Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 4 September
Understanding how concepts are encoded in the internal representations of machine learning models is a central problem in mechanistic interpretability, essential both for the science of deep learning and for the trustworthy deployment of increasingly capable models. Existing approaches to interpret model representations mainly map representations onto more interpretable spaces and do not directly characterize how concepts occupy representation space; various hypotheses have been proposed, but often lack of rigorous validation and largely focus on static representations. In this work, we introduce Capsule Lens, a framework that matches the region a concept occupies with a simple, trackable geometric form, a capsule, defined by several interpretable parameters, fitted in closed form to each concept's geometry and validated on held-out samples. We apply Capsule Lens in two major settings: static and dynamic representations. On static representations, we demonstrate how to locate concept geometry across various models, and how the span and norm curves uncover important geometric characteristics. On dynamic representations, we present three case studies tracking representation drifts induced by distinct training settings, CLIP pretraining, RL post-training on visual question answering, and RL post-training on mathematical reasoning. These analyses reveal qualitatively different geometric dynamics, ranging from broad network-wide restructuring in CLIP pretraining to localized and concept-specific changes in RL post-training. Our results include findings aligned with existing literature as well as novel observations. We believe Capsule Lens stands as a promising tool for locating, analyzing, and tracking concept geometry in both static and dynamic representations.
Machine Learning
Multi-granularity Adaptive Hypergraph Representation Learning via Granular-ball
Hypergraph representation learning aims to capture high-order information in graphs by constructing hyperedges that simultaneously connect multiple nodes. These hyperedges adapt to the graph's topological features, facilitating the extraction of high-order relationships at multiple granularities. Most prior work relies on predefined definitions to generate hyperedges, overlooking the diversity in graph topological structures and the multi-granularity characteristics of hyperedges. As a result, this limits their ability to effectively and adaptively discover high-order relationships and efficiently process complex structural information. To address this limitation, we propose a novel framework called \underline{M}ulti-\underline{G}ranularity \underline{H}ypergraph \underline{R}epresentation \underline{L}earning (MGHRL). MGHRL introduces an Adaptive Granular Hypergraph Generation strategy, which generates hyperedges at multiple levels of granularity through the adaptive splitting of granular-ball, effectively capturing high-order relationships based on the graph's topological structure. Additionally, we propose a Multi-Granularity Hypergraph Network with multiple sub-networks, capturing features from hyperedges at different granularities and integrating them via hierarchical reversible connections. Experimental results show that MGHRL significantly outperforms baseline models on benchmark datasets.
Read the full abstract
Differential Geometry
Go Fishing with Alice: Compatible Carrollian and Poisson Structures
We introduce, motivated by noncommutative Carrollian geometry, the notion of a Carrollian–Poisson bundle understood as a Carrollian manifold with a compatible Poisson structure. Compatibility is expressed as the inclusion of the Carroll distribution in the image of the Poisson anchor. We explore the geometric consequences of this compatibility and present several simple examples of Carrollian–Poisson bundles, including the (outer) Kerr horizon, the BTZ horizon, and future null infinity. Furthermore, we establish that every Carrollian–Poisson bundle can be equipped with a Vaisman (contravariant) connection that is both metric-compatible and clock-compatible, and thus geometric kinematics can be defined and studied.
Read the full abstract
Artificial Intelligence
Deep belief networks are exact
We prove that every strictly positive probability distribution on \(\{-1,1\}^n\) is represented exactly by a sigmoid belief network with finite parameters. This answers a question of Sutskever and Hinton. The proof upgrades their probability-sharing approximation to exact representation using Brouwer's fixed-point theorem.
Read the full abstract
Software Engineering
Grounded Skill Synthesis from Code at Scale for Agentic Intelligence
Reusable skills give agents transferable procedural knowledge, making scalable acquisition essential for extending agents beyond prior experience. Existing methods face two limitations: trajectory-based synthesis requires interactions with specific environments, while document-derived skills may lack executable evidence and verification. Source code offers a complementary path: it requires no prior agent experience yet provides executable evidence for grounding abstractions. We present Code2Skill, a fully automated pipeline that transforms selected code units into implementation-anchored records of atomic operations, composite workflows, and recurring patterns, then verifies each record through source-body-blind reconstruction and source-aware comparison. Applied to 19,769 popular, actively maintained GitHub repositories, Code2Skill produces CodeSkillBank, a grounded bank of 1,006,822 accepted records with workflow, boundary, provenance, and source-evidence metadata. Across 72 protocol-matched evaluations covering nine model settings and eight benchmarks, models augmented with retrieved CodeSkillBank skills improve by 11.7% on average over matched baselines and outperform them in 57 cases. Under a unified downstream interface, Code2Skill also outperforms trajectory-derived skill banks on all seven shared benchmarks, showing that repository-derived skills can provide useful procedural knowledge before agents accumulate sufficient interaction experience. Skills synthesized from tested AI-generated code achieve a 93.50% pass rate, compared with 93.00% for human-written code, providing initial evidence that the pipeline can expand with the growing volume of AI-generated software. Overall, Code2Skill transforms procedural knowledge embedded in repositories into grounded, verifiable, and transferable agent skills.
