arXiv:2606.28145v1 Announce Type: new Abstract: Wearable devices produce large, high dimensional training logs for everyday runners, and interpretation rather than data collection is now the limiting step. This paper evaluates five dimensionality reduction models, three autoencoder variants, PCA, and a Variational Autoencoder, on their ability to compress nine sensor runner profiles into a single scalar performance indicator, the latent score. Because the setting is fully unsupervised, model quality is assessed along two complementary axes: reconstruction error (Mean Squared Error) and latent score interpretability, measured via Spearman and Kendall rank correlations, Mutual Information, and Permutation Importance. These are combined into a composite selection criterion that prevents selecting models on reconstruction accuracy alone. Feature rankings from the four metrics are aggregated via a modified Borda count, and their stability is confirmed by bootstrap validation. A two feature linear baseline is included to anchor the comparison. Deep autoencoder achieved the lowest reconstruction error and the highest composite score. Once the PCA hidden layers were widened, the deeper variants became closely competitive with Deep AE on the composite criterion, indicating that the limiting factor was hidden layer capacity rather than the one dimensional bottleneck. Running pace, aerobic decoupling, and average heart rate emerged as the dominant latent score drivers across all models and resampling runs, consistent with established physiology.
Science Journals
arXiv:2606.28260v1 Announce Type: new Abstract: Microbes in marine environments are often confined to thin near-surface layers while being advected by turbulent flows. Because such constrained advection generates an effectively compressible flow, reproduction and transport interact in a nontrivial way. Here, we focus on populations whose growth is governed by an Allee effect and show that sinks and sources, generated by the compressible flow, have dramatic consequences for the survival of such species. We derive analytical expressions for the carrying capacity as a function of the Allee strength in the limit of small and large Damk\"ohler number, which measures the product of the large eddy turnover time and the organism growth rate. Numerical simulations reveal how these two limits connect. In the limit of small Damk\"ohler number, we find a maximal Allee strength, set by the statistics of the compressible flow, that leads to species extinction in fully developed turbulence.
arXiv:2606.28182v1 Announce Type: new Abstract: Embodied agents operating in decentralized and partially observable environments have attracted growing attention in recent years. However, existing large language model (LLM)-based agents often exhibit behaviors that are misaligned with their partners or inconsistent with the environment state, leading to inefficient cooperation and poor task success. To address this challenge, we propose a novel framework, Learning Laws of Cooperation (LLawCo), that enables embodied agents to autonomously align with both their partners and task objectives. Our framework allows agents to reflect on past failures to extract misaligned behavioral patterns, which are used to derive high-level behavioral laws, such as "Talk when necessary" and "Wait for partner." These laws are explicitly incorporated into the agents' chains of thought via supervised fine-tuning, aligning their reasoning with task requirements and the behavior of other agents. To evaluate our approach, we introduce PARTNR-Dialog, a large-scale multi-agent communicative and cooperative planning benchmark built on the PARTNR environment. Experiments on existing tasks and our new benchmark demonstrate significant improvements in cooperative efficiency and task success rates. Across four backbone LLMs, our method achieves average success rate improvements of 4.5% on the PARTNR-Dialog benchmark and 6.8% on the TDW-MAT benchmark over state-of-the-art open-source communicative agent frameworks. See the LLawCo project page for details: https://www.merl.com/research/highlights/LLawCo
arXiv:2606.28241v1 Announce Type: new Abstract: Background: Conversational AI chatbots designed for mental health may offer an accessible, scalable avenue for supporting psychological well-being, yet prior evaluations have largely focused on clinical symptom reduction rather than broader indicators of day-to-day functioning, and have rarely monitored for potential harms such as inflated self-perception. Objective: We examined within-person change in psychological functioning indicators among real-world users of Ash, a purpose-built conversational AI for mental health support, over the first four weeks of use, and whether these changes were associated with engagement metrics. Methods: In this single-arm observational cohort study, new users (n = 1,284) completed in-app single-item measures of psychological functioning (life satisfaction, relationship satisfaction, sleep quality, behavioral activation), working alliance, and grandiosity (inflated self-perception), at baseline and Week 4. Paired-sample t-tests examined within-person change; ANCOVAs tested engagement-outcome associations at Week 4, controlling for baseline. Results: At baseline, participants reported below-average life satisfaction and fair sleep quality. Significant within-person improvements emerged across all functioning indicators and working alliance (ps < .001; d = 0.14-0.26), with no change in grandiosity. Active days, total sessions, and total minutes consistently predicted Week 4 psychological functioning and working alliance (ps <= .006; partial R^2 range: 0.58-2.15%; controlling for baseline), whereas user message volume did not. Conclusion: Findings provide preliminary data for the potential of evidence-based conversational AI to extend mental health support for broad psychological functioning, extending the existing literature beyond symptom-based outcomes.
