Papers

Vision-Language Models (VLMs) have made substantial progress in spatial reasoning, and self-evolving agents further extend these capabilities to embodied environments by learning from interactions. However, current self-evolution remains particularly challenging in embodied spatial tasks, complex and multi-step decisions from partial visual evidence making it difficult to determine why an execution fails. The failure may stem from misperception, insufficient exploration, or incorrect planning, yet execution feedback alone rarely reveals which component is responsible. As a result, simply accumulating interaction experience can reinforce ineffective strategies. To address this, we propose HiMeco, a brain-inspired self-evolving agent for embodied spatial intelligence, inspired by metacognitive monitoring and hippocampal memory consolidation. Through Hippocampal-Metacognitive Reflection, HiMeco diagnoses failures in perception-exploration-planning (PEP), then compares execution attempts to derive reusable strategies and guidance for avoiding recurring errors. It also abstracts interaction programs into parameterized skills, using evidence contracts and execution-based validation to determine their retention and reuse. These strategies and skills guide subsequent interactions, with memory updated only between rounds and model parameters kept frozen. HiMeco achieves state-of-the-art performance on ESI-Bench, and it also improves accuracy on EmbSpatial-Bench by 3.94 percentage points without evolution.

Advances in generative AI have made it possible to produce music that is nearly indistinguishable from human compositions, raising the question of whether such works can be reliably distinguished from human compositions. We present MUSEBENCH, a large-scale benchmark for AI-generated music detection and artifact analysis, comprising over 361K tracks. The dataset includes human compositions, stylistically matched music generated by ten state-of-the-art open-source and commercial models, such as Suno and Udio, as well as human-AI co-creations and variants processed with adversarial perturbations, compression, and acoustic post-processing. Using MUSEBENCH, we benchmark 11 automated detectors and evaluate their accuracy, generalization, and robustness, under default, out-of-domain, and transformed-variant experiment settings. We further conduct a comprehensive artifact analysis through acoustic feature inspection, identifying recurring signatures such as non-idiomatic harmonic progressions, upsampling induced spectral periodicities, and violations of physical performance constraints. Our findings reveal key strengths and vulnerabilities of existing detectors and provide practical insights for building more reliable AI music detection systems.

Synthetic-aperture imaging (SAI) removes occluders such as foliage by integrating many views of a scene along a camera path. It has so far been used mainly with frame cameras. We ask whether an event camera has advantages for SAI, and answer the question in simulation, where both sensors see exactly the same scene and the same camera movement. A densely rendered Blender sequence is turned into frames with an EMVA~1288 noise model of an industrial global-shutter camera, and into an event stream with the v2e simulator parameterized like a Sony IMX636. The frames are integrated with the existing airborne optical sectioning renderer. For the events we add a new integration path that projects every single event onto the focal plane at its own interpolated pose, so no events have to be binned into frames. We compare the two sensors in three main runs and an additional dark high-speed test. In a bright scene at 25,m/s the frame camera records only three images on the path and cannot remove the occluder, while the event camera reconstructs the target, even after an event-rate controller drops most of its events to fit the camera link. At the same speed in a 5,lux scene the event camera no longer resolves the target either. In a 5,lux scene at 5,m/s both sensors reconstruct the target at similar quality. We also compare the data one traversal of the aperture costs. At high speed the capped event stream still shows the target at the same data volume as the three frames. At 1,m/s we compare both cameras at two matched data budgets, and there the event camera has no advantage over the frame camera in reconstruction quality. The data rate of the event camera depends on the scene and is highest exactly in the fast bright scenes where it is needed most.

