Papers

We present a study of the quality of individual DBpedia triples from the perspective of Natural Language Generation, and propose and evaluate an approach for collecting entity-specific triple sets that filters out questionable triples while minimizing the loss of correct ones. We show in an evaluation against manually annotated data that with validation rules, it is possible to reach 98% precision in triple selection, and with improvements to a few Property definitions, it is possible to improve recall by 40% without harming precision.

Software repositories contain vast amounts of data on code contributions, bug reports, and project activities, yet this information remains challenging for non-technical stakeholders and developers to access due to limited expertise in querying repositories. To address this, we introduce a novel chatbot architecture leveraging OpenAI's GPT-4 model for automated extraction and analysis of repository data. In contrast, our architecture takes a structured path first by parsing the user's query to extract relevant parameters, then selecting the correct tool to employ based on that analysis, and finally invoking the GPT-4 model to create a highly detailed response. In contrast to previous work based on multi-component systems with embedding models and document retrievers, our architecture inverts the process by relying on prompt engineering and tool selection to fit with the query intent. To validate our approach, we conducted experiments on various question types, including Issues, Pull Requests, Commits, Compound Questions, and General Repository Information, evaluating our target prompts' ability to improve the accuracy of responses from the model. Beyond demonstrating the utility of this architecture to a diverse set of users, our findings suggest that this architecture can make repository data more accessible to technical and non-technical audiences through the production of actionable insights.

The fast progress of large language models (LLMs) opens new opportunities in the management of cyber threat intelligence, but their reliability for operational tasks remains unclear. In this work, we benchmark LLMs on the task of threat level determination. First, we construct a curated dataset derived from publicly available MISP OSINT feeds. Next, we design a tailored prompt to systematically compare eight different LLMs under zero-shot conditions. Finally, we apply supervised fine-tuning on each model and perform a comparative analysis between baseline and fine-tuned versions. Our results show that zero-shot models achieve weak performance, with limited ability to correctly assign threat levels. Fine-tuned models, however, demonstrate substantial improvements, reaching F1 scores between 0.40 and 0.58 depending on the base architecture. Despite this progress, the performance is still low for practical deployment, highlighting the need for additional research on data quality, model adaptation, and domain-specific tuning.

Vision-Language-Action (VLA) policies are commonly adapted to new manipulation settings through additional gradient updates, which limits rapid deployment when task-specific data or compute is scarce. We present ICI-VLA, a training and retrieval framework that equips a text-action VLM with few-shot test-time adaptation through in-context demonstrations. Unlike mainstream VLA designs based on action-specific multimodal fusion, ICI-VLA retains the native text-generation interface. ICI-VLA updates its parameters only during offline training; at inference, the policy remains fixed and conditions action generation on retrieved micro-demonstrations. The framework decomposes long trajectories into short, semantically labeled examples and trains an RD-Encoder with positives mined by Dynamic Time Warping (DTW), aligning the retrieved context with the phase and geometry of the current subtask. We further introduce Target Action Masking, a context-corruption objective designed to reduce direct action copying and increase reliance on the current observation. ICI-VLA reaches average success rates of 97.7% on LIBERO and 60.4% on RoboTwin 2.0, exceeding the highest reported baseline average on RoboTwin 2.0 by 19.3 percentage points. It also achieves 83.2% across four physical tasks. These results indicate that a fixed VLA policy can benefit from conditioning on spatiotemporally aligned demonstrations at test time.

Linear complementary dual (LCD) codes form an important class of linear codes with applications in cryptography, classical error correction, and quantum coding theory. In this paper, we study the neighbor relation on LCD codes over finite fields and the graph induced by this relation, where two codes are adjacent whenever they intersect in codimension one. We determine the number of neighbors of an LCD code that are also LCD, and we use this result to analyze the structure of the corresponding neighbor graph. In particular, we prove its regularity over arbitrary finite fields and establish further regularity properties for its main structural subgraphs in the binary and odd-characteristic cases. These results provide a graph-theoretic framework for the study of LCD codes and reveal a strong combinatorial regularity in their neighborhood structure.