arXiv:2606.28078v1 Announce Type: cross Abstract: We demonstrate single-electron trapping and detection in a two-dimensionally scalable dual-plane printed-circuit-board Penning trap. We characterize deterministic electron loading, axial damping, axial temperature, and collision-induced magnetron-radius growth at low magnetic fields. These results establish a practical platform for planar Penning traps and identify key next steps toward applications in quantum information science.
arXiv:2606.28273v1 Announce Type: new Abstract: Vision-language models must reconcile visual evidence with memorized world knowledge when the two conflict. How they resolve this conflict shapes the reliability of multimodal systems, yet prior work characterizes it behaviorally without a component-level causal account. We combine activation patching across three granularities (residual stream, attention heads, and MLP sublayers) with model-component ablation studies and mechanistic analysis. Across three VLM families, we find that visual grounding emerges by default, whereas prior grounding depends on a small set of causally necessary attention heads (2.5-4.8%) concentrated in the second half of the network. These heads enable answers from stored world knowledge (e.g., "red" for a strawberry) despite conflicting visual input. Ablating them flips predictions from knowledge-grounded to visually grounded answers in 68-96% of cases under prior-knowledge prompts, but changes only 0.8-7.5% of visually grounded predictions, establishing an asymmetric causal structure. The identified heads decompose into routing heads, which modulate information flow, and writing heads, which directly project answer tokens into the residual stream. This structure is consistent across model families and scales, revealing a sparse causal circuit underlying perception-knowledge conflict in VLMs.
arXiv:2606.28274v1 Announce Type: new Abstract: Accurate network traffic prediction is a critical element for efficient resource allocation in dynamic urban cellular networks. However, prediction remains challenging because network demand is influenced by complex mobility patterns, congestion dynamics, and heterogeneous user behavior. This paper introduces the Parameter-Efficient Hybrid Transformer (PEHT), a network traffic prediction framework that integrates urban mobility and congestion information into a Transformer-based architecture. PEHT separates primary network communication features from secondary urban mobility features and incorporates Low-Rank Adaptation (LoRA) into the Transformer encoder to reduce the number of trainable parameters while maintaining high predictive accuracy. A multimodal fusion strategy then injects external mobility and congestion features into the decoder to improve traffic forecasting. Experiments on the Telecom Italia Milan dataset and multiple synthetic congestion scenarios show that PEHT outperforms state-of-the-art baselines in terms of RMSE, MAE, and $R^2$. The implementation is available in the GitHub repository.
arXiv:2407.01014v2 Announce Type: replace Abstract: Diffusion models excel in solving imaging inverse problems due to their ability to model complex image priors. However, their reliance on large, clean datasets for training limits their practical use where clean data is scarce. In this paper, we propose EMDiffusion, an expectation-maximization (EM) approach to train diffusion models from corrupted observations. Our method alternates between reconstructing clean images from corrupted data using a known diffusion model (E-step) and refining diffusion model weights based on these reconstructions (M-step). This iterative process leads the learned diffusion model to gradually converge to the true clean data distribution. We validate our method through extensive experiments on diverse computational imaging tasks, including random inpainting, denoising, and deblurring, achieving new state-of-the-art performance.
