diffusion rec

data

kuavo_traj_q_v_tua

what matters in robot imitation learning ?
diffuision scheduler and predict_net is 解耦
directly use data or use latent of data
some acknowledges about nn.parameters()
a model archtecture picture about diffusion policy
details about imitation learning data flow in training
naive bayes 在对两个高斯分布时，naive bayes 是logistic回归的特殊情况
什么是自归回
nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding), nn.Conv2d, nn.Linear(),
Conditional Denoise:

P(x_{t-1}|x_t) = ||\epsilon - f_{\theta}(x_t, t, \text{condition})||, \text{condition} \in \{\text{text}, \text{CLIP}, \text{classifier}, \text{image}\}