SNNs running on neuromorphic hardware use spikes to achieve sparse and energy-efficient communication over a mesh of cores. In turn, system performance heavily depends on the assignment of neurons to cores: the mapping. Since hardware features inter-core multicast and intra-core replication of spikes, we model SNNs as hypergraphs to exploit both opportunities for reducing communication traffic. Mapping thus comprises two NP-hard problems: hypergraph partitioning and placement on the lattice of cores. High-quality solutions to both are critical, yet increasingly difficult as networks scale to millions of neurons. Therefore, we propose a GPU-accelerated pipeline for SNN mapping: a multi-level partitioning scheme is devised around hardware constraints, while placement is initialized through recursive bisection, followed by refinement pulling together strongly connected cores through repeated swaps. Model-based experiments show upwards of 16% lower latency and 42% lower energy for spike movements over existing sequential tools, while our parallel mapper is on average 18-280x faster.

This work introduces an alternative view of efficient exploration and studies its theoretical and empirical implications in the absence of extrinsic rewards. Specifically, we define efficient explorers as agents that prioritize generating generalizable experience, i.e., data that supports learning models capable of predicting and adapting across the environment. This allows us to analyze efficient exploration through the lens of prediction and generalization. Theoretically, we demonstrate that optimally efficient explorers naturally schedule their trajectories to visit the most informative and learnable regions first. Empirically, we show that optimizing for these agents gives rise to an automatic curriculum of progressively more complex behaviors, even in relatively simple environments. These results indicate that pursuing this purely intrinsic objective alone is enough to drive the emergence of highly sophisticated behaviors. We believe that this new framework provides a principled mechanism by which agent-environment systems may sustain an open-ended process of increasingly complex behavior without external rewards, tasks, or objectives.

Can LLM agents simulate how a real public would deliberate? For such simulations to be meaningful, agents must reflect the population's opinions, and their views must be shaped by what others say during discussion. We evaluate these requirements across six LLMs using national surveys, controlled deliberation experiments, and a real deliberative poll. Comparing census-grounded Korean personas against national surveys, we find that agents often produce overly concentrated opinions and fail to reliably reproduce demographic differences. During simulated deliberation, agents exchange arguments, respond to one another, and frequently change positions. Yet similar changes occur even when agents cannot see one another's contributions, while simply restating their starting positions sharply reduces movement. We further test whether these limitations persist in a real-world setting, using agents constructed from actual respondents in the 2017 Shin-Kori deliberative poll. Even with real respondent profiles, agents fail to reproduce the panel's initial opinion distribution or how its opinions change across survey waves. Our findings show that generating deliberative discussions, representing public opinion, and reproducing interaction-driven opinion change are distinct capabilities. These results suggest a different potential role for LLM deliberation: surfacing arguments and perspectives, rather than simulating a real public.

360{\deg} salient object detection (SOD) aims to accurately segment salient regions across a full field of view. However, equirectangular projection (ERP) introduces severe spatial distortion when mapping the spherical domain onto a planar representation. Existing methods mainly focus on compensating projection distortion while overlooking the interaction between panoramic geometry and salient object structure during feature perception and prediction refinement. To this end, we propose SphereSOD, an ERP-native framework that couples panoramic geometry with evolving salient structures. Specifically, spherical geometry governs feature sampling and spatial weighting, while coarse-grained saliency and contour prediction influence context aggregation during the progressive decoding process. SphereSOD first initializes deformable sampling based on spherical projection geometry and then employs bounded, content-adaptive offsets, yielding features that are better aligned with the underlying panoramic geometry. Subsequently, the decoder performs structure-guided context aggregation and progressive refinement to recover complete salient regions and accurate boundaries. Extensive experiments on three public 360{\deg} SOD benchmarks demonstrate state-of-the-art performance and a favorable accuracy-efficiency trade-off, supporting structurepreserving inference directly in ERP space as a promising alternative to projection-heavy panoramic pipelines.