arXiv:2606.28321v1 Announce Type: new Abstract: We present StructSplat, a feed-forward and generalizable 3D Gaussian reconstruction framework that operates directly on uncalibrated images without requiring camera parameters. Existing methods either rely on per-scene optimization or assume known camera poses, and often entangle geometry and appearance within a unified backbone, limiting reconstruction fidelity and generalization. Our key idea is to adopt a structured representation that organizes geometry, semantic, and texture cues with explicit roles in the reconstruction process. Specifically, we introduce a pixel-aligned feature injection mechanism to enable accurate texture modeling from 2D observations, incorporate semantic-aware priors to improve global consistency, and design a camera alignment strategy to prevent information leakage and improve generalization. Experiments show that our method significantly outperforms prior approaches on challenging benchmarks. On DL3DV, our method achieves 28.045 PSNR, surpassing AnySplat (22.377) by +5.67 dB. In cross-dataset evaluation, our method achieves +1.94 dB over AnySplat on ACID and +1.72 dB on RealEstate10K. Project page: https://structsplat.github.io Code: https://github.com/J-C-Zhao/StructSplat
arXiv:2606.27411v1 Announce Type: cross Abstract: We study a quantum autoencoder (QAE) for compression-driven anomaly detection in brain MRI data. The approach leverages angle encoding to map image patches into quantum states, followed by a variational encoder-decoder architecture trained to discard information via auxiliary trash qubits. Anomaly scores reflect the degree to which inputs resist compression relative to normal data, with higher scores corresponding to deviations from the learned normal manifold. Evaluated on publicly available brain MRI DICOM datasets, the method achieves a slice-level ROC-AUC of approximately 0.95 and a patch-level ROC-AUC of approximately 0.813, outperforming classical autoencoder and PCA baselines. Analysis of the learned parameters reveals a pronounced encoder-decoder asymmetry, where effective anomaly detection arises from structured information compression within the encoder rather than increased parameter magnitude or decoder expressivity. This results in a controlled compression-reconstruction trade-off with a clear operating regime that supports principled threshold selection. Qualitative evaluation further shows that the QAE produces spatially localized anomaly heatmaps aligned with tumorous regions. The results, supported by promising baseline performances, demonstrate that quantum autoencoders provide an interpretable and controllable mechanism for anomaly detection based on incompressibility with respect to a learned latent representation. This work highlights the potential of quantum autoencoders as a principled tool for studying compression dynamics in quantum machine learning, with promising implications for decision support in medical imaging workflows.
arXiv:2606.27462v1 Announce Type: cross Abstract: The global minimum-variance portfolio (GMVP) is the canonical decision built from an estimated covariance matrix, yet covariance estimators are universally evaluated by matrix-norm loss, which is not the object the decision depends on. We characterise exactly how covariance-estimation error maps into GMVP suboptimality. We prove an exact regret identity and a non-asymptotic bound showing decision regret depends on the estimation error only through its action on the portfolio weights, scaled by portfolio concentration and the conditioning of the true covariance. From this we derive the decision geometry: GMVP regret is invariant to a (p-1)-dimensional projection of the p^2-dimensional error matrix, with invariance to the covariance-scale direction as an exact special case. We then apply the framework to heavy-tailed returns (tail index kappa in (2,4)), establishing the regret convergence rate implied by the centred operator-norm rate, and confirm the theory on a skew-t/t-copula simulation design with pre-registered analysis. The decision-focused advantage is a sharper constant and a concentration discount rather than a faster rate; we report an honest high-conditioning boundary of the rate prediction. The results complement recent decision-focused learning approaches by supplying the exact estimation geometry and consistency theory they lack.