Homogeneous turbulent flows exhibit recurring multiscale structures across spatially separated regions, suggesting that a single flow realization may contain sufficient information to learn representations of unresolved scales. We investigate this hypothesis through an extreme super-resolution framework in which a wide-activation deep super-resolution (WDSR) network is trained exclusively on overlapping patches extracted from a single high-resolution turbulent realization. Previously unseen flow fields are reconstructed through patchwise prediction followed by Hann-window-weighted overlap-add assembly. The framework is evaluated for two-dimensional homogeneous isotropic turbulence at super-resolution factors of x32 and x64, representing severe spatial under-resolution. Despite the limited training data, the proposed approach reconstructs coherent vortical structures and recovers fine-scale features more accurately than conventional interpolation. Spectral and statistical analyses further demonstrate improved recovery of high-wavenumber energy content and multiscale flow characteristics. To assess the effectiveness of exploiting internal spatial information, we compare the proposed framework against WDSR models trained on 1, 25, 50, and 100 complete high-resolution snapshots without patch extraction. This comparison distinguishes the benefits of localized learning from those associated with increasing the number of training realizations. The results demonstrate that exploiting spatial redundancy within a single turbulent realization enables accurate reconstruction of previously unseen flow fields, even under extreme spatial under-resolution. These findings establish single-realization patch-based learning as a promising data-efficient approach for turbulent super-resolution when high-resolution training data are scarce.

Orthogonal transforms have emerged as an effective tool for low-bit quantization of large language models (LLMs), as they reshape activation distributions without changing the underlying model function. Existing approaches typically rely either on fixed data-agnostic transforms, such as Hadamard rotations, or on calibration-time optimization of learned rotations. We study rotation selection from first principles and show that maximizing covariance leads to spectral alignment and provides a principled criterion for constructing quantization-friendly orthogonal transforms. Based on this insight, we propose SpectrumQuant, a closed-form rotation method that estimates activation covariance from calibration data and uses it to construct a covariance-aware orthogonal transform for post-training quantization. Our formulation characterizes quantization-friendly rotations as compositions of alignment and mixing components, while recovering Hadamard matrices as a special case. Interestingly, a class of butterfly matrices belongs to this category. In the realm of LLM quantization, SpectrumQuant bridges the gap between data-agnostic structured rotations and learned rotations, providing an effective and efficient data-aware alternative for post-training quantization. Experiments on Llama and Mistral models show that SpectrumQuant is competitive with state-of-the-art rotation strategies while requiring less than one-third of the GPU memory and significantly less time than learned rotations approaches like SpinQuant and DartQuant.

In this note, we provide a deterministic algorithm that computes the $(\min, +)$-Matrix Product $C = A * B$ of two $N \times N$ matrices $A$ and $B$ with polynomially bounded integer entries in $\tilde{O}(N^{2.9964})$ time, improving upon the exponent of $O(N^{2.9995})$ time obtained in the recent breakthrough by Anthropic (Alman and Vassilevska Williams, 2026). Our algorithm uses a phased approach first proposed by Bringmann, Grandoni, Saha, and Vassilevska Williams in 2016. The phased approach produces an approximation $\tilde{C}$ of $C$ in truly subcubic time such that the number of remaining triples $(i,k,j): A_{i,k}+B_{k,j} < \tilde{C}(i,j)$ is truly subcubic. This provided strong evidence that a truly subcubic $(\min, +)$-Matrix Product and therefore, a truly subcubic algorithm for All Pairs Shortest Paths (APSP) may exist. However, to find the remaining triples in subcubic time, a sparse certificate is needed whose construction was not known. We use the thin matrix product theorem of the recent breakthrough to provide such a certificate. This gives a simple modular subcubic algorithm for APSP.