Read the full abstract
Instrumentation and Detectors
Nanoscale NMR of DNA-Functionalized Diamond Using Individual NV Centers
Nuclear magnetic resonance (NMR) spectroscopy is a powerful probe of molecular structure and dynamics but typically requires large spin ensembles. Here, we demonstrate nanoscale NMR detection of fluorinated DNA using individual near-surface nitrogen-vacancy (NV) centers in diamond nanopillar waveguides under ambient conditions. We covalently tether $^{19}$F-labeled oligonucleotides to mixed nitrogen- and oxygen-terminated surfaces, localizing target spins within the nanoscale sensing volume. Probabilistic spin simulations suggest that the measured signals are consistent with those expected from a few molecules and depend strongly on NV-molecule separation, configuration, and local surface heterogeneity. These results establish a framework for NMR of surface-tethered biomolecules and identify key challenges for single-molecule structural elucidation.
Read the full abstract
Robotics
Information-Guided Safe Reinforcement Learning for Autonomous Gas Source Localization using sUAS
The autonomous localization of fugitive gas emissions using small Unmanned Aircraft Systems (sUAS) constitutes a fundamentally ill-posed inverse problem. In turbulent atmospheric boundary layers, highly intermittent scalar concentration fields violate the assumptions of classical gradient-based navigation, causing data-driven estimators to suffer from severe noise and spurious local minima. To address these challenges, we introduce an Information-Guided Safe Reinforcement Learning framework evaluated within a custom, GPU-accelerated 3D simulation environment coupling an Eulerian wind solver with a Lagrangian puff dispersion model. We identify a critical vulnerability in deterministic information-seeking planners - a Gramian bias where agents act greedily upon flawed early estimates, starving the estimator of spatial diversity. To systematically break this degeneracy, our architecture integrates a classical empirical observability Gramian (EMGR) planner with a learned Soft Actor-Critic (SAC) exploratory policy. A deterministic meta-supervisor actively monitors estimator reliability via Kullback-Leibler (KL) divergence, dynamically blending deterministic exploitation with learned exploration to steer the sUAS into high-information zones. Trained via a progressive curriculum and safeguarded by a strictly enforced Robust Control Barrier Function (RCBF), our RL framework achieves nearly 80% localization success on complex, mobile sources - drastically outperforming classical baselines (~30%) - while ensuring zero safety violations.
Read the full abstract
Combinatorics
The Erd\H os-Sós conjecture in dense graphs
The Erd\H os–Sós conjecture states that every $n$-vertex graph with more than $(k-2)n/2$ edges contains every $k$-vertex tree. We prove that for every $\gamma$ there is an $n_0$ such that for all $n\ge n_0$ and $k \ge \gamma n$ the conjecture holds. As a corollary of our result, we obtain a solution of a 51-year-old problem of Erd\H os and Graham on the multicolor Ramsey numbers of trees.
Read the full abstract
Computer Vision
WorldSculpt: Generating Compositional Worlds from Grounded Videos
We study the problem of generating a compositional 3D representation of a cluttered scene containing hundreds of objects. The goal is to represent the scene as a collection of individual object meshes placed in a shared world frame, as required by downstream applications such as gaming, AR/VR, simulation, and robotics. This task is challenging in densely cluttered scenes, where objects heavily occlude one another and each view reveals only a fraction of their geometry. Geometry-based approaches typically reconstruct the scene as a single representation and leave incomplete geometry in occluded regions, while existing compositional methods with generative priors are largely limited to relatively simple scenes. We show that complex scenes with hundreds of objects can instead be generated compositionally by adapting a strong single-object 3D generative prior to multi-view observations. We instantiate this paradigm with Pixal3D, extending it with a multi-view conditioning pathway that grounds object generation in multiple posed observations. Although the model is finetuned entirely on single objects in canonical space, it generalizes to large scenes with severe occlusion without any scene-level training, demonstrating the feasibility and scalability of this paradigm. We further introduce UE-MeshyScene, a photorealistic benchmark of densely cluttered scenes with hundreds of objects, per-object annotations, and ground-truth meshes. Across single-object, controlled multi-object, and UE-MeshyScene evaluations, our method consistently outperforms prior approaches, with larger gains as scene complexity and occlusion increase. Finally, we demonstrate broader applicability by converting generated 3DGS worlds, such as Marble and HY-World 2.0, into compositional mesh scenes.