Large language models (LLMs) are increasingly deployed for translation tasks, yet their implicit political positioning in such contexts remains understudied. We ask whether a single politically charged framing term, such as aggressor, enemy, neighbour, or coloniser is sufficient to trigger implicit political alignment in an otherwise apolitical task. We present a fully crossed factorial study in which eight models spanning Western, Chinese, and European origins are prompted to translate culturally attributed recipes into a target language left deliberately unspecified. Across 17 languages, four framing conditions, eight models, and 15,680 responses, we find that models do not simply decline or ask for clarification but resolve the ambiguity. Language resolution and reasoning behavior cluster meaningfully along model families: Western models hedge and deflect with vague justifications, Chinese models resolve conflicts silently, and Mistral Large emerges as a distinct profile combining high compliance with conflict-grounded reasoning. Sensitivity to framing terms is consistent across models: even subtle framing variation is sufficient to modulate behavior. Our findings urge caution when deploying LLMs for translation in conflict-adjacent contexts, where implicit political judgments may be made without any signal to the user.

Caching systems often rely on simple eviction policies such as Least Recently Used (LRU) and Least Frequently Used (LFU), which perform well in complementary request regimes. Recent policies such as LeCar and Cacheus combine LRU and LFU using ideas from the experts problem in online learning. Specifically, upon a miss, they randomize between the two eviction rules using probabilities derived from scores updated by tracking the history of past evictions. While these policies exhibit strong empirical performance, it remains unclear whether they are guaranteed, on every request sequence, to perform asymptotically as well as the better of LRU and LFU, i.e., whether they achieve sublinear regret with respect to this benchmark. We first show that LeCar suffers linear regret against an oblivious adversary, even with unbounded history. We then propose H-MC, a Hedge-based mixture of virtual LRU and LFU caches that preserves Hedge's selection probabilities, and hence its regret guarantees, while minimizing the switching cost among all joint selection rules with these marginals.

We study a factorization notion for Lipschitz functions between metric spaces in which such a function is written as a composition of a Hölder function and a function that suitably “undoes” the Hölder regularity. We show simple ways to construct such “de-Höldering” factorizations. If the identity mapping on a metric space $\mathcal{M}$ admits a de-Höldering factorization through a metric space $\mathcal{Z}$ that has a conical geodesic bicombing, then the class of metric spaces from which one can extend $\mathcal{Z}$-valued Lipschitz functions is shown to be contained in the corresponding class of metric spaces for $\mathcal{M}$-valued Lipschitz functions. As a quick consequence of these abstract permanence properties, we deduce that every $L_1$-valued Lipschitz function from a subset of $\ell_2$ can be extended to a Lipschitz function that takes values in $L_1$ and is defined on all of $\ell_2$, answering a 1992 question of Ball. By work of Makarychev and Makarychev, this implies that every weighted graph has a vertex cut sparsifier of size $n$ and quality $O(\sqrt{\log n})$, improving Moitra's 2009 bound. We also show that for every metric space $\mathcal{Z}$ that has a conical geodesic bicombing, any metric transform of a metric space $\mathcal{M}$ has $\mathcal{Z}$-valued Lipschitz extension modulus at most a universal constant multiple of the $\mathcal{Z}$-valued Lipschitz extension modulus of $\mathcal{M}$ itself, improving the 2002 bound of Brudnyi and Shvartsman.

We study semantic-aware remote estimation of multiple finite-state Markov sources observed by K sensors with overlapping coverage. The sensors share a time-division multiple-access uplink and differ in transmission reliability, delivery delay, and transmission budget. In each slot, the scheduler jointly selects a source and one of its monitoring sensors, or remains idle, to minimize the long-run average cost of actuation error subject to global and per-sensor transmission-frequency constraints. We formulate this problem as a finite average-cost constrained Markov decision process. We show that the transmission resource functions have rank at most K, although the global constraint may still restrict the feasible region. Consequently, the Lagrangian depends only on K effective transmission costs, and an optimal constrained solution can be represented using at most K+1 deterministic policy-recurrent-class components. We further characterize the piecewise-affine concave Lagrangian value function and derive projected dual subgradient ascent over an explicitly bounded multiplier set. Numerical results illustrate the value-function structure, the need for policy randomization in a representative instance, and the interaction between global and per-sensor transmission budgets.

