Unverified50% confidenceFactExact time
The generalized Gauss-Newton approximation is a method for approximating the curvature (Hessian matrix) of a neural network loss function, using the outer product of Jacobian matrices to approximate the Hessian, thereby estimating curvature at acceptable computational cost.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedNewton's method uses the inverse of the Hessian matrix to determine the update step size and direction Δθ = -H⁻¹∇L, and can theoretically find an extremum at a quadratic convergence rate.76% similarUnverifiedThe full Newton method requires O(d³) computational complexity to solve for the inverse of the Hessian matrix, which is unacceptable in high-dimensional problems.68% similarUnverifiedThe Muon optimizer, proposed by Jeremy Bernstein et al., is centered on the idea of orthogonalizing gradient matrices via Newton-Schulz iteration.64% similarUnverifiedThe Jacobian matrix describes all first-order partial derivatives of a vector-valued function with respect to the input vector, where the element J_ij = ∂f_i/∂x_j characterizes the influence of a small change in the j-th component of the input on the i-th component of the output.63% similarUnverifiedEckart-Young-Mirsky定理证明，在所有秩为k的矩阵中，由SVD前k个奇异值构成的截断近似矩阵是在Frobenius范数意义下对原矩阵的最佳近似62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102741API
curl https://kongchang.com/api/v1/knowledge/claims/1102741MCP
get_claim(id=1102741)