RB-LDC：面向扩散模型的冗余均衡潜空间编码
RB-LDC: Redundancy-Balanced Latent Coding for Noise-Robust Diffusion Latents
这篇论文把扩散模型的加噪过程按解码器敏感度重新分配，改动只是加一层可逆线性变换，却让带噪重建的 FID 平均降了 17.1。做图像生成或潜空间传输的开发者可以看看，谱求解器有解析解，方法可复现。
一项新研究提出 RB-LDC，在 VAE 编码器与扩散过程之间加入一个可逆线性层，按解码器对各方向的敏感度重新分配冗余。作者推导出白化潜变量下 LMMSE 恢复风险的精确谱表达式，在能量预算和可逆性约束下，通过确定性谱求解器最小化按解码器敏感度加权的风险，得到唯一最优分配。
在大预算情形下，分配遵循敏感度的幂律，指数由损失加权的低 SNR 尾部决定：cosine 调度下噪声预测为 1/2，数据预测为 2/3，截断调度则出现交叉点。在 ImageNet 256x256 上，使用 SD-VAE 和 VA-VAE 训练的 RB-LDC 在 15 个 SNR、三种流水线种子、五种编码配置下全面优于等预算各向同性编码，VA-VAE 分块局部编码下扰动 FID 平均降低 17.1。
论文还报告了部分中间 SNR 下扰动 FID 变差的情形，并验证在冻结扩散模型下，12 种采样进度与信道 SNR 组合的损坏采样中 FID 点估计均更低。
RB-LDC: Redundancy-Balanced Latent Coding for Noise-Robust Diffusion Latents
Latent diffusion models decode approximately correct latents, making decoder sensitivity to perturbations relevant to image quality. After whitening, forward diffusion corrupts every direction at the same signal-to-noise ratio (SNR), despite unequal decoder sensitivities. We study an invertible linear layer between the encoder and diffusion process that shifts each direction's log-SNR. Its linear minimum mean-square-error (LMMSE) recovery risk has an exact spectral expression for any whitened latent distribution. Under an energy budget and a floor ensuring invertibility, minimizing decoder-sensitivity-weighted LMMSE risk integrated over the diffusion schedule yields a unique optimal allocation. The layer aligns with the decoder's expected pullback metric and equalizes marginal risk reduction across directions, computed by a deterministic spectral solver. At large budgets, the allocation follows a power law in sensitivity whose exponent is determined by the low-SNR tail of the loss weighting: one half for noise prediction and two thirds for data prediction under the cosine schedule, with a crossover on truncated schedules. We instantiate this principle as RB-LDC using Jacobian-based sensitivity estimates. On ImageNet at 256 x 256 resolution with SD-VAE and VA-VAE, trained RB-LDC improves reconstruction from noisy latents over budget-matched isotropic coding in all four metrics and five coding configurations, averaged over 15 SNRs and three pipeline seeds. It reduces average perturbed FID by 17.1 for VA-VAE block-local coding while preserving clean reconstruction. We also report intermediate SNRs where perturbed FID worsens. With a frozen diffusion model, RB-LDC yields lower FID point estimates in all 12 tested combinations of sampling progress and channel SNR under corruption during sampling.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include 扩散模型, Latent Diffusion, VAE, ImageNet.
Source attribution: curated from arXiv cs.LG (@Hyunseok Jeong, Jaeho Jeon, Young-Sik Kim).