Late-interaction models such as ColBERT achieve strong effectiveness by representing each document with many token-level vectors, but this expressivity leads to large indexing cost, storage footprints and expensive MaxSim scoring. We show that late-interaction representations exhibit an intrinsic low-rank structure: document token embeddings concentrate in a low-dimensional subspace that preserves most of the retrieval signal. Leveraging this observation, we introduce EigenLI, a spectral approximation framework that compresses late-interaction representations via document-specific low-dimensional subspaces. Unlike clustering or pooling methods, EigenLI identifies the dominant eigendirections of each document and uses them to construct reduced interaction representations. Empirically, $k$-EigenLI with $k \le 32$ outperforms k-means and Ward clustering based pooling methods on ColBERTv2 and AnswerAI-ColBERT-small; GTE-ModernColBERT exhibits a different tradeoff at $k=32$, where clustering methods perform better. The same spectral construction also yields EigenLI-SV, an ANN-compatible single-vector representation derived from the second-order summary of the reduced structure. Across multiple datasets and all three text models, EigenLI-SV consistently outperforms comparable single-vector surrogates such as MUVERA.

The avalanche safety literature has identified human factors that contribute to accidents, yet researchers note a persistent gap between these insights and effective product design. Meanwhile, interactive technologies are shaping backcountry decision making with or without grounding in theory and research. We broaden the discussion of human factors into “human-technology factors," examining how technology can both support and undermine judgment in avalanche terrain, from a human-computer interaction (HCI) perspective. We summarize relevant HCI concepts along four dimensions - attention, cognition, trust, and social interaction - and use them to revisit McCammon's FACETS framework, cataloging ways in which specific technologies may mitigate or exacerbate classic heuristic traps, grounded in accident reports and literature where possible. We identify recurring patterns, including technologies with dual-sided effects and a pervasive “digital expert halo."

How do you validate a cheap, deterministic proxy for an oracle that is expensive, rate-limited, and non-stationary? We present a protocol built on adversarial falsification gates (negative control, dose response, bounded amplification, duplication penalty, length neutrality) that define and select the proxy, fitted on a training split and confirmed held-out; around them it bounds what the proxy can never resolve, and re-measures external causal evidence on the current oracle rather than assuming it. We demonstrate it end to end on Generative Engine Optimization, where the proxy is a deterministic content score, and one step fails on that domain exactly as the protocol is built to detect: re-measuring the only published causal anchors (2023 effect sizes) on ten modern engine families shows their levers move citation on none, so the anchors are an expired external check; recalibrating to the near-zero modern vector strips the score of its lever-responsive components. What survives is the gate-enforced response surface. The gates buy a measured property: on a 500-source benchmark of adversarial edits, amplifying the score's calibrated levers gains an attacker at most 6 points, and decreases with dose; single-lever amplification is provably bounded, while the cap and cross-lever sub-additivity are empirical findings consistent with it. On detection, web-spam baselines dominate and out-of-distribution attacks evade the score, so the deployable filter layers it over them. A query-conditioned skyline bounds the score's citation signal (within-query Spearman 0.11), repositioning query-agnostic scores as quality filters rather than citation predictors. A query-leakage bug in our first ranking evaluation and a failed confidence flag are disclosed and corrected; every number reproduces offline from released artifacts at zero marginal API cost.

