Reproduce “Connected Self Forcing: Beyond Local Learning in Video Autoregression” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dongbin Zhang, Chaoda Zheng, Kangjie Chen, Xiangyu Li, Shijia Chen, Jinhao Deng, Yuqi Zhang, Guangfeng Jiang and 8 more
Connected Self Forcing (CSF) trains autoregressive video generation by replaying computations so later chunks can send gradients through earlier cached context to its generator. It supplements direct chunk supervision without retaining the full rollout graph or changing inference. With teacher-forcing initialization, CSF scores 0.8701 on 240-second chunk-wise generation, versus 0.8545 for Self Forcing; its Dynamic Degree is lower than both baselines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.