视频选题榜 2026-07-15

• https://arxiv.org/abs/2607.07386 · Sparse Delta Memory: Scaling the State of Linear RNNs through Sparsity · 用 PKM 稀疏寻址把 GDN 状态容量扩三阶 · isoFLOP 下 1M tokens 长召回显著超 GDN · round=R2
• https://arxiv.org/abs/2607.05394 · Weak-to-Strong Generalization via Direct On-Policy Distillation · 用 policy shift Δ=logπ^RL−logπ^ref 当稠密隐式奖励跨规模蒸馏 RL 经验 · round=R3