An agent can improve its behavior within an unresolved problem by changing how it acts and which experience it retains. We propose MMLA-RSI, a recursive self-improvement scheme with two implementations in Memory-Mediated Learning Architecture: Reasoning-Time Training (RTT) changes the policy state $\Phi$, while Reasoning-Time Memory Adaptation (RTMA) changes authoritative memory $M$. Each route also makes its adaptation method an object of improvement. A current numerical updater trains the policy proposing its successor; a current memory manager chooses the records available to the process proposing its successor. An independently admitted successor takes over the same role in the next generation. We give bounded executable rule languages, shared task/proposal interfaces, and one algorithm for either route or their composition. The RTT instance changes scalar schedules in a clipped-momentum update. The RTMA instance chooses NULL or a complete-row write with possible eviction. A fixed interpreter, trusted committer, and independent evaluator bound what can change. State versions, read closures, and conditional properties describe causal execution, accepted transitions, and finite work. Existing numerical-component studies motivate the construction. In four-candidate program selection, retaining LARC feedback updates reduces expected error by 24.65 and 36.65 percentage points for two trained initializations. In chronological continuous-integration replay, feedback binding is positive while retention increases loss. Completed SQL studies improve substantially over an unadapted substrate but do not establish the required additional online benefit. These results characterize their declared components. The two recursive routes and their joint realization are specified here; their beneficial recursive chains and discovery efficiency remain to be evaluated.

Large language models increasingly generate multiple retrieval actions jointly, while downstream systems often provide only one reward for the resulting set. Broadcasting this reward provides a valid score-function gradient estimator for the expected set-level objective, but does not determine a coalition-dependent within-set attribution. We introduce CASI (Credit Assignment via Set Intervention for Agentic Retrieval), which constructs the missing counterfactual information through snapshot-grounded set intervention: generate and retrieve once, hold replayable snapshot components fixed, and evaluate alternative subsets under common conditions. Shapley aggregation converts these marginals into within-set credit, which bounded zero-sum advantage shaping incorporates into policy optimization. For m >= 4, singleton, leave-one-out, and full-set endpoints do not generally determine the Shapley allocation; under additive set values, extra coalition evaluations are redundant. In a five-action sponsored-search setting, exhaustive auction replay over 1,200 requests reveals substantial coalition dependence. Under the training-time evaluator, CASI achieves the highest median cosine similarity to a replay-defined Shapley reference among evaluated credit rules; under a sensitivity backend, CASI does not outperform Singleton. Across five independent training runs, CASI-trained policies achieve higher held-out auction-replay set value than Singleton- or Broadcast-trained policies. In a randomized seven-day production A/B test on 10% of traffic, the CASI-trained policy yields an observed 3.51% lift in the primary advertiser-value metric over production control.

Reliable medical reasoning requires LLMs to dynamically access external biomedical evidence to improve factual trustworthiness and reasoning accuracy. However, indiscriminate retrieval can introduce noise and latency, raising a critical retrieval-timing problem: at each reasoning step, the system must decide whether the model's internal knowledge is reliable or external evidence is needed. We propose PUlSE, a lightweight probing method that uses the model's internal representations as intrinsic signals for step-level retrieval control. Our posterior trajectory analysis reveals that erroneous steps exhibit substantially higher inter-layer volatility than correct ones, suggesting unstable internal knowledge. Based on this finding, PUlSE trains a compact probe on inter-layer geometric features to detect factual errors upon step completion, and uses it as an adaptive trigger for retrieval and step regeneration. Across four medical benchmarks, PUlSE-timed retrieval matches exhaustive per-step retrieval in accuracy while reducing retrieval calls by 72%, and can transfer to an unseen legal domain, indicating potential domain-agnostic generality. Overall, our study provides an initial step toward shifting retrieval-trigger design from surface-level confidence estimates to internal activation dynamics for adaptive access.

This work proposes a multi-target tracking solution that reasons over combinations of sensor-level tracks to perform distributed multiple hypothesis tracking (MHT) in an asynchronous multi-sensor network. This approach handles clustered target tracking and allows for efficient information pedigree summarization for distributed data fusion (DDF). By decomposing the association problem into partitioned hypothesis sets and fusing across sensor track combinations, the method reduces combinatorial complexity compared to traditional MHT approaches. Evaluation on a satellite constellation-based tracking scenario demonstrates improved estimation accuracy and data association quality compared to a reference target tracker with global nearest neighbor association.

