Science & Mathematics

On-policy self-distillation (OPSD) enables a language model to learn from a Teacher instantiated from the same model but given access to privileged information, such as a reference solution. We find that access to sufficient privileged information allows the Teacher to generate highly accurate trajectories; however, accuracy alone does not guarantee that these trajectories are suitable for Student learning. Direct access to a complete solution may encourage shortened, reference-conditioned derivations that omit the exploration, verification, and self-correction required to solve problems without privileged information. We introduce TeachBack, a bidirectional preference-learning framework for on-policy self-distillation that learns not only from the Teacher but also how the Teacher should teach. TeachBack shapes the Teacher's generation behavior by controlling the length of the reference-reasoning suffix injected into the prompt and by varying the prompt style. While maintaining comparable Teacher accuracy, it encourages the Teacher to explore and re-solve problems, thereby reducing its reliance on the reference solution. At each rollout refresh, the shared policy generates both a privileged Teacher trajectory and a problem-only Student trajectory. A pairwise preference objective uses privileged Teacher trajectories to improve Student learning, while the reversed preference uses the Student's behavior to shape future Teacher supervision under the privileged condition. Experiments on mathematical and logical reasoning tasks show that TeachBack consistently improves Student performance over existing OPSD methods and related baselines.

World action models (WAMs) are shown to substantially diverge under generalization performance based on their generative backbones. We investigate this gap through the representational geometry that WAMs inherit from their backbones and how policy training reorganizes it. Our central hypothesis is that a useful representational prior would structure geometry that separates nuisance information from the in-distribution manifold in advance of training as a policy. We quantify this with nuisance orthogonality; orthogonality of the feature under visual variation outside the empirical in-distribution manifold. Across various image and video backbones, we show that a higher orthogonality from the representational prior is associated with stronger robustness to visual variations after WAM training, enabling policy generalization measurable prospectively for world models. In addition, analyzing the resulting robust WAMs reveals nuisance suppression; relative suppression of feature displacement caused by such variations and robust action generation of the policy. Together, these findings provide a geometric perspective on how the representational prior of the world model shapes the generalization performance of WAMs.

Retargeting human motion to humanoids requires preserving motion semantics while accounting for the robot's physical constraints. Kinematic pipelines prescribe a robot trajectory through geometric matching, without establishing whether it can be executed under the robot's dynamics. Prior physics-supervised neural methods train bidirectional full-sequence models on physics-refined kinematic trajectories, remaining tied to prescribed kinematic solutions and requiring noncausal context. We propose SMR, a streaming neural motion retargeter learned from semantic-conditioned physics experts. We convert human motion into robot-aware task-space targets for body landmark positions, orientations, and velocities. Physics experts track these targets directly in simulation under contact, dynamics, and actuator constraints. We train a single strictly causal Transformer on their executed state trajectories that complete the motion and pass semantic and physical validity checks. At each incoming frame, the student uses only current and past human observations to generate a robot motion reference for a downstream whole-body controller. Experiments demonstrate that semantic-conditioned physics experts generate motion with higher semantic fidelity and physical validity than direct optimization-based retargeting, while the resulting streaming retargeter achieves improved closed-loop performance over kinematic and physics-refined baselines under a common whole-body controller.

Selecting an effective quantum compiler configuration requires coordinated decisions over both hardware targets and compiler settings, with the ultimate goal of improving circuit execution fidelity. Existing SOTA approaches typically decompose this problem into two sequential stages: a device is chosen first, after which compiler options are tuned on that device through RL-based search. This bottom-up strategy can be fundamentally limited by its initial device decision, since a poorly matched backend may constrain the best fidelity achievable regardless of how effectively later compiler options are optimized. In contrast to classical compiler tuning, where the search space is often prohibitively large, the configuration space of quantum compilers remains manageable enough to support exhaustive search. Consequently, rather than repeatedly searching from candidate configurations, one can derive reusable configuration knowledge from offline oracle results and apply it directly to new circuits. In this paper, we introduce a top-down configuration-selection paradigm and build QuOPTBench, an exhaustively searched benchmark containing oracle configurations for 268 quantum circuits. We first study whether LLMs can directly infer suitable configurations and evaluate ten representative LLMs. The results reveal that vanilla LLMs remain far from the oracle. We then employ an LLM to extract transferable rules from sampled circuits governing configuration selection. Using these rules together with predictors trained on the enumerated configuration data of the sampled circuits, we design QuTOP, a rule-guided ML framework that performs fast configuration selection without iterative online search. Across our evaluation, the distilled rules consistently strengthen LLM-based selection, while QuTOP provides the strongest overall balance between solution quality and selection latency among the baselines.

