Unverified50% confidenceSolutionExact time
The input gradient kernel reuses the weight blocks resident on the NPU and the dequantizer of the forward pass, avoiding transposed weight copies and ensuring that the forward and backward passes use the same effective weights.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe efficiency of backpropagation comes from reusing intermediate values and error terms stored during the forward pass—the core idea of dynamic programming—allowing gradients for all parameters to be obtained in a single backward pass.63% similarUnverifiedResidual connections add each submodule's input directly to its output (output = F(x) + x), alleviating the vanishing gradient problem in deep networks62% similarUnverifiedBalanced optimal transport forces all local clusters to maintain fixed marginal mass, and forced matching introduces misleading alignment gradients that degrade attack effectiveness.62% similarUnverifiedAfter determining the ordering, HOST recovers the parent set of each node by recursively stripping indirect effects from the total effect between pairs of nodes.61% similarUnverified权重共享使卷积操作等价于对输入信号做互相关运算，并引入平移等变性这一归纳偏置61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1074633API
curl https://kongchang.com/api/v1/knowledge/claims/1074633MCP
get_claim(id=1074633)