Anisotropic magnetorheological elastomers are magnetoactive smart materials whose complex magneto-mechanical behavior is strongly influenced by their underlying microstructure. Efficient and accurate multiscale simulations of these materials rely on small unit cells that reproduce the behavior of the real material. In this work, we present a framework for the generation and validation of such two-dimensional representative volume elements (RVEs) from a given microstructure image, assuming circular particles. The cells are generated through an optimization procedure where the two-point correlation function serves as a differentiable descriptor, while a minimum inter-particle distance is enforced as a constraint. To assess the equivalence of the generated cells, we compare their effective magneto-mechanical response with those of the reference microstructure. We demonstrate that the generated cells reproduce the reference response with high accuracy, also with substantially fewer particles. Consequently, the computational cost of subsequent multiscale simulations is reduced.

Conditional generative models are increasingly used for scientific prediction beyond the direct training support. Yet scientific questions often concern whether one intervention changes the effect of another. Such interactions can be of primary interest while explaining little of the total variation, allowing models to appear accurate overall without recovering the relevant scientific effect. It remains unclear how these models learn interactions and when these successfully extrapolate to unseen conditions. We therefore study interaction recovery in flow-matching models using a physiological time-series simulator, drug dose-response data, combinatorial cell perturbations, and molecular conformer generations, where we separate recovery from hallucinated interaction structure. In the controlled setting, individual effects are learned much earlier than interactions and models hallucinate interactions where none exist. Across settings, we find qualitatively different generalization problems, with successful prediction depending on whether condition structure continues from observed to held-out targets. When it does, interactions are recoverable. If not, models can generate plausible responses while missing the interaction. We furthermore find that hallucinations increase as targets become less identifiable, and that successful interaction recovery in output space can fail under nonlinear readouts. Finally, we show that aggregate fidelity metrics can conceal whether models learn and extrapolate interactions.

The Platonic Representation Hypothesis (PRH) holds that increasingly capable models converge toward a common representation. However, recent studies have shown different views on aligning representations via their neighborhoods, spanning trees, and distances. We show that these views can differ because the alignment criteria behind them sit at different points of one lattice, and we formalize this explanation as the Lattice Representation Hypothesis (LRH). Under the LRH, the convergence claimed by the PRH is agreement at the top of the lattice, whereas its evidence comes from neighbor overlap, a criterion near the bottom. The Aristotelian Representation Hypothesis (ARH) restricts convergence to local neighborhoods because calibration keeps the growth of neighbor overlap with model size but largely removes that of centered kernel alignment. We show that this removal can come from calibrating a maximum over layers, so global convergence remains open. Back into Plato's Cave (BPC) infers coarse shared structure from the collapse of neighbor overlap on larger evaluation sets, whereas we trace it at least partly to a fixed number of neighbors covering shorter distances. We show that the spanning-tree overlap used by Structure over Geometry (SOG) mostly repeats one-nearest-neighbor overlap. The LRH also extends to independently trained models, whose training procedure and objective decide the criteria met reliably across runs. We prove this claim for the training procedure in a linear autoencoder and for the objective in a finite contrastive Gibbs model. Experiments on dozens of pretrained vision and language models and thousands of runs of both theoretical models support these conclusions. We hope this work paves the way toward a general theory of representation hypotheses, with a framework to analyze them and compare their evidence. Project page: https://eethanshi.github.io/LRH

A central goal in solving a polynomial eigenvalue problem is to compute eigenpairs that are exact for a nearby polynomial, with each coefficient perturbation bounded by a modest multiple of unit roundoff times that coefficient's norm. A backward stable eigensolver for a linearization does not automatically provide this coefficientwise relative guarantee. Scaling can improve accuracy, but a single scaling may not work across a widely spread spectrum. We construct a family of scaled companion linearizations that covers the nonzero complex plane. Each linearization is paired with a region and an eigenvector recovery rule for which the backward-error amplification factor is bounded at exact pencil eigenvectors, independently of coefficient norm disparities. Tropical roots guide the construction, and their multiplicities reduce the number of required solves. These results provide a basis for coefficientwise relative backward stability across the spectrum.