We study how to improve training efficiency on Ascend NPUs. In the NPU training programs we study, framework-level compilation optimizations such as operator fusion are only partially applied, leaving fragmented computation and avoidable execution overheads. Profiling-guided tuning can address these inefficiencies, but it often requires costly trial and error. To make training on NPUs more efficient, we propose TrainBridge, an agentic approach for transferring GPU optimization guidance to NPU training. TrainBridge uses the highly optimized execution produced by mature GPU compilation stacks as a task-specific reference for NPU tuning. It extracts optimization intents from the GPU training execution to guide NPU program rewrites and further refines the rewritten program through iterative profiling. Our GPU-to-NPU migration experiments show that TrainBridge achieves an average 1.7x speedup across four workloads covering LLM pretraining, image classification, and vision-language training, and an average speedup of nearly 2.0x across 27 Modded-NanoGPT speedrun records with highly optimized GPU implementations.

Conditional Flow Models (CFMs) are increasingly being used in place of deterministic predictors across machine learning, including for tasks that were classically formulated as regression. Given their success in such tasks, we ask the question: what useful information do the conditional distributions produced by CFMs carry beyond what a point predictor would provide? We identify a systematic phenomenon: in many natural regimes, CFM samples routinely have larger variance than the true conditional distribution, even in nearly deterministic regression problems. This effect is most pronounced where the model has not resolved the dependence on the conditioning variable, and thus the additional variance can be viewed as converting epistemic uncertainty in fitting the ground truth conditional distribution into apparent aleatoric variance in the generated samples. Through controlled synthetic and real-data experiments, and comparisons with explicit smoothing estimators, we show that the induced variance tracks model underfitting of complex or hard-to-fit targets, and decreases with additional training or data. We provide a theoretical account of the phenomenon in a simplified setting and demonstrate a consequent practical use-case of conditional flow-matching: designing high-coverage predictive sets that cover the ground-truth regression target.

Multimodal large language models (MLLMs) have made rapid progress in general visual understanding, yet they still struggle with fine-grained instance-level consistency verification, where the goal is to determine whether two images depict the exact same physical object under varying viewpoints, illumination, and backgrounds. This task requires both global semantic perception and precise examination of subtle local evidence, since semantically similar objects may differ only in small suspicious regions. To address this challenge, we propose LocalAgent, a collaborative agentic framework that integrates the complementary strengths of a generalist MLLM and lightweight specialist verifiers. The MLLM first reasons globally to identify suspicious local regions, which are then delegated to a Local Evidence Verifier (LEV) toolkit for fine-grained appearance and geometric matching. LEV comprises Local Appearance Consistency (LAC), which estimates appearance similarity under illumination and color changes, and Local Geometric Correspondence (LGC), which estimates spatial correspondence under viewpoint changes. LEV returns structured scalar evidence that the MLLM integrates for the final decision. We further introduce an auxiliary-capped reinforcement fine-tuning (RFT) strategy, in which answer correctness dominates auxiliary tool-use and format signals, enabling the policy to learn when verification is necessary. Experiments show that LocalAgent improves mean accuracy by 4.9% over the strongest CoT-RFT baseline while invoking a verifier on only 12.13% of samples and preserving the general multimodal capabilities of the underlying MLLM. The open-source repository is available at https://github.com/Firefly-zoo/LocalAgent-main.

The Regular Syndrome Decoding (RSD) problem is a structured variant of classical Syndrome Decoding and underlies a growing number of efficient cryptographic constructions, including zero knowledge protocols, pseudorandom correlation generators, and MPC in the Head signatures. Although RSD is generally believed to be computationally hard, recent works have shown that algebraic attacks can outperform information set decoding on important RSD parameter regimes. Building on this line of work, we revisit the concrete hardness of RSD and show that many widely used parameter sets are substantially easier to solve than previously estimated. This improvement stems from a refined hybrid algebraic attack that exposes and removes a previously overlooked source of concrete complexity: the joint blockwise placement of zero coordinate guesses and syndrome based variable eliminations. Rather than treating the two guessing stages and linear elimination as largely separate choices, our method coordinates them across all regular blocks so that the guessing penalty is balanced while sufficient equations are retained for linearization. This gives a systematic block balancing principle for the entire guess reduce solve pipeline. We reassess several RSD based applications, including Ferret, Wolverine, and committed VOLE. Our estimates place at least one parameter set per application below its 128 bit target, with losses approaching 67 bits. On PCG benchmarks, we improve previous best attack costs by up to 29.77 and 25.91 bits over non binary and binary fields, respectively. Moreover, for the revised 128 bit PCG parameter sets of Liu et al. evaluated here, estimated attack costs are all below $2^{111}$ field operations. These results substantially revise the concrete security landscape of RSD and indicate that several existing parameter choices overestimate the hardness of their underlying RSD instances.