Read the full abstract
Computer Vision
UniMate: One Unified Model to Animate Diverse Skeletons
Recent advances in automatic rigging now deliver animation-ready 3D assets at scale, yet generating the motion to drive them remains a bottleneck. Existing learned animators are topology-constrained: they rely on category-specific templates or require per-skeleton fine-tuning and reference motions at inference. We present UniMate, a unified foundation model that synthesizes articulated motion for arbitrary skeletons from a rigged 3D asset and a text prompt, with no test-time optimization or per-skeleton retraining. UniMate introduces a topology-aware diffusion transformer, which integrates skeletal topology into attention via three mechanisms: (1) a graph-aware attention bias from pairwise joint relations and geodesic distances; (2) a spectral rotary position embedding generalizing RoPE to arbitrary kinematic trees via the graph Laplacian; and (3) a global topological conditioner attention-pooled from the rest-pose skeleton. We also curate UniML3D, 13,006 motion sequences spanning bipedal, quadrupedal, avian, marine, insectoid, serpentine, and articulated rigid objects with unified canonicalization and text pairing. Trained on this dataset, UniMate outperforms state-of-the-art baselines in quality, generalization, and efficiency, and supports zero-shot cross-topology transfer, in-betweening, expansion, and text-guided editing. Our project page is available at https://linzhanmou.com/unimate/.
Read the full abstract
Combinatorics
Non-Hamiltonian $\frac{3}{2}$-Tough Plane Triangulations
By Tutte's classic theorem of 1956 that every 4-connected planar graph is Hamiltonian, every planar graph of order at least three with toughness greater than $\frac{3}{2}$ is Hamiltonian. In 1999, Owens constructed a sequence of maximal planar graphs whose toughness approaches $\frac{3}{2}$ from below and which do not contain even a 2-factor, and he asked whether there exists a maximal planar graph with toughness exactly $\frac{3}{2}$ and with no 2-factor. In 2025, Shan constructed a $\frac{3}{2}$-tough plane triangulation with no 2-factor. In that construction, there are many pairs of vertices of degree $3$ that have a common neighbor. By imposing a distance condition on the vertices of degree $3$, Hao, Ma, Shan, and Yang recently proved that every $\frac{3}{2}$-tough plane triangulation of order at least three whose vertices of degree $3$ are pairwise at distance at least $3$ has a 2-factor, and they asked whether every such graph is Hamiltonian. We answer this question in the negative, and in fact prove the following stronger statement: for every positive integer $\ell$, there exists a $\frac{3}{2}$-tough non-Hamiltonian plane triangulation whose vertices of degree $3$ are pairwise at distance at least $\ell$. Thus, although the distance condition guarantees the existence of a 2-factor, it does not guarantee that the graph is Hamiltonian: the essential obstruction to a Hamiltonian cycle is a certain local configuration involving a vertex of degree $3$, rather than the proximity of such configurations in the graph.