High-quality data is essential for improving instruction tuning in threat intelligence analysis. However, existing methods for generating cybersecurity instruction data lack reliable quality assessment mechanisms, and gradient influence-based data selection methods primarily focus on single-sample matching, failing to cover diverse and complex security capabilities. To address these issues, we propose CTI-TDS, a Targeted Data Selection method for instruction tuning in Cyber Threat Intelligence analysis. Specifically, we first filter high-quality CTI instructions by integrating instruction style consistency and security knowledge verification. Furthermore, we formulate gradient influence as a distribution alignment problem, learning a gradient subspace of target tasks and applying entropic regularized optimal transport to optimize distribution matching between training samples and target tasks, enabling diversity-aware data selection. We conduct extensive experiments on CTIBench across multiple large language models. Experimental results demonstrate that when fine-tuning with only a small set of high-value instruction data, CTI-TDS outperforms existing data selection methods and achieves competitive performance on threat intelligence analysis tasks.

AI-powered image editing increasingly allows users to express high-level editing intentions, yet how these intentions are translated into concrete editing actions often remains difficult to understand. We present Atelier, a semantic image editing tool that uses image semantics and user intent to determine which editing controls to surface and how to initialize them. Guided by a formative study with eight image-editing practitioners, Atelier composes context-specific, directly manipulable controls that users can refine rather than committing AI interpretations directly to finished edits. In a within-subjects study with 12 practitioners, participants preferred AI-composed controls, described them as supporting creative exploration, and actively refined AI proposals. However, experiences of control were mixed, and interviews highlighted limited insight into why particular controls and values were proposed. Our findings suggest that making AI interpretations actionable through manipulable controls can support exploration and refinement, but actionability alone does not make those interpretations transparent to all users.

Nanosatellites (e.g., CubeSats) have the potential to carry out useful science and exploration activities for outer solar system exploration, but their small size leads to tradeoffs between attitude control and science capabilities. Magnetorquers offer a unique compromise as compact actuators that enable three-axis attitude control, but suffer from a loss of control authority and require full knowledge of the planetary magnetic field. In this work, we present a learning-based adaptive model predictive control (MPC) algorithm that plans rotational trajectories for the satellite while accounting for this loss of control authority. Key to our approach is the inclusion of a learned magnetic field model that is trained in simulation using low-fidelity magnetic field data. Using measurements collected online, our proposed algorithm adapts this learned magnetic field model using a simple last-layer update scheme, allowing the MPC controller to quickly adapt to the true planetary magnetic field model and more effectively plan attitude trajectories. We demonstrate the results of our proposed approach for two use cases, (1) Earth-orbit and (2) Uranus-orbit, and show that our learning-based adaptive MPC algorithm improves controller success for these system simulations by 5% and 61%, respectively.

Part-of-speech tagging is a fundamental sequence labeling task and an important component for many downstream natural language processing applications. However, Saraiki, written from right to left in a Perso-Arabic script, remains an under-resourced Indo-Aryan language, despite being spoken as a mother tongue by 28,849,579 people (approximately 28.9 million) according to Pakistan 7th Population and Housing Census 2023. This paper presents a manually annotated Saraiki Universal POS tagging dataset developed from plain-text. We annotated the text with 17 POS categories. We evaluate three pretrained transformer encoders under the same fine-tuning setup: XLM-RoBERTa base, Urdu RoBERTa small, and multilingual DistilBERT. Experimental results over five random seeds show that XLM-RoBERTa base achieves the best performance, with 87.29% +/- 0.21% development accuracy and 88.60% +/- 0.21% test accuracy. Urdu RoBERTa small and multilingual DistilBERT also provide competitive results, suggesting that multilingual and related-script pretrained encoders are effective for Saraiki UPOS tagging. The dataset, evaluation setup, and baseline results introduced in this work provide a benchmark for future research on Saraiki NLP and low-resource sequence labeling.

