Unverified50% confidenceFactExact time
During training, VAEs use the reparameterization trick, which transforms sampling from a probability distribution into z = μ + σ·ε, where ε is standard normal noise, allowing gradients to backpropagate normally.
1
Sources
50%
Confidence
Long-term
Relevance
9/25/2026
First Seen
Sources
Related Entities
Related Claims
Unverified重参数化技巧将采样表示为 z = μ + σ × ε（ε从标准正态分布采样），使随机性外置于ε，从而让μ和σ的梯度可正常反向传播实现端到端训练73% similarUnverifiedThe core innovation of VAE lies in encoding inputs as a probability distribution (typically a Gaussian distribution described by mean μ and variance σ) rather than a deterministic point, making the latent space continuous, smooth, and sampleable.68% similarUnverified贝叶斯推断、最大似然估计、KL 散度等概率统计概念在现代生成模型（如 VAE、Diffusion Model）中广泛应用66% similarUnverifiedThe reparameterization trick, proposed by Kingma and Welling in 2013, enables gradients to backpropagate through the sampling step by transforming the stochastic sampling operation into a deterministic computation plus noise.66% similarUnverified扩散模型的反向过程通常使用U-Net或DiT（Diffusion Transformer）架构来预测每一步应去除的噪声量64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/948658API
curl https://kongchang.com/api/v1/knowledge/claims/948658MCP
get_claim(id=948658)