Read the full abstract
Earth and Planetary Astrophysics
Two extremely irradiated volatile-rich sub-Neptunes with companions in the TOI-426 and TOI-1839 systems: Insights into arrival and survival near the lower edge of the Neptunian desert
Using TESS photometry and 147 HARPS radial velocities, we confirm two extremely irradiated, volatile-rich sub-Neptunes near the lower edge of the Neptunian desert: TOI-426 b and TOI-1839 b, orbiting the solar-type stars HD 34390 and TYC 304-865-1. We modelled stellar activity with shared-timescale and multidimensional Gaussian processes. The planets have orbital periods of approximately 1.32 and 1.42 days, radii of $2.19\pm0.09$ and $2.27\pm0.10$ Earth radii, and masses of $6.7\pm1.8$ and $7.10\pm0.78$ Earth masses, respectively. They receive approximately 1700 and 1050 times Earth's irradiation, yet their densities require substantial volatile content. We also detect a transiting sub-Neptune, TOI-1839 c, at approximately 4.02 days and a giant companion, TOI-426 c, with a period of $235.8^{+9.2}_{-8.5}$ days and a minimum mass of $444^{+18}_{-17}$ Earth masses, assuming a circular orbit. Interior-structure and atmospheric-escape models yield mass-loss timescales of 1-100 Myr for hydrogen/helium envelopes, compared with 100-1000 Gyr for water-dominated envelopes, favouring a steam-world interpretation for all three transiting planets. A population analysis reveals a striking excess of detected outer giant companions to low-mass sub-Neptunes near the desert edge, where high-eccentricity tidal migration is expected to circularize surviving planets: 36% (9/25) within the tidal survival band versus 6% (9/148) outside it (Fisher exact test $p=1.4\times10^{-4}$). This excess is robust to changes in sample selection and companion definitions. The survival of these volatile-rich planets challenges a boundary set solely by evaporation of primordial hydrogen/helium envelopes, while the companion excess supports an important role for high-eccentricity tidal migration in shaping the lower desert edge.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Tracing the Cosmic Origins: Machine Learning Reconstruction of the Primordial Density Field from EoR Observations
Reconstructing the initial conditions of the Universe from late-time tracers would unlock cosmological information buried by non-linear structure formation and astrophysics. We reconstruct the initial density field corresponding to $z=300$, with its linear amplitude extrapolated to $z=0$, from simulated 21-cm and CO(1-0) line-intensity maps at $z=8$ generated with LIMFAST. Using a three-dimensional U-Net, we reconstruct the initial conditions and evaluate their impact on cosmological parameter constraints. The two tracers probe complementary environments: 21-cm emission traces neutral, low-density regions of the IGM, while CO traces overdense, star-forming regions. We also train and evaluate the reconstruction on maps incorporating SKA1-Low- and COMAP-ERA-like instrumental response and thermal noise. In the noiseless case, combining both tracers delivers the most accurate reconstruction across ionisation states, with the cross-correlation coefficient between the reconstructed and true initial density fields, $|C(k)| \gtrsim 0.90$ for $k \lesssim 0.75\,{\rm Mpc}^{-1}$. With observational effects, small-scale information is degraded, but the joint reconstruction retains substantial large-scale coherence, with $|C(k)| \gtrsim 0.70$ for $k \lesssim 0.3\,{\rm Mpc}^{-1}$. We quantify cosmological information gains through simulation-based inference using power-spectrum summaries pre- and post-reconstruction. The largest improvements occur for $\sigma_8$ and $n_{\rm s}$, with a median aggregate marginal compression of $1.17$ bits in the noiseless case, corresponding to a $56\%$ reduction in the factorised product of prior-normalised $68\%$ marginal widths. With observational effects, the corresponding median compression is $1.29$ bits. These results demonstrate that joint 21-cm-CO reconstruction recovers cosmological information beyond pre-reconstruction summaries.
Read the full abstract
Combinatorics
The extremal cases of the Erd\H os–Sós conjecture
The Erd\H os–Sós conjecture states that every $n$-vertex graph $G$ with more than $(k-2)n/2$ edges contains every $k$-vertex tree. We solve the extremal cases of this conjecture, showing that for some fixed $\mu>0$, the conjecture holds for each $G$ that minimally satisfies the assumptions of the conjecture and has a subgraph~$H$ of minimum degree $\delta(H)\ge (1-\mu)k$. In our proof, we mainly have to deal with $H$ taking two different shapes: either $H$ is close to the complete graph $K_k$ or $H$ is close to the complete bipartite graph $K_{k,k}$.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Early against Late: A contrast on dark energy in the light of DESI DR2
Recent findings from Dark Energy Spectroscopic Instrument (DESI) Baryon Acoustic Oscillation (BAO) measurements, combined with Type Ia supernovae and Cosmic Microwave Background (CMB) data, suggest potential parameter-level deviations from $\Lambda\text{CDM}$. However, parameter exclusions omit prior-volume penalties, whereas Bayesian evidence uncovers a stark dichotomy between early- and late-time dynamics. To quantify this effect, we perform a Bayesian model comparison that explicitly accounts for the cosmological epoch of dark energy dynamics, contrasting two schemes acting in opposite epochs of cosmic history: pre-recombination Early Dark Energy (EDE) and late-time Chevallier–Polarski–Linder (CPL), both measured against a baseline $\Lambda$CDM. We validate the learned harmonic-mean estimator against UltraNest (at background level) before applying it to the CMB analysis. Background-only data already disfavor both extensions, yielding {$\ln B_{\Lambda\text{CDM},\text{EDE}} = 1.48 \pm 0.18$ and $\ln B_{\Lambda\text{CDM},\text{CPL}} = 2.36 \pm 0.24$}. Including CMB sharpens this result: {EDE is very strongly rejected ($\ln B_{\Lambda\text{CDM},\text{EDE}} \approx 12.61 \pm 0.17$)} with its early-time fraction constrained to $10^3\,\Omega_e^{\rm EDE} = 2.0^{+0.6}_{-1.0}$, whereas CPL remains only mildly disfavored ($\ln B_{\Lambda\text{CDM},\text{CPL}} \approx 2.34 \pm 0.18$), even though 2D parameter posteriors for both models display a multi-$\sigma$ deviation from $\Lambda\text{CDM}$. Thus, Bayesian evidence weakens reported preferences for dark energy dynamics, with the impact depending on the cosmic epoch involved, confirmed by our functional reconstructions of $w_{\rm de}(z)$ and $\Omega_{\rm de}(z)$ from posterior samples.