Facial image watermarking can support traceability and certification in biometric pipelines, but provenance or manipulation detection alone does not guarantee reliable biometric identity. We introduce a watermark-preserving face evasion attack that targets this gap under a black-box watermarking setting, without access to the watermarking components. The proposed PGD-L_2 attack guides adversarial perturbations with spatial and spectral preservation terms to reduce face-recognition similarity while limiting pixel-level distortion and high-frequency changes that may interfere with hidden watermark cues. Evaluated on controlled facial images with two watermarking schemes, two face-recognition models, and two datasets, the attack achieves evasion rates above 92% at operational thresholds while preserving watermark recovery above 93% for StegaFormer and above 99.7% for FaceSigns, with visual quality above 47dB PSNR at epsilon_2=2.0. These results expose a structural gap in watermark-based facial certification: a face image can remain watermark-valid and visually faithful while becoming biometrically unreliable.

Designing RNA sequences that satisfy target tertiary-structure objectives is an important problem in RNA engineering, with broad applications in synthetic biology and therapeutics. Existing 3D RNA design methods have made substantial progress, but they largely treat the target fold as a geometry-only objective. This is incomplete for RNA, whose functional folds are specified not only by tertiary geometry but also by secondary base-pairing topology. We therefore formulate RNA inverse design as a topology-and-geometry co-design problem and propose TANGO, a multi-objective multi-agent reinforcement learning framework for this task. TANGO adopts a staged curriculum: it first learns a topology-aware policy from secondary-structure targets, then transfers this policy to tertiary-structure targets to train a general co-design model, and finally applies frontier-guided fine-tuning to refine target-specific trade-offs among secondary-structure fidelity, tertiary-structure fidelity, and sequence diversity. Experiments show that TANGO generates diverse and novel designs, improving tertiary- and secondary-structure performance by 18.3% and 52.7%, respectively, over the strongest baseline.

Learnable downsampling and spectral-domain filter parameterizations have each made convolutional networks more flexible, but their interaction has not been systematically examined. We bridge this gap by examining convolutional kernels learned in the spatial and spectral domains, together with learned downsampling ratios under a cropped resizing implementation. We find that spectral and spatial kernels respond differently to learned resolution allocation, with larger gains for the spectral kernel. We observe this consistently on three datasets (EuroSAT, CIFAR-10 and STL-10) and at two model sizes (differences in accuracy gain of +0.013 to +0.028 on held-out test sets). Part of this difference reflects the greater gain of the spectral kernel from retaining more spatial area, but the difference remains positive when total retained area is matched. This pattern is observed across datasets and model sizes, although its magnitude and dependence on the complexity budget vary with model size and the resizing implementation.

Large language models (LLMs) have demonstrated multilingual capabilities across diverse language generation and understanding tasks. Yet effective code-switching requires more than multilingual proficiency: models must decide when, where, and why to switch languages in ways that fit the surrounding context. This context-sensitive ability remains insufficiently evaluated. We introduce CSWBench, a benchmark for evaluating context-sensitive code-switching in creative writing through two complementary settings: Focused Narrative and Extended Narrative. We further propose CSWEval, a theory-informed framework with two complementary levels: event-level evaluation of Licensing, Contextual Appropriateness, and Functional Gain, and whole-text evaluation of task-specific quality—overall literary and narrative quality in Focused Narrative and quality emphasizing multilingual organization in Extended Narrative. Evaluating twelve LLMs, we find that models exhibit meaningful but uneven code-switching competence across narrative contexts. Model differences are particularly pronounced in multilingual literary integration, while multilingual-focused dimensions show stronger difficulty effects overall than the general narrative dimensions; narratively salient switches are associated with stronger contextual and functional realization; and event-level CS quality is moderately associated with whole-text performance.