arXiv:2606.27481v1 Announce Type: cross Abstract: We present a first study of a diffusion-based approach to accelerated sampling of the $N_f = 2$ lattice Schwinger model. Our work is inspired by recent and growing successes in developing such generative models for ensemble generation in LFT to overcome the well-known critical slowing down problem. We train a U(1)-equivariant score-based generative model to sample gauge link configurations from the marginal Schwinger model. By computing model likelihoods, we obtain unbiased estimates for observables that closely match those produced by MCMC simulations. We also demonstrate improvement over HMC as measured qualitatively by a reduction in topological freezing near critical parameters.
arXiv:2606.27928v1 Announce Type: cross Abstract: We present $\texttt{SMaSH}$, a $\texttt{Mathematica}$ package to do spinor helicity computations in four spacetime dimensions $\href{https://github.com/aakash-kmr/SMaSH}{\text{(github)}}$. It can handle massive spinor helicity computations with explicit little group indices which is a novel feature. It can also handle massless as well as off-shell spinor helicity variables. It is designed to compute perturbative computations; it comes with predefined three point amplitudes and propagators for any masses and spins (arXiv:1709.04891). It can implement the high energy limit over an expression, check the discrete $\tt{C,P,T}$ transformations, compute contact terms and impose gauge invariance for any scattering process. We have shown the usage of such functions for computing gauge invariant Weinberg minimal amplitudes (arXiv:2506:12431, arXiv:2504:06343). The package can also generate both real and complex numerical kinematics for any $n$-point scattering for arbitrary masses and energy scales by implementing the $\tt{RAMBO}$ algorithm. It is also rich with basic spinor helicity manipulations like Schouten simplification, Clifford algebra manipulation, conversion between spinor helicity and Lorentz vectors, derivative w.r.t. spinors and their scalars, helicity scaling etc.
arXiv:2310.12375v3 Announce Type: replace Abstract: We study monotonicity testing of functions $f \colon \{0,1\}^d \to \{0,1\}$ using sample-based algorithms, which are only allowed to observe the value of $f$ on points drawn independently from the uniform distribution. A classic result by Bshouty-Tamon (J. ACM 1996) proved that monotone functions can be learned with $\exp(\widetilde{O}(\min\{\frac{1}{\varepsilon}\sqrt{d},d\}))$ samples and it is not hard to show that this bound extends to testing. Prior to our work the only lower bound for this problem was $\Omega(\sqrt{\exp(d)/\varepsilon})$ in the small $\varepsilon$ parameter regime, when $\varepsilon = O(d^{-3/2})$, due to Goldreich-Goldwasser-Lehman-Ron-Samorodnitsky (Combinatorica 2000). Thus, the sample complexity of monotonicity testing was wide open for $\varepsilon \gg d^{-3/2}$. We resolve this question, obtaining a nearly tight lower bound of $\exp(\Omega(\min\{\frac{1}{\varepsilon}\sqrt{d},d\}))$ for all $\varepsilon$ at most a sufficiently small constant. In fact, we prove a much more general result, showing that the sample complexity of $k$-monotonicity testing and learning for functions $f \colon \{0,1\}^d \to [r]$ is $\exp(\Omega(\min\{\frac{rk}{\varepsilon}\sqrt{d},d\}))$. For testing with one-sided error we show that the sample complexity is $\exp(\Theta(d))$. Beyond the hypercube, we prove nearly tight bounds (up to polylog factors of $d,k,r,1/\varepsilon$ in the exponent) of $\exp(\widetilde{\Theta}(\min\{\frac{rk}{\varepsilon}\sqrt{d},d\}))$ on the sample complexity of testing and learning measurable $k$-monotone functions $f \colon \mathbb{R}^d \to [r]$ under product distributions. Our upper bound improves upon the previous bound of $\exp(\widetilde{O}(\min\{\frac{k}{\varepsilon^2}\sqrt{d},d\}))$ by Harms-Yoshida (ICALP 2022) for Boolean functions ($r=2$).