Fibre optic sensing, such as distributed acoustic sensing (DAS), has become a widespread technology for geophysical studies. To process the large-scale datasets produced by DAS, several machine learning methods have been proposed. However, without standardization of data and models, these methods lack comparability and interoperability. This introduces a gap between model developers and practitioners analyzing DAS data and inhibits adoption of deep learning for DAS. To address these limitations, here we present SeisBench DAS, an extension to the SeisBench library for machine learning in seismology. SeisBench DAS defines standard formats for DAS benchmark datasets, including standardised metadata and labels, and DAS models. It builds on the xdas framework for data ingestion and virtual array handling, and on PyTorch for reading and applying the machine learning models. Importantly, SeisBench provides an engine to efficiently apply deep learning models to diverse formats of DAS data, bridging the gap between model developers and practitioners. SeisBench DAS is designed as an open and extensible framework, allowing to easily incorporate future developments in deep learning for DAS.

3D Gaussian Splatting has recently revolutionised novel view synthesis as well as many other 3D vision methods and applications. Drawing inspiration from this representation, we now rethink textures to overcome the main issues of UV mapping while considerably lowering their memory footprint. Heat Kernel Textures (HKTex) eliminate UV unwrapping as well as their persistent issues of wasted UV space, seams, distortions, vertex-duplication, and varying resolution. Grounded in discrete Riemannian geometry and intrinsically defined on any manifold surface discretised as a triangular mesh, HKTex uses anisotropic heat kernels as geodesic equivalents to Gaussians. Like our kernels, also the optimisation of their position and the adaptive densification strategies were redefined to operate on the surface of the object to be textureised. Our novel representation is also fully integrated with a physically based renderer and can be optimised either from existing textures or multi-view images. Our project page and code are available at circle-group.github.io/research/HeatKernelTextures.

In approval-based committee voting, proportional approval voting (PAV) is a well-studied rule that combines proportional representation with Pareto efficiency. However, computing a PAV committee is NP-hard, raising a natural question: Can the proportionality and efficiency properties of PAV be achieved via computationally efficient procedures? We make two contributions toward answering this question. First, building on the known proportionality guarantees of the local-search-based variant of PAV (or local PAV), we systematically study its efficiency properties. We show that local PAV committees are weakly Pareto optimal, meaning that no other committee is strictly preferred by every voter. We also identify limitations: Local PAV guarantees only a $2$-approximation to fractional Pareto optimality ($2$-fPO) and a $2/3$-approximation to the optimal PAV score, and both bounds are tight. In contrast, global PAV is Pareto optimal and satisfies the stronger $\alpha^\star$-fPO guarantee, where $\alpha^\star \approx 1.346$ is the unique solution of $\int_0^{\alpha^\star} \frac{1-e^{-y}}{y} \, dy = 1$, and this approximation is tight. Second, we design a polynomial-time algorithm that combines the best of these guarantees. The committee returned by our algorithm satisfies EJR$+$ (a proportionality guarantee), $\alpha^\star$-fPO, and weak Pareto optimality. It also achieves a $0.79$-approximation to the optimal PAV score, matching the best possible polynomial-time approximation assuming $P \neq NP$. Our algorithm works by pipage rounding a concave relaxation of the PAV objective and using that committee to initialize local PAV, thereby combining global approximation guarantees with local search stability.

In recent years, automatic speech recognition (ASR) has witnessed transformative advancements driven by three complementary paradigms: data scaling, model scaling, and deep integration with large language models (LLMs). However, bridging the gap between academic benchmark performance and real-world production utility remains a persistent challenge, particularly in handling diverse regional dialects, dynamic entities and hotwords, long-range contextual information, and disfluent spontaneous speech. In this report, we present Qwen-Audio-3.0-ASR, a Mixture-of-Experts (MoE) LLM-based ASR system designed to address these production demands through a unified, instruction-following framework. The model is built upon the Qwen backbone, and is trained on tens of millions of hours of large-scale speech data. Qwen-Audio-3.0-ASR supports transcription across 30 languages and 16 Chinese dialectal varieties spanning eight major dialect regions. Beyond multilingual and dialectal recognition, the model provides production-oriented capabilities including industry-domain entity recognition, hierarchical hotword customization, native single-pass transcription polishing, and long-audio contextual modeling. We further develop a dedicated streaming variant, Qwen-Audio-3.0-ASR-Streaming, for latency-sensitive applications. Extensive evaluations on Chinese, English, multilingual, and real-world industrial test sets demonstrate state-of-the-art or highly competitive recognition performance across a broad range of evaluation conditions, with strong performance relative to leading commercial and proprietary systems including GPT-4o Transcribe and Gemini 3.1 Pro.