Single-channel radio-frequency (RF) signal separation is important for interference mitigation when a signal of interest overlaps with a co-channel interferer in time and frequency. Classical matched filtering and LMMSE receivers are limited under structured non-Gaussian interference, while diffusion and score-based methods often require iterative sampling or optimization. We propose FlowSep-RF, a rectified flow matching framework that learns a clean-endpoint transport from the received mixture to the signal of interest. The model combines time-domain IQ samples with frequency-domain features and recovers the separated waveform using one endpoint velocity evaluation. We evaluate FlowSep-RF on QPSK mixtures with OFDM-BPSK, OFDM-QPSK, and RFChallenge CommSignal2 interference against classical, discriminative, and generative baselines. FlowSep-RF achieves the lowest average BER and MSE among evaluated baselines across all three settings with one-step inference.

Deep Gaussian processes (DGPs) use iterative sampling to create non-stationary random fields. This is notoriously expensive because the covariance at each non-initial layer depends on the realised field from the layer below, so it changes with every sample. Moreover, standard fast sampling techniques like circulant embedding or matrix compression fail since the underlying kernel usually lacks the necessary regularity and structure. We develop and analyse a wavelet compression method for convolution Matérn DGPs. At each layer the covariance matrix is compressed using wavelets and a Lanczos algorithm computes the random sample. For specific parameter ranges involving the smoothness of the Matérn kernel as well as the smoothness of the wavelets, we show that the accumulative compression and sampling errors through all layers decay exponentially with increasing truncation parameter. We obtain a practical DGP sampling algorithm by choosing a truncation that gives log-linear covariance storage and matrix-vector costs, enabling DGP sampling at nearly linear computational cost.

Client-level fair federated learning, in the form of $q$-fair federated learning ($q$-FFL), agnostic federated learning, or tilted empirical risk minimization, has emerged as an attractive solution to equalize utility across heterogeneous clients. This paradigm grants high-loss clients greater relative influence, while a central server aggregates their loss-weighted updates into a single shared model. Yet, the Byzantine robustness of $q$-fair federated learning has been largely neglected, leaving the server unable to tell legitimate fairness-induced amplification from malicious update scaling. To address this critical gap, we formalize the problem of Byzantine-robust $q$-fair federated learning under an update-only interface, in which the server observes nothing but the fairness-scaled client updates. Our analysis under shared class-conditional label skew shows that fairness scaling preserves the nonnegative class-mixture structure of honest updates, while replacing their unit-sum constraint by a client-dependent radial scale. Moving beyond magnitude-based screening, we show that this warped geometry supplies two complementary tests, one radial and one structural, that a single norm cannot provide. We then introduce FUSE, a structure-aware robust aggregator that uses a small trusted labeled dataset for coefficient recovery, scale and cone-consistency screening, and post-screening step-size control. Across four benchmarks and seven Byzantine attacks, FUSE improves seed-wise worst-attack accuracy over the strongest baseline on each benchmark by 3.9–30.1 percentage points, while largely preserving the influence of the high-loss honest clients that fairness is meant to protect.

Every multilingual speech pipeline begins with a language decision, an error there propagates silently through recognition, translation and routing. In India it spans 22 scheduled languages across four language families, heard over short, noisy, narrowband and code-mixed audio. Foundational speech models increasingly present language identification (LID) as an inherent latent, generalized capability. Autoregressive decoders in speech models emit a special language ID token, and modern LLMs with audio capabilities can name the language when prompted. We test this assumption on identical clips and find that whisper and similar speech models has tokens for only 15 of the 23 languages, on one second inputs Gemini 3.5 Flash reaches 0.52 macro-F1 and GPT-audio 0.14. We present Anuvani LID, a dedicated 115M parameter LID conformer model trained on 54,643 hours of Indian speech spanning 22 Indian languages and Indian accented English. Along with designed and controlled studies on 102 condition benchmark with speaker and corpus held out tests. Across 37 conditions, it reaches 0.914 mean macro-F1, 5.2 points above Gemini 3.5 Flash and ahead of 14 other systems. It is 24 points ahead of Gemini at one second, scores 0.897 on all 23 languages where no other system exceeds 0.713, and is calibrated to an ECE of 0.008. Reliable Indian LID still requires a robust, dedicated and calibrated model.

