~ similar to 2606.13240· 20 results
DASH introduces a dual-branch distillation framework to effectively compress class-conditional diffusion models by independently supervising both score branches, significantly preserving guidance fide…
Renhao Zhang, Haotian Fu, Mingxi Jia, George Konidaris +2 more
The Parameterized Diffusion Policy (PDP) framework transforms diffusion models from general stochastic generators into precise, steerable tools for learning and adapting complex robotic behaviors by e…
This paper introduces a novel method to improve the performance of finite-dimensional diffusion policies by lifting them to a Cameron-Martin space and using a backward Kolmogorov equation.
This paper establishes a connection between neural network approximation of score functions and approximation of probability distributions generated by reverse diffusion models.
This paper proposes two strategies to improve feedback efficiency of reinforcement learning from human feedback (RLHF) in diffusion models.
The paper proves that diffusion models adapt to low-dimensional structure robustly across various update coefficient choices.
Bingnan Li, Haozhe Wang, Haozhong Xiong, Fangtai Wu +4 more
This paper introduces Positive--Direction Matching (PDM), a branch-aware on-policy distillation objective to address Negative Branch Asymmetry (NBA) in diffusion models under classifier-free guidance.
This paper proposes a method to stabilize training of flow-matching models in reinforcement learning by calibrating importance-ratio behavior based on the predicted path variance.
Hongru Hou, Tiehua Mei, Denghui Geng, Jinhui Huang +4 more
The paper proposes ProRL, an effective Reinforcement Learning framework that rectifies gradient estimation deficiencies to optimize proactive recommendation paths, significantly outperforming existing…
This paper develops an adjoint-sensitivity framework for positional influence in causal residual Transformers and provides theoretical results on the residual-to-depth-flow estimate and finite-token-t…
This paper explores generative modeling of humanoid control networks through their weights, enabling the creation of deployable networks for various motions.
This paper introduces a variance-weighted batch distribution to improve the diversity of samples in diffusion models.
The paper introduces MDM-VGB, a reward-guided sampler for Masked Diffusion Models, which extends the Jerrum-Sinclair backtracking Markov chain to a masked-state graph for effective high-reward generat…
The paper repurposes a pre-trained speech classifier as the backbone for diffusion generation, reducing the need for two separately trained models.
The paper proposes a unified, constrained optimization framework using KL divergence and likelihood constraints to achieve effective and principled unlearning in diffusion models.
This paper investigates the effect of pseudorandom number generator orbits on diffusion models' training and generation, revealing learnable input and significant impact on loss and quality.
Zehan Zhang, Neng Zhang, Yaoyi Li, Jia Cai +1 more
The paper proposes a diffusion-based planning framework called Diffusion Forcing Planner (DFP) to stabilize motion plans in autonomous driving.