[KongchangAI]
ConceptHessian flow

Hessian flow

Refers to the mechanism by which the local Hessian matrix is transported linearly along the SGD iteration trajectory, used to analyze how large gradient shocks influence the tail distribution of iterates through local curvature.

Timeline (last 90 days)

Oct 19

The paper's main result is that for regularly varying innovations (index t ∈ (1,2)), the tail probability of the last iterate is dominated by a single large gradient shock transported through the local Hessian flow.

Unverified50%

All Facts (1)

Unverified

The paper's main result is that for regularly varying innovations (index t ∈ (1,2)), the tail probability of the last iterate is dominated by a single large gradient shock transported through the local Hessian flow.

50%

Source Articles

Exact Tail Laws for SGD with Infinite-Variance GradientsOct 19