arXiv:2405.13890v4 Announce Type: replace Abstract: This paper was a Workshop Paper. See the full paper which will be presented at CHI 2026: arXiv:2501.10551; As large language models (LLMs) become more powerful and ubiquitous, systems like ChatGPT are increasingly used by students to help them with writing tasks. To better understand how these tools are used, we investigate how students might use an LLM for essay writing, for example, to study the queries asked to ChatGPT and the responses that ChatGPT gives. To that end, we plan to conduct a user study that will record the user writing process and present them with the opportunity to use ChatGPT as an AI assistant. This study's findings will help us understand how these tools are used and how practitioners -- such as educators and essay readers -- should consider writing education and evaluation based on essay writing.
arXiv:2406.09733v3 Announce Type: replace Abstract: Searching for a unified scene representation remains a research challenge in computer graphics. Traditional mesh-based representations are unsuitable for dense, fuzzy elements and introduce additional complexity for filtering and differentiable rendering. Conversely, voxel-based representations struggle to model hard surfaces and high-frequency details. We propose a general-purpose rendering primitive based on 3D Gaussian distributions for unified scene representation, featuring versatile appearance ranging from glossy surfaces to fuzzy elements, as well as physically based scattering to enable accurate global illumination. We formulate the rendering theory for the primitive based on non-exponential transport and derive efficient rendering operations to be compatible with Monte Carlo path tracing. The new representation can be converted from different sources, including meshes and 3D Gaussian splatting, and further refined via transmittance optimization thanks to its differentiability. We demonstrate the versatility of our representation in various rendering applications such as global illumination and appearance editing, while naturally supporting arbitrary lighting conditions. With suitable simplification, we further adapt our method to radiance field reconstruction and rendering. We conduct comprehensive comparisons of our representation with existing scene representations, highlighting its efficiency in capturing details and representing aggregate elements.
arXiv:2504.01882v2 Announce Type: replace Abstract: The use of DNS over HTTPS (DoH) tunneling by an attacker to hide malicious activity within encrypted DNS traffic poses a serious threat to network security, as it allows malicious actors to bypass traditional monitoring and intrusion detection systems while evading detection by conventional traffic analysis techniques. ML techniques can be used to detect DoH tunnels; however, their effectiveness relies on large datasets containing both benign and malicious traffic. Sharing such datasets across entities is challenging due to privacy concerns. In this work, we propose CO-DEFEND framework that enables multiple entities to collaboratively train a classification machine learning model for DoH threat detection while preserving data privacy, enhancing scalability and resilience against single points of failure. The proposed DFL framework provides a realistic implementation for DoH threat detection, enabling multiple entities to train their local models online with incoming DoH flows in real-time batches as they are processed - an approach that fits naturally within modern Internet architectures. This framework adapts four classical machine learning algorithms, Support Vector Machines, Logistic Regression, Decision Trees, and Random Forest, for federated scenarios and efficient training. In addition, a key methodological feature of CO-DEFEND is the use of DT and RF as model selection rather than aggregation mechanisms, allowing each participant to retain interpretable and locally optimal decision structures while benefiting from collective updates. We compare our proposed method by using the dataset CIRA-CIC-DoHBrw-2020 with existing machine learning approaches, including more computationally complex alternatives such as neural networks, to demonstrate its effectiveness in detecting malicious DoH tunnels while improving scalability and computational efficiency.
arXiv:2506.01442v2 Announce Type: replace Abstract: Reinforcement learning (RL) remains fundamentally limited by poor data efficiency and weak generalization. Prior episodic RL methods attempt to alleviate this via external memory modules, yet they suffer from two key limitations: a representation bottleneck caused by shallow encoders, and a retrieval dilemma where episodic memory is accessed indiscriminately. To address these challenges, we propose Agentic Episodic Control (AEC), a novel architecture that integrates large language models (LLMs) into episodic RL. AEC uses an LLM-based semantic augmenter to generate semantic representations from raw observations, and a critical state recognizer to selectively retrieve valuable experiences. This transforms memory usage from passive similarity matching into strategic, context-aware recall. Across five BabyAI-Text environments, AEC achieves 2-6x higher data efficiency than baselines and is the only method to solve complex tasks like UnlockLocal with over 90% success. It further demonstrates strong cross-task and cross-environment generalization, maintaining performance even under distribution shifts. AEC shows that combining LLM-derived priors with reinforcement learning yields more sample-efficient and adaptable agents.