Multi-object tracking (MOT) has advanced rapidly in urban surveillance and autonomous driving, yet many trackers rely on ReID- and transformer-based appearance encoders and are designed for standard FoV cameras. These assumptions break down for low-cost omnidirectional deployments, where equirectangular projection introduces seam discontinuities and targets appear to be small and fast-moving. We address multi-object tracking of flying animals captured in remote environments using omnidirectional cameras. We propose a lightweight framework that re-engineers SORT-based tracking for this geometry, including (i) a Seam-Aware Motion Model that keeps the Kalman state continuous across the seam, (ii) a composite seam-aware association cost that pairs a wrapped Euclidean term with GIoU, and (iii) OmniSmall, a new benchmark of omnidirectional wildlife footage. On our new dataset, with ground-truth detections, our modifications improved over OCSORT by +8.51 HOTA, +9.41 MOTA, and +10.17 IDF1; with YOLOX detections the gain narrows to +1.95 HOTA. Our proposed methods improved tracking performance on OmniSmall and remained competitive on JRDB without adding appearance encoders while keeping the tracking stage CPU-only. Our dataset and source code are available at: https://github.com/Xin-Shu/OmniSORT.git.

Flying humanoids are expected to perform tasks in diverse environments, while their existing locomotion is mainly limited to aerial flight and ground walking. The capability to move in complex three-dimensional space can greatly expand their application range. For such walking motion on ceilings and similar anti-gravity environments, whole-body MPC is effective. However, the discontinuous changes in dynamic structure accompanying contact switching during walking can induce thrust spikes, resulting in control instability. Therefore, in this work, we propose and implement a real-time whole-body MPC framework for anti-gravity bipedal walking. First, we formulate whole-body MPC using the time derivative of thrust, namely thrust-rate, as the control input. This formulation guarantees continuity of the thrust trajectory during contact switching while preserving the sparse structure of the optimal control problem for fast computation. Second, we address the lack of natural support forces in anti-gravity environments. We introduce lower bounds on the foot-normal component of the contact force, and smoothly transfer them during the doublesupport phase. Finally, we implement the proposed framework and demonstrate anti-gravity walking by a flying humanoid through simulation and a hardware experiment. To the best of our knowledge, this is the first demonstration of multi-contact whole-body MPC for a transformable aerial robot and walking by a flying humanoid beyond the ground.

Open source communities are converging on a new governance artifact: the AI contribution policy. These policies barely existed a few months ago and are now being written and adopted. We analyzed 281 AI contribution policies, and manually classified them along the six dimensions; to study how policies change, we also tracked 92 dedicated AI policy files over time. We answer four research questions on (1) AI usage allowance, (2) AI disclosure practices, (3) AI slop countermeasures, and (4) AI policy evolution. We find that, first, permission is the norm rather than the exception: 83.3% of policies permit or encourage AI in code contributions. But permission comes with conditions, as 67.3% require a high level of human involvement and 43.4% assign accountability. Second, AI disclosure is required by 48.8% of policies, most often in pull request descriptions and commit messages, but what must be disclosed varies widely. Third, we identify ten countermeasures against AI slop, targeting pull requests, users, and autonomous agents. Finally, policies are not static: half of the dedicated AI policy files have already been revised since creation. Our results give maintainers and researchers a baseline and a labeled corpus for studying the impact of AI policies.

