DeepJEPA：通过内部分配计算深度扩展世界模型

1. 12:00
   arXiv · Robotics

   DeepJEPA：世界模型的内部计算缩放新方法

   summary_zh: DeepJEPA 提出一种权重共享的联合嵌入预测世界模型，将过渡深度作为测试时缩放轴，学习判断何时值得对候选和回放步骤进行递归更新。在五个视觉控制任务中，DeepJEPA 以每步平均仅 1.00-1.26 次更新匹敌或超越最强固定深度规划器，计算集中在接触开始和持续物体交互等决策关键事件上。

Current heat 8·Comparable peak 10（Oct 2）·Comparable change over 24 hours –

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.