arXiv:2512.01693v3 Announce Type: replace Abstract: Metal-organic framework (MOF) databases have grown rapidly through experimental deposition and large-scale literature extraction, but recent analyses show that nearly half of their entries contain substantial structural errors. These inaccuracies propagate through high-throughput screening and machine-learning workflows, limiting the reliability of data-driven MOF discovery. Correcting such errors is exceptionally difficult because true repairs require integrating crystallographic files, synthesis descriptions, and contextual evidence scattered across the literature. Here we introduce LitMOF, a large language model-driven multi-agent framework that validates crystallographic information directly from the original literature and cross-validates it with database entries to repair structural errors. Applying LitMOF to the experimental MOF database (the CSD MOF Subset), we constructed LitMOF-DB, a curated set of 189,567 computation-ready structures, including the successful repair of 9,227 invalid entries, which accounts for 69.1% of the CSD-derived not-computation-ready MOFs in the latest CoRE MOF DB. Additionally, the system uncovered 8,771 experimentally reported MOFs absent from existing resources, substantially expanding the known experimental design space. Using direct air capture screening as a case study, we demonstrate that structural errors severely distort predicted adsorption energies and CO2/H2O selectivity, leading to systematic misranking of materials, false positives, and the omission of high-performance candidates. This work establishes a scalable pathway toward self-correcting scientific databases and a generalizable approach for LLM-driven curation in materials science.
arXiv:2603.12001v2 Announce Type: replace Abstract: Distributed AI and IoT applications increasingly execute across heterogeneous resources spanning end devices, edge/fog infrastructure, and cloud platforms, often under different administrative domains. Fluid Computing has emerged as a promising paradigm for enhancing massive resource management across the computing continuum by treating such resources as a unified fabric, enabling optimal service-agnostic deployments driven by application requirements. However, existing solutions remain largely centralized and often do not explicitly address multi-domain considerations. This paper proposes an agnostic multi-domain orchestration architecture for fluid computing environments. The orchestration plane enables decentralized coordination among domains that maintain local autonomy while jointly realizing intent-based deployment requests from tenants, ensuring end-to-end placement and execution. To this end, the architecture elevates domain-side control services as first-class capabilities to support application-level enhancement at runtime. As a representative proof of concept, we instantiate the architecture through a distributed AI use case; specifically, we consider a multi-domain Decentralized Federated Learning (DFL) deployment under Byzantine threats. Under this setting, we leverage domain-side capabilities to enhance Byzantine security by introducing FU-HST, an SDN-enabled multi-domain anomaly detection mechanism that complements Byzantine-robust aggregation. We validate the use-case workflow via simulation in single- and multi-domain settings, evaluating anomaly detection, DFL performance, and computation/communication overhead.
arXiv:2601.14632v2 Announce Type: replace Abstract: Contact tracing (CT) is a frontline measure against emerging epidemics, yet in practice it is never complete. The quantitative impact of missing information -- such as untraced cases or unnotified contacts -- on the effectiveness of CT remains insufficiently understood. Using a stochastic agent-based model with sociodemographics from metropolitan areas in South Korea, we simulate how different forms of information loss affect epidemic spreading dynamics. We construct information-loss scenarios based on two types: infector-omission (IO), the omission of infected individuals from the tracing process, and contact-omission (CO), the omission of specific contact events even when the infected individuals themselves are identified. The sensitivity of epidemic dynamics to increasing omission rates differs markedly between the two types: IO produces substantially stronger and more abrupt changes in transmission structure and epidemic outcomes, whereas CO produces more gradual effects. Notably, CT effectiveness breaks down beyond a city-specific threshold -- an IO rate of approximately 4% in Seoul but about 10% in less populous Busan -- underscoring that CT strategies must be tailored to regional population and mobility structure. Both IO and CO scenarios also lead to an increase in the transmission network diameter as information loss grows, indicating that a small network diameter reflects effective contact tracing that limits the depth of transmission chains. Collectively, our results offer threshold estimates and practical guidance for designing robust CT systems in the real world.