Contrastive learning approaches achieve strong performance by training models to bring similar samples closer while pushing dissimilar samples apart. A crucial component of contrastive learning is the temperature hyperparameter $\tau$, which controls the penalty strength applied to negative samples. However, most existing methods either fix this hyperparameter or learn a global value during training. In this paper, we introduce TeMo, Temperature Modulation framework, a similarity-based modulation approach that adaptively adjusts the temperature for each positive-negative pair according to their similarity, enabling more fine-grained multimodal contrastive learning. Our approach seamlessly integrates temperature-modulated multimodal and unimodal losses with the standard multimodal contrastive loss by gradually transitioning between them. This design allows the model to capture both coarse- and fine-grained semantics at different training stages. Extensive experiments demonstrate that each component of TeMo consistently enhances performance across diverse zero-shot retrieval and classification tasks, establishing new state-of-the-art results.

We present a partially fault-tolerant implementation of QAOA based on the $[[4,2,2]]$ error-detection code, targeting the Max-Cut problem on a square graph. Our main contribution is a novel ancilla-mediated logical $R_{ZZ}$ gate enabling interactions between qubits in different $[[4,2,2]]$ blocks. We evaluate unencoded and encoded circuits under five noise models, with both all-to-all and grid-routed connectivity, using the Cirq and qsimcirq frameworks with parallel CPU execution. Post-selection on stabilizer measurements consistently improves the probability of sampling optimal bitstrings, with five measurements providing the strongest benefit. These results support error-detection as a practical near-term strategy for improving the quality of variational quantum algorithms.

Fine-grained computation–communication overlap in distributed Mixture-of-Experts (MoE) inference allows communication to begin as partial compute results become ready. However, cooperative thread arrays (CTAs) performing computation and communication contend for finite residency capacity on streaming multiprocessors (SMs). Because a resident CTA generally retains its allocated SM resources until completion, CTAs that cannot be co-resident must wait for resources, resulting in wave-like execution. A fixed resource partition cannot adapt to changes in input size, routed expert load, and kernel configuration, potentially causing a communication backlog or reducing expert compute parallelism. We present a wave-quantized analytical model and launch-time resource manager for dependency-coupled overlap pipelines. Using routed-tile counts, kernel occupancy, GPU residency constraints, and split-level readiness dependencies, it selects the communication-CTA count and resource partition before each launch without candidate execution, per-workload profiling, or kernel recompilation. We integrate the method into the public COMET A100 implementation in FLUX. We evaluate three MoE models on four NVIDIA A100 GPUs under several parallelism strategies at the GEMM2+GatherRS operator, complete post-router MoE layer, and complete-model prefill levels. Across 15 real-p90 workloads, the analytical selector achieves 3.22 percent mean regret relative to the measured oracle with a mean solver overhead of 0.157 microseconds. Over COMET, our method achieves geometric-mean speedups of 2.528x at the GEMM2+GatherRS operator, 1.771x at the complete post-router MoE layer, and 1.185x for complete-model prefill, with maxima of 4.218x, 2.584x, and 1.439x, respectively. At every feasible TP=2/EP=2 sequence length of at least 4,096, our implementation outperforms COMET, Megatron core-TE, and FastMoE TP+NCCL.

Contact-rich assembly remains challenging because it requires submillimeter spatial accuracy and reliable interpretation of forces during sustained contact. Although simulation-based reinforcement learning offers a scalable training paradigm, discrepancies in visual observations, contact dynamics, and force/torque (F/T) measurements often limit policy transfer. We observe that proprioception is comparatively consistent across domains because joint positions are expressed in a shared calibrated coordinate system and joint velocities are computed consistently in simulation and on hardware. Based on this observation, we present PACE (Proprioception-Anchored Cross-Modal Encoder), which supervises temporal visual and F/T representations by predicting proprioceptive state transitions. Static domain-specific factors, including lighting, texture, and sensor bias, contain little information about joint motion; optimizing the proposed objective therefore suppresses their influence on the learned representation while retaining task-relevant motion cues. Policies trained on frozen PACE features are directly deployed on hardware without real-world fine-tuning or object-pose tracking. Across four contact-rich assembly tasks, PACE attains an average real-world success rate of 93.3% and only a 2.7-percentage-point sim-to-real drop, while remaining robust to perturbations that substantially degrade pose-based and learned-fusion baselines.

