Bingran You, Wenjun Ke, Ria Rosenauer, Bo Liu, Wenbo Chen, Min Chen, Shuqi Xu, Yue Ma 等 72 位作者
AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality alongside correctness. We introduce FrontierPhysics, an end-to-end benchmark with 56 tasks made by 42 physicists based on their own research, covering nine domains across theoretical, computational, and experimental physics. Each task requires agents to review the literature, develop a research plan, conduct research and experiments, and produce structured outputs and scientific manuscripts. Combining deterministic verifiers with expert-authored rubrics, FrontierPhysics evaluates whether agents' research is correct and publishable. Across 22 models in 26 configurations, no configuration passes more than 46 of its 168 rollouts. Manual review of agent trajectories shows that frontier agents' main weaknesses are insufficiently rigorous reasoning and limited expert intuition, rather than gaps in physics knowledge or execution errors.
Safe reinforcement learning (RL) aims to learn policies that optimize rewards while satisfying constraints. Predominant approaches rely on soft-constrained policy optimization, which has achieved empirical success but does not provide formal safety guarantees for the learned policy. In contrast, methods with strict guarantees typically rely on explicit certificate functions, whose construction requires the direct synthesis and verification of control-invariant sets, a process that scales poorly with state dimension and often yields overly conservative behavior. In this paper, we present the Provably Safe, yet Scalable RL (PS2-RL) framework, a novel two-phase architecture for learning provably safe policies in a scalable manner, designed to overcome the key bottlenecks of prior methods. Rather than explicitly computing invariant sets, PS2-RL leverages a learned backup policy to forward-integrate the system dynamics, generating an implicit control-invariant set online. In the first phase, the backup policy is trained with the safe-arrival value function to achieve rapid, safe recovery to the certified base set for invariant-set construction. In the second phase, an RL policy is trained end-to-end through a differentiable projection layer that strictly enforces the safety guarantees induced by the learned backup policy. By maximizing the volume of the implicit control-invariant set in the first phase, the resulting PS2 policy from the second phase is performant and scalable, while maintaining provable safety. Crucially, PS2-RL imposes no restrictions on the underlying RL algorithm and can be plugged into any existing training pipeline. We establish theoretical guarantees for the proposed framework and evaluate it on robotic control tasks with state dimensions up to 10, a regime in which prior provably safe RL methods struggle or become impractical.
Xiang 'Anthony' Chen, Savvas Dimitrios Petridis, Ruofei Du, Yang Li
Generative user interface (GenUI) tools use large models to create UI mockups from high-level descriptions; a recurring interaction paradigm starts with an unstructured prompt, followed by depth-first generation of a polished, high-fidelity prototype. To examine how it supports early-stage, 0-to-1 design exploration, we contrast this paradigm with structured input, breadth-first workflow, and low-fidelity generation. In our study, 24 industry professionals in early-stage product and UX design used a commercial GenUI system and our research prototype as study-era instances of each paradigm for a comparative evaluation. Findings show that 1) structured input surfaces key facets but raises entry barriers; 2) breadth-first workflow reveals more possibilities, but previewing multi-screen UX ideas remains hard; and 3) despite low fidelity's value, participants rated high fidelity as better suited to their design stage and to communicating ideas. Our findings contribute a model-agnostic, design-level account of how GenUI users specify intent, traverse alternatives, and interpret fidelity.
Chang Liu, Ke Han, Niccolò Biondi, Gloria Isotton, Elisa Ricci, Davide Talon, Nicu Sebe
Sign language generation (SLG) maps written sentences into signing motion while preserving both lexical content and fine-grained articulation. This requires a model to decide what signs should be expressed and to realize them through precise handshape, trajectory, facial expression and body coordination. Existing pose-based systems and reconstruction-only motion tokenizers often entangle these two decisions, leaving the generator without an explicit text-aligned intermediate target for what to sign before refining how to sign it. To address this issue, we introduce Semantic Residual Vector Quantization (SRVQ), a framework for SLG that frames generation as a two-level process moving from semantics to kinematics. The newly introduced SRV Quantizer learns a hierarchy in which the first residual level encodes sign content and subsequent levels refine pose, handshape, trajectory, and body coordination, all within a single quantizer and without separate semantic representations or external planning stages. Through word-aware conditioning for hierarchical code generation, a masked semantic generator predicts semantic codes and a kinematic generator completes the residual codes for full SMPL-X motion synthesis. We evaluate using retrieval and FID in a shared text-motion embedding space, alongside standard geometric metrics. On CSL-Daily and Phoenix-2014T, SRVQ achieves R@1 of 82.3 and 70.3, outperforming the previous SOTA by 30 and 46 points, respectively.
