arXiv：cs.LG· Achyuthan Sivasankar·· 3 小时前
深度能否在组合预测中存活，路由器能否利用它？潜空间世界模型中的自适应计算
Does Depth Survive Composition, and Can a Router Use It? Adaptive Compute in Latent World Models
AI 导读
研究在九个 DeepMind Control 任务上用十二 block 的 early-exit 潜空间预测器检验自适应计算的两个假设：更深的预测器是否更准确、能否提前判断哪些步骤需要深度。
正文
Abstract:Adaptive-compute world models run fewer predictor blocks on some rollout steps than on others. This rests on two assumptions: that a deeper predictor gives a more accurate rollout, and that the steps which need depth can be identified before the predictor runs. We test both on nine DeepMind Control tasks with a twelve-block early-exit latent predictor, eight seeds per task. The first assumption fails on three tasks, where a shallower exit rolls out more accurately than the full stack on every seed, and composing predictions over ten steps moves the depth-error relation away from its one-step shape in both directions. In hindsight, choosing depth per step helps: a greedy oracle that sees the true next latent has 10-37% lower rollout error than the best mixture of fixed depths at equal average compute, and 8-28% lower than the best step-index schedule, on all nine tasks. No router we trained captures this. The difficulty router these models carry stays within 5% of fixed-depth mixtures; a router trained to predict each exit's error beats them by up to 12% but no more than a fixed schedule does; and in closed-loop planning at equal compute the routed planner matches a fixed middle depth. On cheetah, where depth 2 rolls out more accurately than the full stack (an inversion), supervising the early exits only at the first rollout step removes the inversion, while a control with the same loss weight that supervises them at one random step keeps most of it (rho 0.91 vs 0.85 standard and 1.14 first-step-only, three seeds): the inversion depends on supervising exits from composed latents. These results change with the encoder, backbone, horizon and training data. The fixed-depth campaign was pre-registered; the routing analyses, planner runs, ablation control and one-step decomposition are exploratory.
| Comments: | 26 pages, 12 figures, 12 tables. v4: major revision. Adds equal-compute routing comparisons (oracle, step-index schedules, three trained routers), equal-compute planner runs and a supervision-matched ablation control; narrows several v3 claims accordingly; corrects four perturbation-baseline numbers in v3 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| ACM classes: | I.2.6 |
| Cite as: | arXiv:2607.10203 [cs.LG] |
| (or arXiv:2607.10203v4 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2607.10203 arXiv-issued DOI via DataCite |
Submission history
From: Achyuthan Sivasankar [view email]
[v1]
Sat, 11 Jul 2026 08:27:33 UTC (102 KB)
[v2]
Tue, 14 Jul 2026 14:26:03 UTC (102 KB)
[v3]
Sat, 18 Jul 2026 16:02:20 UTC (112 KB)
[v4]
Wed, 14 Oct 2026 02:15:14 UTC (178 KB)
来源：arXiv：cs.LG · arxiv.org