In video action detection, modeling the temporal dependencies in continuous video streams is crucial. Previous methods based on deep neural networks can perform frame-level video recognition tasks. However, these methods struggle to model dynamic temporal patterns, facing challenges such as redundant frame information, difficulty balancing global and local features, and insufficient fusion of visual and textual features. To address these issues, we propose a video temporal attention module based on the Multi-Scale Latent Temporal Compression Attention, which aggregates continuous temporal sequences into feature representations across diverse temporal scales while efficiently reducing redundancy. We also introduce a Video-Text Cross Attention Fusion module that leverages cross-attention at both latent and standard feature levels, achieving fine-grained fusion between video and textual features to enhance semantic alignment and feature interaction. Experimental results across multiple video recognition benchmarks demonstrate that our proposed methods achieve superior performance compared to existing methods, validating the effectiveness and generalization capability.

We propose a learning-based generative model for completing curvilinear features under occlusion by environmental obstacles. To this end, we adopt a flow matching-based approach that supports the inpainting of occluded regions with the appropriate curvilinear features. To eliminate the reliance on privileged information about the occlusion a priori, we also train a UNet-based mask prediction model, which can identify regions for inpainting given only the occluded observation. The components of our approach are validated via experiments using procedurally-generated test cases, and the complete pipeline is validated using large public datasets for curvilinear objects. Against off-the-shelf inpainting baselines, ROCOCO is the strongest method by a wide margin when no occlusion mask is given, and is competitive with mask-supervised inpainting when one is, achieving best on full-image error and second only to LaMa on the occluded region, while using a completer of $7.7$M parameters against LaMa's $51$M and Stable Diffusion's $860$M.

Before an AI agent takes a consequential action, such as paying, messaging or changing code, several monitors check it. Their verdicts are combined by aggregating scores or by counting approvals, whatever inputs the monitors share. Monitors that share a base model or a poisoned input, however, may fail together. We borrow the fail-prone systems of distributed computing: lists of monitor sets that may fail together. A rule is safe if it executes no bad action and live if it blocks no good one. We measured twelve small open-weights monitors on payment approval under two pre-registered protocols. Without pre-registration, we also measured three closed frontier models (Claude Opus 5.5, GPT-6 Astra, Gemini 4 Argon) and the open DeepSeek-V3.2. Every reliable fooling of a monitor into approving a fraud came through an evidence source and reached only that source's readers. Fooling it into blocking a correct payment depended on injection resistance, a monitor's refusal to follow instructions planted in its data. Small models and DeepSeek obeyed such notes. The three closed models ignored them in all but one case, and Opus also resisted persuasion tailored to it. On new data, a rule that counts sources instead of votes blocked 0.9% of correct payments, against 7.4% for the best counting rule, at equal safety. The pre-registered worst-case test failed. One attack left the approvers on a single source, so no rule over the twelve language-model monitors survives every measured attack. Adding a rule engine, which cannot be persuaded, restored the condition for a safe and live rule on the measured attacks. Diversify the inputs, harden the monitors, count sources, not votes.