Read the full abstract
High Energy Physics (Theory)
Decompactification and the Flat-Space Limit of AdS$_5\times$ S$^5$ Master Correlators
We study the flat-space limit of AdS$_5$/CFT$_4$ holographic correlators while accounting for the simultaneous decompactification of the internal S$^5$. We show that this limit is naturally formulated in terms of master propagators and correlators, which resum the infinite tower of Kaluza–Klein modes. For boundary correlators, the resulting expressions can be interpreted in terms of amplitudes with kinematics restricted to 5d. We then consider master correlators whose external insertions are kept in the AdS$_5\times$S$^5$ bulk before taking the limit, and argue that they recover flat-space amplitudes with unrestricted kinematics in 10d Minkowski space. Finally, we show that the flat-space limit commutes with the boundary limit if we analytically continue the sphere so that the bulk geometry becomes AdS$_5 \times$dS$_5$, and the resulting formulae suggest an interpretation in terms of amplitudes in flat space with two time directions and with 10d kinematics satisfying two simultaneous null constraints. We discuss the implications of these constructions for flat-space holography and its relation to the Carrollian limit of the dual CFT.
Read the full abstract
Quantum Physics
Towards Scaling Quantum Fine-Tuning of Foundational Time Series Models for Classification
Time-series foundation models produce rich embeddings, but whether quantum models can exploit them, and how far hybrid classical-quantum architectures scale, remains unclear. We address this by fine-tuning Chronos for power-grid event classification (PSML-5) with a quantum head on the model's embeddings. Grouping embeddings by physical sensor type before summarization already surpasses the best published baseline built for this benchmark, and with finer-grained features the quantum head outperforms a larger classical multilayer perceptron on identical inputs by 1.7-2.0 percentage points of balanced accuracy. Yet the gains saturate: past a point, feeding more information to the same fixed-width register yields no improvement. We show the bottleneck is neither the supply of information nor circuit expressiveness, but the bandwidth of the data intake. To overcome this limitation, we introduce the wing module, a self-contained few-qubit circuit that feeds additional information into the core circuit through a sparse, one-way coupling. Under a preregistered four-seed protocol, we attach wings to a fixed 12-qubit core with fixed features. Balanced accuracy increases with each added wing, from 83.6% with no wings (13 qubits, including a post-selection qubit) to 85.2% with two (19 qubits). Ablations establish that a circuit enlarged without new information gains nothing, while a wing fed information from the wrong sample harms accuracy. These results reframe scaling for quantum fine-tuning: added qubits help when they carry added inputs, not merely more parameters. Wings offer a modular and stable route to widening that bandwidth.
Read the full abstract
General Relativity and Quantum Cosmology
Nonlinear evolution in Galileon EFTs: Regularization and screening
In a scalar effective field theory (EFT) with Galileon symmetry, we distinguish the strong-field expansion from the derivative expansion and study spherically symmetric classical nonlinear dynamics. More concretely, we consider two models related via perturbative field redefinitions: one with second-order equations that can also exhibit screening, and one that includes a higher-derivative term that introduces a propagating ghost. The latter term is of the type that can “regularise” the equations to render them well-posed as an initial value problem. For initial data that respect the derivative expansion, we find that the first model (ghost-free, unregularised) develops ill-posed regions during evolution only when the derivative expansion breaks down. This result persists in the strong-field regime. We further find that the regularized model reproduces the evolution of the unregularised one when the latter remains well-posed, while it exhibits tachyonic behaviour when the unregularised one becomes ill-posed. We also consider initial data that corresponds to stationary states that exhibit screening in the unregularised, ghost-free theory. In this case, we find that the regularised theory can remain well-posed for initial data that the unregularised one developed ill-posed regions. However, the ghost term naturally dominates over the Galileon term in the screening regime, contrary to common expectation.
