Unverified50% confidenceFactExact time
In gradient attribution, the number of steps in Integrated Gradients, whether the baseline is all zeros or the mean, whether gradients are smoothed, and how ReLU saturation regions are handled all significantly change the attribution map.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGradient accumulation is crucial in automatic differentiation: when a variable influences the loss function through multiple computational paths, the gradients from each path must be summed73% similarUnverifiedGradient clipping works by proportionally scaling down the entire gradient vector when its norm exceeds a preset threshold, so that its norm equals exactly that threshold.73% similarUnverifiedIn zeroth-order optimization, converting function values into gradient estimates via finite differences or Gaussian smoothing and then applying first-order methods usually incurs a cost linear in the dimension d.72% similarVerifiedGradient descent iteratively adjusts weights in the opposite direction of the loss function's gradient, causing the prediction error to converge progressively.69% similarUnverifiedGradient descent mathematically involves partial derivatives and the chain rule, and its intuitive essence is a 'downhill' process of finding the lowest point.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1063427API
curl https://kongchang.com/api/v1/knowledge/claims/1063427MCP
get_claim(id=1063427)