EuPd2Si2 serves as an ideal system to probe the interplay between lattice dynamics and electronic instabilities through its tunable valence transition, responsive to temperature, pressure, and magnetic field. Here, we focus on investigating the origin of the temperature-induced europium valence instability in pristine EuPd2Si2, alloyed EuPd2Si1.95Ge0.05, and EuPd2Si1.9Ge0.1 and its correlation with structural responses. Compared with the parent compound EuPd2Si2 which exhibits the europium valence transition at Tv = 143 K, magnetic susceptibility measurements reveal a systematic shift in Tv with Ge substitution, decreasing to 87 K for EuPd2Si1.95Ge0.05. Temperature-dependent synchrotron X-ray diffraction confirms phase coexistence between Eu2+Pd2Si2-xGex and Eu3+Pd2Si2-xGex phases across the valence transition for all samples measured (x = 0, 0.05, 0.10). The pronounced increase in the full width at half maximum (FWHM) well above Tv reflects the gradual nature of the transformation and highlights the role of structural strain in stabilizing the coexisting phases, supporting the presence of a phase-segregated ground state, in contrast to a homogeneous intermediate-valence or mixed-valence state. The variation in Tv across different compositions correlates with changes in lattice parameters, highlighting the robust coupling between the structural and electronic degrees of freedom. The phase diagram suggests an inverse linear relationship between the lattice parameter and Tv, suggesting a possible predictive framework for estimating transition temperatures in other chemically substituted variants of this family.

Executing vision-language models (VLMs) aboard low-Earth orbit satellites enables flexible processing of observation data without first downlinking it to the ground. On commercial off-the-shelf payloads, however, a request may span multiple execution windows: power and thermal constraints, fault-triggered reboots, and competing payload demands can interrupt inference before it completes. Each shutdown discards volatile activations and KV caches, forcing an unprotected request to repeat work after restart. We present OrbitWeaver, a framework that preserves VLM inference progress across onboard interruptions through coordinated checkpointing and recovery. Its phase-aware checkpointing selects recovery boundaries according to the state lifetimes and computational characteristics of vision encoding, language-model prefill, and decoding, balancing storage traffic against recomputation. Its dependency-aware dynamic recovery schedules restoration from partially completed states across the layer and token-chunk dimensions. By overlapping checkpoint and weight reads with recomputation, it reduces restart latency; the recovery planner computes an optimal execution order under its scheduling model within 100 ms. Experiments on satellite payload hardware using on-orbit telemetry traces show up to a 4.07-fold end-to-end speedup and a 3.56-fold reduction in SSD write volume over the evaluated baselines. A 72-hour deployment on two low-Earth orbit satellites demonstrates that OrbitWeaver retains inference progress and resumes execution across repeated bus-enforced shutdowns.

We study labeling schemes for $p \times q$ $Γ$-free binary matrices, which do not contain a submatrix of the form $Γ= \begin{pmatrix} 1 & 1 \\ 1 & 0 \end{pmatrix}$. We give a labeling scheme that uses $O(\log^2 N)$-bit labels for rows and columns, where $N=p+q$, such that any matrix entry can be determined from the corresponding two labels in constant time in the word RAM model. The worst-case label size matches the counting lower bound up to constant factors. We also show that the entrywise least $Γ$-free supermatrix of an arbitrary $p\times q$ binary matrix, together with its labels, can be constructed in $O(pq)$ time. Our construction represents a $Γ$-free matrix by a sequence of rooted forests, where each column corresponds to a node and the $1$-entries of each row form a root path in its corresponding forest. For $n$-vertex strongly chordal and chordal bipartite graphs, our result gives asymptotically optimal $O(\log^2 n)$-bit adjacency labeling schemes with constant-time queries. These results answer both representation questions posed by Spinrad in 1995.

Root cause analysis (RCA) and mitigation prediction are essential for incident management, enabling the identification of underlying causes of failures and ensuring the reliability of complex cloud-based systems. With rising incident volume and system complexity, incident reports have become a valuable resource for these tasks. Recent studies have shown promising results using Large Language Models (LLMs) to predict root causes and mitigation actions from textual descriptions in incident reports. Despite these advances, further progress is hindered by the lack of publicly available datasets; existing work relies on private data, which limits transparency, reproducibility, and generalization. To address this gap, we introduce CloudIR-Bench, the first publicly available benchmark providing a dataset and evaluation framework for incident report-based RCA and mitigation prediction. Curated from incident reports across five leading cloud-based systems, CloudIR-Bench captures a diverse range of real-world production scenarios representing authentic failures and engineers' reasoning. We also release an open-source evaluation framework that includes eight recent LLM-based approaches. Our comprehensive evaluation includes zero-shot and in-context learning experiments, along with a qualitative analysis to identify where and why LLMs fail. Benchmarking results reveal the capabilities and limitations of existing LLMs in this domain. The benchmark, including the dataset, framework and supplementary materials, is available at https://github.com/hoangtnm/CloudIR-Bench.