 y = w * x + b，其中w是权重（斜率），b是偏置（截距）60% similarUnverified在鲁棒回归中将似然函数从高斯替换为t分布，等价于在损失函数中降低离群点的权重，能提升模型对异常值的鲁棒性60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/901606API
curl https://kongchang.com/api/v1/knowledge/claims/901606MCP
get_claim(id=901606)