Deploying a general-purpose model as an agent means running it inside a product-specific runtime harness that decides which actions are executable, how state advances, and when the loop may terminate. Ordinary trajectory supervision shows which action was taken but leaves the runtime semantics behind that action implicit. We introduce Harness Policy Distillation (HPD), which makes these runtime control semantics an explicit training target alongside actions to align the harness policy. HPD replays reference trajectories through the harness to recover the runtime semantics that are implicit in ordinary trajectory data, and represents them as control documents at each model-call boundary. These documents provide privileged information for a teacher to generate actions, while harness verification ensures that the resulting targets are executable under the runtime contract. We jointly train a shared model on the runtime semantics recovered from control-state privileged information and native history, thereby internalizing these semantics into the model policy. At deployment, the trained model uses the original harness interface without requiring the control document or any additional supplement. On TRAJECT-Bench with the Pi harness, Qwen3-8B trained with HPD reaches 22.98% autonomous task success and 49.23% runtime compliance, against 3.09% and 19.90% for the unadapted model. Harness-verified teacher actions substantially improve the native deployment policy, while joint control-state supervision provides further gains in task completion and runtime compliance. HPD turns executable runtime semantics into dense supervision for learning harness-compatible actions at each decision boundary.

Long-horizon agents must coordinate reasoning, tool use, and environment feedback over extended interactions. Learning from sparse outcome rewards requires weighting trajectories of different lengths and updating the policy while rollouts are still being collected. We examine a single-rollout asynchronous recipe combining Muon, token-averaged loss, truncated importance sampling, and auxiliary length and tool-error penalties. Single-rollout sampling allows asynchronous learning without waiting for multiple trajectories for the same issue. We study optimizer choices and recipe components for repository-level issue resolution across two model families. The evaluated training runs show higher training success with the full recipe than with the baselines.

Despite the Markov assumption, safety failures in robot reinforcement learning rarely occur instantaneously; they typically emerge over several steps as the policy enters unfamiliar states under disturbances and domain shift. We address this limitation by learning a receding-horizon safety critic that predicts whether a state will remain non-terminating over the next receding horizon, enabling anticipatory detection of impending locomotion failures. The critic is trained from rollout-derived horizon labels with a composite objective that combines horizon classification, control-barrier-function-inspired temporal consistency, and smoothness regularization, and we aggregate an ensemble of critics with a conservative Conditional Value at Risk estimate to account for uncertainty. Building on the learned critic, we derive a closed-form, differentiable CBF-style safety filter that minimally modifies the policy action to improve predicted safety, and integrate it into PPO with an anticipation loss to reduce policy-filter mismatch during training. In simulated quadruped locomotion, our method improves robustness under action noise, external perturbations, and terrain shifts, reducing catastrophic failures while maintaining competitive returns.

Large view synthesis models (LVSMs) reconstruct scenes from sparse views and render novel views quickly, but they incur unnecessary storage and rendering costs due to token redundancy. We present GLoD-LVSM, a compact, geometry-grounded scene representation for LVSMs with continuous levels of detail. First, instead of learning a fixed set of scene anchor initializations, GLoD-LVSM learns a continuous field that initializes a flexible number of anchors to match the desired budget. A condenser then aggregates scene information into these anchors, with both the field and condenser trained end-to-end. Second, the condenser infers reference-token and anchor geometry to guide compression through geometry-grounded attention, improving rendering quality and enabling frustum culling during decoding. Compared with existing LVSM scene-compression baselines, GLoD-LVSM achieves the best trade-off between rendering quality and scene token count. Compared with the uncompressed base model, GLoD-LVSM reduces the scene-token count by 33.4% and accelerates rendering by $1.23\times$ with no quality loss. Under a relaxed quality-loss budget, the reduction increases to 85.5%, with $3.00\times$ rendering speedup. Further analysis reveals that view-synthesis supervision yields emergent scene depth and self-organized anchors that align with surfaces and adapt their density to local texture complexity. Beyond rendering, we further show that our scene representation also supports scene-semantic understanding, highlighting its potential for broader downstream tasks.