Posterior-First Neural PDE Simulation proposes a new paradigm where downstream tasks dictate the minimal problem state to infer, and a posterior over that state conditions future predictions. The approach replaces traditional field-to-future predictors, which collapse distinct latent states, with a task-to-target-to-interface framework that preserves necessary ambiguity. Experiments on synthetic and PDEBench tasks show that posterior recovery improves rollout accuracy, reducing error by 59.4% of the direct-to-oracle gap.
arXiv:2606. 16923v1 Announce Type: new Abstract: Simulation-based inference (SBI) of latent parameters is often hindered by simulator misspecification, the mismatch between simulated and real-world observations caused by inherent modeling simplifications.
The paper introduces SPT-Flow, a continuous‑time world model that samples a single latent per trajectory and aligns a free‑running ODE rollout with demonstration states at multiple readout times. Experiments on LIBERO show that SPT-Flow improves best‑of‑10 PSNR by 2.0–2.4 dB at mid‑trajectory and final frames and reduces mean LPIPS by 23 % compared to a retrained ODEWorld baseline, while maintaining comparable inference cost. The authors attribute these gains to better utilization of trajectory‑associated latent information and demonstrate that learned latent conditioning adds diversity and best‑of‑K performance. whyItMatters":"SPT-Flow demonstrates that careful alignment of ODE rollouts with demonstration states can significantly enhance reconstruction quality and diversity in continuous‑time world models, offering a practical improvement over existing baselines."
arXiv:2609.38902v1 Announce Type: cross Abstract: Augmenting small concurrent studies with external or historical cohorts is attractive in drug development, where enrollment is slow, follow-up is exp...
The paper investigates when in‑context learning (ICL) can succeed even if the instruction or target task is not covered by a model’s pretraining data. It introduces a theoretical framework that distinguishes two types of distribution shift—unsupported instructions and unsupported target tasks—and shows how demonstrations can mitigate these shifts. The authors further extend the analysis to long‑horizon tasks that can be decomposed into pretraining‑supported subtasks, providing error bounds that explain how ICL can handle tasks beyond the model’s original support.
arXiv:2606. 27354v1 Announce Type: cross Abstract: Neural surrogate models offer fast approximate mappings from PDE parameters to solutions, but they typically treat solving as a purely statistical task: once trained, they struggle to correct their own constraint violations and extrapolate beyond the training distribution.
The paper introduces a framework for evaluating how well learned states from neural PDE solvers can be reused in new contexts. By separating solution accuracy, learning contribution, and numerical utility through paired state comparisons and cost accounting, the authors audit 12 papers and conduct experiments on linear PDE families, Fourier neural operators, and convolutional networks. Their results show that matched state comparisons and downstream computational evidence can lead to significant savings in conjugate‑gradient iterations and demonstrate finite‑budget pretraining value.
arXiv:2610.27280v1 Announce Type: cross Abstract: Offline goal-conditioned reinforcement learning enables skill acquisition by reaching any goal from any state without dense rewards. Model-based meth...
The paper introduces Align Latent Rollout (ALR), a regularization technique for neural PDE solvers that encourages consistency across different rollout origins. ALR uses a second‑order Taylor analysis to define two objectives: zeroth‑order consistency aligning latent states at a shared target time, and first‑order consistency aligning local evolution responses to perturbations. The method can be applied to various neural PDE solvers without altering their architectures and demonstrates improved prediction accuracy and reduced discrepancies across rollout origins in multiple benchmarks.
Draft-OPD introduces an on‑policy distillation method for speculative draft models, addressing the mismatch between supervised fine‑tuning and inference by letting the target model supervise the drafter on draft‑induced states. The approach uses target‑assisted rollouts for stable continuations and replays drafting from error positions exposed during verification, enabling the drafter to learn from both accepted and rejected proposals. Experiments demonstrate that Draft‑OPD achieves more than five‑fold lossless acceleration across diverse tasks, outperforming prior draft models such as EAGLE‑3 and DFlash by 23 % and 13 % respectively.
Offline goal-conditioned reinforcement learning enables skill acquisition by reaching any goal from any state without dense rewards. Model-based methods allow for planning by scaling test-time compute...
The paper introduces AdaptiveSpec, a training‑free speculative decoding method that simultaneously adapts the per‑step verification rule and the draft‑tree shape using signals generated during decoding. It replaces the fixed token‑match rule with a margin‑based threshold and adjusts tree depth, width, and node count based on draft confidence and recent acceptance history, allowing the total draft count to vary. Experiments on SGLang show up to 56% throughput gains over EAGLE‑3 while maintaining 93% of lossless task accuracy on GSM8K, MATH‑500, and HumanEval across three models.
The paper challenges the common practice of selecting latent world models based on one‑step prediction loss, arguing that this metric poorly predicts planning performance. Using a frozen‑decoder protocol, the authors decompose one‑step error into encoder‑inherent and predictor‑added components and evaluate goal‑conditioned planning across three continuous‑control environments. Their experiments show that improvements in one‑step fidelity do not correlate with planning success, with correlation coefficients near zero and inconsistent exchange rates across tasks.