Tag
Salt++ 是一个两阶段后训练框架，通过 CausalSelf-Flow 和上下文对齐的自回归 DMD，解决少步流式音视频生成中教师强制与因果分布蒸馏的上下文错配问题，在 480p 4-step 设定下相对 OmniForcing 提升视觉质量 57%、运动质量 45%。
ViRDM is a method for few-step causal video generation using representation distribution matching, achieving better results than DMD-based baselines with efficient training on single or multiple GPUs.
PixVerse R2 introduces a unified scaling architecture for real-time audiovisual world models, leveraging block-sparse attention and continuous pretraining to enhance video generation and interactive control.
The paper proposes a deterministic framework that compiles games specified in VGDL into Dynamic Structural Causal Models to guarantee causal fidelity, enabling transparent causal pathways for counterfactual reasoning and causal reinforcement learning.
This paper provides a mathematical analysis of the Jacobian lens (J-lens) for reading verbalizable representations from language models, identifies its sparse causal geometry, and proposes improvements to enhance concept readout.
Domino is a speculative decoding framework that decouples causal dependency modeling from autoregressive drafting, using a parallel backbone and lightweight causal refinement head to achieve up to 5.49× end-to-end speedup on Qwen3 models.
This paper provides an algebraic formalization of the Theory of Dyadic Morality using structural causal modeling, and demonstrates applications to AI policy design.
CausalCine is a new academic framework for real-time, interactive multi-shot video generation that uses causal modeling and dynamic memory routing to improve cross-shot coherence in autoregressive models.