arXiv:2604.21211v2 Announce Type: replace Abstract: Current text anonymization evaluation relies on span-based metrics that fail to capture what an adversary could actually infer, and assumes a single data subject, ignoring multi-subject scenarios. To address these limitations, we present SPIA (Subject-level PII Inference Assessment), the first benchmark that shifts the unit of evaluation from text spans to individuals, comprising 675 documents across legal and online domains with novel subject-level protection metrics. Extensive experiments show that even when over 90% of PII spans are masked, subject-level inference protection drops as low as 33%, leaving the majority of personal information recoverable through contextual inference. Furthermore, target-subject-focused anonymization leaves non-target subjects substantially more exposed than the target subject. We show that subject-level inference-based evaluation is essential for ensuring safe text anonymization in real-world settings.
arXiv:2606.27406v1 Announce Type: new Abstract: Software engineering, whether performed by humans or by AI agents, requires reasoning about how software behaves. We call the internal model that supports such reasoning the software world model, and view current code-execution benchmarks as covering one well-studied slice of it -- control flow. In this paper, we take a step toward a broader evaluation by shifting the observable axis to execution resources: alongside test outcome and exception class, we predict peak memory, wall-clock time, and ranked profiler outputs at method and line granularity. We use SWE-bench Verified as the source of data to hold the test close to real-world software engineering tasks. All tested models, frontier ones included, show modest performance and brittle behaviour, suggesting a notable lack of understanding of how software is executed, as opposed to how its source code is written.
arXiv:2606.02379v2 Announce Type: replace Abstract: Metric scale monocular geometry estimation has seen significant progress through large-scale data aggregation, yet current foundation models suffer from a persistent ''scale-collapse'' phenomenon: distant landmarks and vast landscapes are metrically underestimated. We hypothesize that this performance gap stems from a training data bottleneck, where existing metric-scale datasets are hardware-constrained to homogenous vehicle-captured LiDAR or short-range indoor scans, or consist of synthetic data that lacks the semantic complexity of the physical world. To bridge this gap, we curate a new metrically-grounded, in-the-wild dataset that we call MetricScenes, gathered from a variety of sources including Internet photo collections and stereo imagery. We estimate camera poses and initial depth maps for each scene using off-the-shelf methods, and recover absolute scale from geo-tagged metadata as well as known stereo camera baselines. We also improve the quality of depth maps derived from MetricScenes via a new two-stage Poisson completion method. Fine-tuning MoGe-2 on our dataset significantly mitigates scale-collapse and achieves superior metric accuracy in unconstrained, open-domain scenes while maintaining state-of-the-art performance on standard benchmarks.
arXiv:2606.17524v2 Announce Type: replace Abstract: Large language models show strong reasoning ability, but their internal reasoning process can remain unstable in complex multi-step settings, where early hidden-state errors may propagate to incorrect predictions. We propose ReLAR, a reinforcement-guided latent refinement framework that iteratively updates hidden representations before decoding. ReLAR maintains a compact latent reasoning state and uses learned depth and action controllers to adaptively determine both the number and direction of refinement steps. The controllers are trained with a policy gradient objective based on step-wise likelihood improvement, enabling efficient input-dependent reasoning without explicit chain-of-thought generation. Experiments on medical, mathematical, multi-hop reasoning, and open-ended generation benchmarks show that ReLAR improves accuracy, generation quality, and reasoning stability with substantially lower inference overhead than explicit reasoning baselines.