Read the full abstract
Systems and Control
Beyond Scalar Flexibility: From Eligible AI Workloads to Dependable Load Relief
Grid studies often represent data-center flexibility as a fixed percentage of load, although no public production trace has shown how much eligible load persists across event durations or co-moves across clusters. We reconstruct 4,439 hourly power observations from a 185-day trace of 155,410 GPUs and derive a workload-semantic flexibility envelope. The fleet's time-averaged Monte Carlo median facility demand is 55.8 MW, while immediate eligible curtailment averages 3.55 MW after retaining allocated-GPU idle power: 12.1% of workload power and 6.35% of median facility power. Under full realization of that eligibility, 95%-available relief falls from 2.51 MW for one hour to 2.32 MW for four hours and 1.95 MW for 24 hours; a common realizable fraction q scales every value exactly by q. A mean-calibrated scalar overstates these quantities by 17%, 25%, and 47%, while a scalar tail-calibrated at four hours understates the one-hour product by 6% and overstates the 24-hour product by 17%; the share that reproduces the surface varies by a factor of 1.6 across durations and reliability levels. Aggregating 13 clusters raises four-hour firmness from 0.38 to 0.66, but cross-cluster covariance limits the gain. The production scheduler exposes almost no additional delay-based capacity: newly deferrable arrivals average 0.008 MW and have zero 95%-available capacity. These results replace an assumed flexibility percentage with duration, reliability, portfolio, and realizability terms that can be written into interconnection and demand-response contracts.
Read the full abstract
Computation and Language
WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data
Recent advances in wearable sensing enable continuous monitoring of physiological and behavioral signals, yet existing benchmarks rarely evaluate whether AI systems can reason over a real user's longitudinal wearable record. We introduce WearableQA, a benchmark comprising 4,084 10-option multiple-choice questions constructed from the wearable time series, blood biomarkers, and demographics of 200 real users, each with up to 500 days of daily measurements. WearableQA preserves authentic wearable distributions that include device noise and inter-individual variability. To evaluate distinct reasoning capabilities, we introduce 16 question types organized along two complementary axes: data versus health reasoning, which distinguishes computation over longitudinal measurements from physiological interpretation; and single- versus cross-signal reasoning, which separates reasoning about individual signals from the integration of multiple signals. To construct reliable questions at scale, we adopt a dual-grounding framework that combines literature-grounded physiological findings with statistically validated population-grounded physiological patterns. This enables the capture of meaningful relationships observed in real-world wearable data. Evaluation of 14 proprietary and open-source LLMs demonstrates that WearableQA effectively differentiates model capabilities, with performance ranging from 19.6% to 72.9% against a 10% chance baseline. Moreover, WearableQA remains far from solved: most models achieve accuracies below 60%. Overall, WearableQA provides a realistic and diagnostic benchmark for evaluating LLM reasoning over real-world wearable data.
Read the full abstract
Human-Computer Interaction
Diffusion TV: Experiencing Diffusion Models through Tangible, Embodied Interaction
Diffusion TV is an interactive AI art installation that offers a tangible and embodied experience of diffusion models through a modified CRT TV. By physically manipulating the TV's antenna, audiences control the clarity of AI-generated images and sounds, metaphorically enacting the denoising process that underlies diffusion-based generation. Using the tuning knob, participants switch between three channels featuring AI-generated animals from the Past (extinct species), Present (endangered species), and Future (speculative creatures), situating the interaction within a temporal and ecological narrative. Through continuous audiovisual feedback and physical interaction, Diffusion TV foregrounds the generative process over final outputs, allowing audiences to explore intermediate states as experiential material. Rather than providing explicit technical explanation, the work presents an alternative, embodied mode of explainable AI that invites exploratory engagement with and reflection on generative technologies.
