Self Gradient Forcing: Native Long Video Extrapolation
Recent autoregressive video diffusion methods are increasingly built upon Self Forcing, where the student is trained on histories produced by its own rollout rather than ground-truth video contexts. This reduces exposure bias, but the historical key-value cache is still used by future frames only as frozen rollout state. As a result, future losses cannot supervise how earlier generated latents should be written into more useful keys and values for later video-latent generation. We call this the historical context-gradient gap. We propose Self Gradient Forcing (SGF), a two-pass training strateg
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- FuzzyOverlapping authors or contributors · 62%keras-team/keras →
“Shared author/contributor keys: jin”
- FuzzyOverlapping authors or contributors · 62%HKUDS/LightRAG →
“Shared author/contributor keys: jin”
- FuzzyOverlapping authors or contributors · 62%modular/modular →
“Shared author/contributor keys: liu”
- FuzzyOverlapping authors or contributors · 62%sgl-project/sglang →
“Shared author/contributor keys: luo”
- FuzzySimilar title/name (fuzzy) · 59%Developer-Y/cs-video-courses →
“Fuzzy title match (0.73): “Self Gradient Forcing: Native Long Video Extrapolation” ≈ “Developer-Y/cs-video-courses””
- LinkedLinked via arxiv author · 85%Xianglong He →
“Self Gradient Forcing: Native Long Video Extrapolation”
- LinkedLinked via arxiv author · 85%Xuying Zhang →
“Self Gradient Forcing: Native Long Video Extrapolation”
- LinkedLinked via arxiv author · 85%Haoran Li →
“Self Gradient Forcing: Native Long Video Extrapolation”