Unverified50% confidenceFactExact time
文本引导的去噪过程本质上是Classifier-Free Guidance（CFG）机制在起作用，模型同时预测有条件和无条件的噪声估计，两者差值被放大后引导生成方向
1
Sources
50%
Confidence
Long-term
Relevance
8/5/2026
First Seen
Sources
Related Claims
UnverifiedThe formula for Classifier-Free Guidance is: predicted noise = unconditional prediction + guidance strength × (conditional prediction − unconditional prediction), and the higher the guidance strength, the more the output deviates from the natural distribution.75% similarVerifiedDiffusion models circumvent the instability of adversarial training through a Markov chain mechanism that progressively adds Gaussian noise and then trains the model to learn the reverse denoising process.71% similarUnverified扩散模型的训练过程分为加噪与去噪两个阶段，推理阶段的条件可引入额外视觉信号（如参考视频、深度图、骨骼图）约束去噪方向70% similarUnverifiedHuBERT使用离线聚类算法（如k-means）为音频帧生成伪标签，训练模型预测被遮蔽帧的聚类ID68% similarUnverified过拟合的统计原理是当模型参数量相对训练数据过多时，模型会记住噪声而非学到规律68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/689123API
curl https://kongchang.com/api/v1/knowledge/claims/689123MCP
get_claim(id=689123)