Read the full abstract
Machine Learning
RegionFed: Federated Learning for Personalized Query Understanding in Heterogeneous Retail Environments
Retail search systems serve diverse geographic regions with distinct query patterns, vocabularies, and product preferences, creating significant data heterogeneity that challenges both privacy-preserving training and model personalization. Federated learning offers a natural solution for privacy, but standard FL methods produce global models that sacrifice regional performance, while existing personalized FL approaches operate at the parameter level and catastrophically collapse on modern transformers (below 10% accuracy on T5) due to tied embeddings and LayerNorm interactions. We introduce RegionFed, an architecture-robust federated learning framework that sidesteps this failure by operating entirely at the gradient level. RegionFed uses the $\ell_2$ conflict between regional and global gradients as a unified signal that (i) diagnoses heterogeneity, (ii) routes each region to the cheapest sufficient personalization strategy, and (iii) adaptively controls personalization strength. Because it treats models as differentiable black boxes, RegionFed deploys on T5-Small, T5-3B, RoBERTa, and CNN with zero code changes, providing large gains on transformers (where parameter-level methods collapse) and consistent improvements on CNNs. Across three public datasets (Amazon ESCI, Amazon Reviews, LEAF-FEMNIST) and four architectures, RegionFed-Meta achieves 92.27%, closing the gap to the privacy-violating centralized upper bound (Centralized + Regional Weighting: 92.04%, $\Delta$=0.23pp, within 1$\sigma$) while providing $(\epsilon{\approx}0.60)$-differential privacy and $\mathcal{O}(1/\sqrt{T})$ convergence.
Read the full abstract
Quantum Physics
Squeezed light and coherent feedback enhanced multiparameter quantum estimation in a double-cavity optomechanics: steady and dynamical states
Multiparameter quantum estimation in open optomechanical systems is fundamentally constrained by dissipation, thermal fluctuations, and measurement incompatibility. In this work, we investigate a coupled cavity optomechanical platform in which two mechanical modes interact with driven optical cavities, two mode squeezed vacuum, intracavity degenerate parametric amplification, and coherent optical feedback. Using the continuous variable Gaussian state formalism, we derive the linearized quantum Langevin dynamics and steady state covariance matrix and evaluate the quantum Fisher information matrices associated with simultaneous estimation of the optomechanical coupling strength and cavity dissipation rate. We characterize the precision bounds using the symmetric and right logarithmic derivative formalisms and employ $\mathcal{B}_{\rm MI}=\max\{\mathcal{B}_S,\mathcal{B}_R\}$ as a comparative figure of merit within the SLD/RLD framework. We find that parametric amplification can substantially reduce the most informative bound, $\mathcal{B}_{\rm MI}$, thereby enhancing multiparameter estimation precision over a broad range of operating conditions. Coherent feedback can further reduce $\mathcal{B}_{\rm MI}$ and improve the estimation precision, particularly in the strong feedback regime. Moreover, the simultaneous presence of coherent feedback, parametric amplification, and squeezed light injection provides a combined mechanism for controlling quantum fluctuations and parameter dependent correlations, leading to enhanced and more robust multiparameter estimation. We further analyze the dynamical and steady state regimes to identify the parameter regions in which these quantum resources provide the largest metrological gain. Our results establish coupled cavity optomechanics as a flexible platform for quantum-noise engineering in multiparameter sensing.
Read the full abstract
Robotics
Same Trajectory, Contradictory Rewards (ROBORMBENCH): Paraphrase Fragility in Vision Language Reward Models
Vision-language models are increasingly used as reward functions for robotic learning, but this role requires paraphrase invariance: the same trajectory should receive the same reward under semantically equivalent goal descriptions. We show that current VLM reward models often violate this property. Paraphrasing the instruction alone can substantially change predicted progress scores, and can even flip identical robot behavior between failure and success. To measure this failure mode, we introduce ROBORMBENCH, a benchmark with 2,390 real-robot trajectories, ground-truth progress labels, and 21,673 verified paraphrases spanning lexical, syntactic, and action-goal rewrites. Across proprietary and open-source VLMs, paraphrase-induced instability is widespread and severe, grows under more divergent rewrites, and is not reliably reduced by scale or explicit reasoning. Dedicated reward models trained with trajectory-grounded supervision are substantially more stable. These results show that paraphrase robustness is a core requirement for reliable VLM-based reward modeling in robotics.
