Unverified50% confidenceFactExact time
Xavier 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).
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedXavier initialization was proposed by Xavier Glorot in 2010, sampling weights from a distribution with mean 0 and variance 2/(n_in + n_out), and is suitable for symmetric activation functions such as Sigmoid/Tanh.77% similarUnverifiedXavier initialization was proposed by Glorot in 2010 and He initialization was proposed by He et al. in 2015; their variances are designed to be 1/n and 2/n respectively (where n is the input dimension).75% similarUnverifiedHe初始化针对ReLU激活函数设计，通过将权重方差设为2/n来保证各层激活值方差稳定59% similarUnverifiedSTE (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.58% similarUnverifiedThe distribution gap (domain shift) was first systematically described by Ben-David et al. in domain adaptation theory in 2007.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/831825API
curl https://kongchang.com/api/v1/knowledge/claims/831825MCP
get_claim(id=831825)