Unverified50% confidenceFactExact time
Reverse-mode automatic differentiation can compute the gradients of a scalar loss with respect to all parameters in O(n) time, whereas forward mode requires O(n) forward passes
1
Sources
50%
Confidence
Long-term
Relevance
9/9/2026
First Seen
Sources
Related Entities
Related Claims
Verified反向传播算法的本质是通过链式法则在计算图上高效计算标量损失对每个参数的偏导数70% similarUnverified自动微分的前向模式适合输入少输出多的场景，反向模式适合输入多输出少的场景，深度学习因参数多而损失为标量属于后者69% similarVerifiedBackpropagation is essentially a dynamic programming algorithm for efficiently computing gradients, which reduces the computational complexity of gradient calculation by propagating error signals layer by layer from the output layer to the input layer and reusing intermediate computation results.67% similarUnverified自动微分分为前向模式（从输入向输出传播导数，适合输入维度远小于输出维度）和反向模式（从输出向输入反向传播梯度，适合输出维度远小于输入维度）65% similarVerifiedAutomatic differentiation builds a computation graph by recording all tensor operations at runtime, then uses reverse-mode differentiation to automatically compute exact gradients of arbitrarily complex functions.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/884441API
curl https://kongchang.com/api/v1/knowledge/claims/884441MCP
get_claim(id=884441)