In data-driven predictive maintenance (PdM), feature extraction is usually treated as fixed preprocessing: a descriptor set is chosen once and reused while the downstream model or forecasting horizon changes. This paper isolates the representation-learning stage and presents a quantile-led feature-extraction framework based on a dual-stage MLP-QRNN hierarchy. QRNN1 learns a broad ten-quantile conditional distribution for each sensor channel, while skip-connected QRNN2 refines a retained mid-tail quantile set into compact, channel-resolved, distribution-aware features. A fixed thirteen-pipeline ablation spans 1-hour, 70-hour, and 30-day regimes across 72 machines in 9 industrial facilities, with the downstream temporal classifier held fixed within each regime. Increasing the retained mid-tail set from two to four quantiles improves 30- and 60-minute F1-score, reaching 75.92% and 72.44% with attention enabled. The results also show that representations do not transfer reliably beyond their design horizon unless feature capacity, temporal embedding, activation strategy, and sensor breadth are scaled with the forecasting task. The unmodified short-horizon extractor falls to 42.90% F1 at 70 hours, whereas horizon-conditioned extractors reach 60.38% at 70 hours and 79.97% at 30 days. The framework therefore supports treating PdM feature extraction as a horizon-dependent representational stage rather than fixed preprocessing.

Learning physically plausible dynamics from visual observations is essential for interactive world models and embodied agents. However, modeling real-world deformable objects remains challenging because their dynamics often arise from complex, spatially heterogeneous material responses. To address this challenge, we propose PhysReal, a video-driven framework for learning and simulating the underlying physics of real deformable objects. PhysReal integrates a spatially varying hybrid expert-neural constitutive model with a differentiable MPM simulator and 3DGS renderer. Analytical expert models provide interpretable physical priors, while neural constitutive residuals capture material responses beyond predefined formulations. Spatially distributed patches parameterize the constitutive field, enabling a continuous representation of local material variations. To organize the identification of this model from sparse visual observations, we adopt a progressive curriculum that sequentially optimizes global material properties, spatially varying local parameters, and neural constitutive residuals, together with complementary motion and mask supervision. Extensive experiments on diverse deformable-object interactions demonstrate that PhysReal achieves superior performance in dynamic reconstruction and future-state prediction, while showing strong potential for downstream robotic applications.

Language-model agents are vulnerable to indirect prompt injection (IPI) during tool use: adversarial instructions hidden in untrusted tool outputs can covertly redirect legitimate task execution. Existing work often trains and evaluates defenses against fixed attacks that do not adapt to the defender's behavior, so the resulting defenses may struggle against adaptive attacks in real-world settings. We argue that a strong defense against adaptive IPI must adapt during training to a continually evolving attacker. Building on this insight, we propose CoER, a verifier-grounded co-evolution and refinement framework that models interleaved tool calls and adaptive injections within a task as a general-sum Markov game: the defender advances the task through successive tool calls, while the attacker can inject multiple times within the same task and adapt subsequent attacks to the defender's responses and prior execution traces. After initializing the attacker from successful trajectories, bilateral adversarial reinforcement learning (BA-RL) retains historical policies from both roles as opponent populations and mixes current and historical opponents, extending training beyond the latest matchup. Attackers from these populations are then reused to challenge teacher agents, and only demonstrations verified for both safety and task completion are used to fine-tune the co-evolved defender. Across seven domains and three evaluation seeds, CoER reduces adaptive attack success from 41.3% to 0.2% and raises safe task completion from 39.6% to 76.2%; external benchmarks also show improved attack resistance. Further experiments validate the effectiveness of bilateral historical-opponent mixing and population-guided refinement. Attacker analyses show that co-evolution strengthens attack capabilities and that the trained attacker uses execution feedback to adapt subsequent injections.