Unverified60% confidenceFactExact time
STE (Straight-Through Estimator) performs truncated rounding in the forward pass but passes gradients straight through in the backward pass (approximating the gradient as 1) to address the non-differentiability of quantization operations, and was formally systematized in a 2013 paper by Bengio et al.
2
Sources
60%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
AI Application Engineer Interview Guide: Core Topics in Model Quantization and Inference Deployment
redditr/learnmachinelearning7/9/2026
Related Claims
UnverifiedThe Straight-Through Estimator (STE) was proposed by Bengio et al. in 2013 and is the theoretical foundation of extreme quantization training.70% similarUnverifiedThe conjugate moment measure differs from the original decomposition solely in that the transport uses the gradient of the Gaussian transform of the potential function, rather than the gradient of the potential function directly.61% similarUnverifiedA 2010 paper by Glorot and Bengio systematically revealed the mechanism of the vanishing gradient problem.61% similarUnverifiedCordero-Erausquin and Klartag proved the moment measure decomposition theorem in 2015: any probability measure ρ with a first moment, zero barycenter, and support not contained in a hyperplane can be decomposed into a specific form, where the transport map is the gradient of a convex function u.60% similarUnverifiedXavier initialization (Glorot, 2010) and He initialization (He et al., 2015) are based on Gaussian or uniform distributions and maintain stable gradient flow by designing variances of 1/n and 2/n respectively (where n is the input dimension).58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/611104API
curl https://kongchang.com/api/v1/knowledge/claims/611104MCP
get_claim(id=611104)