Algorithm-driven services, including search, recommendation, and advertising, rely on multi-stage model pipelines. While research prioritizes model optimization, post-deployment monitoring focuses on operational reliability. Yet operationally healthy systems may still produce poor-quality results, exposing a gap in existing evaluation and monitoring practices. However, a systematic problem formulation and benchmark for this task remain lacking. To bridge this gap, we formulate System Quality Diagnosis (SQD) for algorithm-driven services, a new task that extends quality evaluation beyond identifying poor outcomes to attributing their causes across system components. Focusing on industrial search, we propose WISE, an agentic framework for end-to-end quality diagnosis that discovers user-facing issues and traces their origins to underlying algorithmic components. To improve diagnostic reliability, we introduce ReTrace for constructing evidence-consistent trajectory supervision and Adaptive Harness-Policy Optimization (AHPO) for failure-guided optimization of the agent harness and model policy. To enable systematic evaluation, we construct WISE-BENCH, a real-world multimodal search benchmark that combines user-facing evidence, internal pipeline traces, and expert diagnosis annotations. Extensive experiments show consistent gains across model scales, with WISE achieving 75.20% end-to-end accuracy on Qwen3.5-35B-A3B. In production, WISE supports quality auditing, online quality assurance, and model optimization, reducing diagnosis costs and improving downstream relevance models.
Strong single-turn editing performance does not guarantee reliable multi-turn interaction: as edited outputs are repeatedly reused, small unintended changes can accumulate into severe visual drift. This long-term drift is often mitigated through training on multi-turn data or cross-turn rollouts, potentially altering pretrained priors, or through specialized inference-time mechanisms that are difficult to scale to foundation models. We introduce TailSteer, a final-stage training framework that improves multi-turn robustness while retaining the generative and editing capabilities acquired through earlier training. TailSteer follows a simple principle: keep early denoising dynamics close to those of the base model and concentrate preservation corrections near the trajectory tail, where fine-grained deviations can be suppressed. These corrections are learned from the editor's own on-policy denoising rollouts, using unchanged source content as a preservation reference without requiring curated ground-truth editing pairs. Experiments across the FLUX, SenseNova, and Qwen families show that TailSteer substantially reduces cumulative drift over 15-20 editing turns while largely preserving base-model performance on extensive text-to-image and single-image editing benchmarks. These results demonstrate that long-horizon editing robustness can be unlocked in foundation models after TailSteer, rather than by relearning their existing capabilities.
Muon has recently demonstrated strong empirical performance in large language model training, but the theoretical role of momentum in Muon remains unclear. Existing analyses of Muon either remove momentum to study spectral updates in isolation, or retain momentum without explaining why it improves empirical performance. Our work bridges this gap by showing momentum in Muon acts as a spectral filter. Under a structured signal-plus-perturbation gradient model, we prove that momentum suppresses perturbations while preserving the dominant signal, thereby enlarging the spectral gap between them. This enlarged gap stabilizes the singular subspaces of the matrix passed to Muon's orthogonalization step, making the resulting update more reliable. We further show that applying momentum before orthogonalization achieves provably stronger alignment with the signal component of the gradient than either reversing this order or simply removing momentum. Experiments across diverse tasks, including LLM pretraining, support our theoretical analysis. More broadly, our theory offers a starting point for understanding the benefits of momentum in other matrix-based optimizers.
Jingyan Shen, Ang Li, Salman Rahman, Yifan Sun, Micah Goldblum, Matus Telgarsky, Pavel Izmailov
Reinforcement learning (RL) has become central to improving large language models (LLMs) on complex reasoning tasks, yet RL post-training is largely studied in isolation from the pretraining that precedes it. As a result, two basic questions remain open: (1) how do pretraining choices (model size, data) shape the returns to RL compute, and (2) what does RL actually do to the model? These questions are difficult to study in the standard LLM setting: pretraining corpora are vast and uncontrolled, making it hard to attribute behaviors to pretraining versus RL, and systematic compute sweeps across both stages are prohibitively expensive. To address these challenges, we use chess as a controlled testbed for studying reasoning across the full pretraining-to-post-training pipeline. We follow the standard LLM training pipeline by pretraining language models from 5M to 1B parameters on human chess games, supervised fine-tuning on synthetic reasoning traces, and running RL on chess puzzles with verifiable rewards. Using this framework, we find that the post-RL performance at given RL compute level is well-predicted from the pretraining loss, and slope of the RL reward curves improves approximately linearly with the pretraining tokens. Beyond scaling, we find that RL does not simply sharpen the SFT policy: on easy puzzles it amplifies correct moves the SFT policy already preferred, while on hard puzzles it surfaces correct moves that were nearly absent under SFT. We further test whether our findings transfer beyond chess by training a 1B language model on math-domain text, where the same predictive pattern emerges: longer-pretrained checkpoints reach higher post-RL performance and improve faster under RL. In sum, we provide a quantitative account of the pretraining-to-RL interface and a controlled testbed for studying the science of reasoning across the full pretraining-to-post-training pipeline.
NVIDIA, :, Anis Ahmad, Jon Allen, Malik Aqeel Anwar, Jesse Archer, Maciej Bala, Piotr Barejko 等 90 位作者
We present a shared framework that adapts Cosmos world foundation models for real-time interactive world simulation in Physical AI. By streaming visual observations in response to actions and scene controls, the framework enables users and embodied agents to interact with evolving virtual environments. It supports both predicting the visual consequences of actions and generating observations from physical state maintained by an external simulator. We demonstrate this approach across camera navigation and robot manipulation, controllable video transfer, and multi-view autonomous vehicle (AV) simulation. These examples provide a shared foundation for developing interactive virtual environments for Physical AI training and evaluation. Our code and model checkpoints are available at https://github.com/nvidia/cosmos and https://huggingface.co/collections/nvidia/cosmos3-examples. The project website is available at https://research.nvidia.com/labs/cosmos-lab/cosmos-simulation.