Read the full abstract
Computer Vision
A Generalizable Feature Extractor for Alzheimer's-Related Brain MRI Tasks
When there is not enough labeled data to properly train deep learning models, transfer learning can help. We still do not fully understand how effective it is in neuroimaging, especially for Alzheimer's disease research. It is also not clear if these transferred models can work on new datasets without being retrained for each specific task. We evaluate whether a compact, supervised pretrained model can serve as a reusable foundation model for downstream neuroimaging tasks. We freeze the 7.18 million weights of a 3D CNN previously trained for brain-age prediction, and adapt it to each task using Low-Rank Adaptation (LoRA), requiring only ~1% additional trainable parameters. We evaluate generalizability in six experiments. Adapting the model to classify cognitively normal versus Dementia on ADNI gave an AUC of 0.964 on held-out folds (Experiment #1). Applying that adapted model unchanged to OASIS-3, with no retraining, gave an AUC of 0.871 (Experiment #2). Reusing its output logit together with age and a cognitive score distinguished stable from progressing MCI with an AUC of 0.828 (Experiment #3). Adapting the same backbone to predict amyloid positivity from structural MRI gave an AUC of 0.804 (Experiment #4). Finally, the same approach estimated ICV-normalized hippocampal and white matter hypointensity volumes directly from the T1w image, with R^2 of 0.80 and 0.91 respectively, tasks normally addressed with much larger U-Net networks (Experiments #5 and #6). A compact model supervised on brain age can therefore serve as a reusable backbone, adapting to each task with ~1% additional parameters and transferring to an unseen cohort without any training. Our findings suggest that a carefully trained brain age model can serve as an effective foundation model for Alzheimer's related tasks, even under strict data constraints.
Read the full abstract
Computer Vision
From Interpretability Methods to Interpretable Models
More than a decade in, explainable AI (XAI) for computer vision has assembled a mature toolbox: attribution, feature visualization, concept-based, and circuit-based methods. Yet almost all of the field's effort has gone into building and comparing these methods, and little into the question they were meant to answer—how interpretable are our models, and are we making progress as they evolve? We argue for shifting the field's focus from methods to models, along two complementary lines. One is already within reach: existing tools let us characterize and compare what different models represent and compute. The other is harder, and largely neglected: whether a model can actually be understood by the humans who rely on it—the independent evaluators on whom trust and certification depend, not the experts confirming what they already expect. It can only be measured, not inferred. We review why the toolbox is mature enough to support both, survey the thin body of work comparing models, draw a parallel to systems neuroscience, and close with a model-centric XAI agenda.
Read the full abstract
Solar and Stellar Astrophysics
Metallicity dependence of Wolf-Rayet binaries using detailed binary models: An absence of long-period systems at low metallicity
Observations of Wolf-Rayet (WR) stars in binaries in the SMC and LMC suggest a preference for short period ($P\lesssim30$ days) orbits, with an apparent absence of WRs in long-period systems. The Galactic population does extend to longer periods but shows a deficit in long-period WR binaries compared to their progenitors, the O-star population. Across these populations a nearly constant binary fraction has been observed. We aim to characterize the population of WR binaries across a range of metallicity and determine the metallicity dependence of the formation of these systems, specifically focusing on the period distribution. We use detailed binary evolution models from POSYDON to predict the population of WR binaries at 0.01, 0.1, 0.2, 0.45, and $1Z_\odot$, analyzing the resulting period distribution and formation channels, while comparing them against the observed population in the SMC, LMC, and Milky Way. We find that the absence of wider WR binaries in the SMC, and potentially in the LMC, can be explained by stable, Case-B mass transfer only partially stripping the donor star, combined with WR winds at low metallicity being insufficiently strong to strip the remaining envelope. As a result, the long-period peak from Case-B mass transfer, present at Zsun, disappears at low metallicity. We additionally find that stable mass transfer (SMT) produces a short-period peak through Case-A mass transfer (P~5-10 days) that closely matches observations across metallicity. Furthermore, SMT and non-interacting systems are the dominant formation channels of WR binaries at all metallicities, with their relative contribution showing no metallicity dependence. This result implies that the SMT channel has the same metallicity-dependence as isolated WR formation. At the same time, we find that common envelope evolution primarily produces short-period ($P<1$ day) WR binaries with black-hole companions.
Read the full abstract
Computer Vision
CrossDepth: Geometry-Constrained Attention for Generalizable Multi-View Surround Depth Estimation
Reliable 3D understanding of the surrounding environment is a core requirement for autonomous driving. Multi-view surround camera rigs provide broad scene coverage, but the spatially adjacent images typically overlap only minimally. Consequently, the depth of most pixels must be inferred from monocular appearance cues. These cues can appear differently across images and may therefore be interpreted differently by the depth estimation model. We target two main sources of cross-image inconsistency: differences in camera intrinsics and the limited receptive field of each image. We address the former by conditioning the features on per-pixel camera-aware ray embeddings, enabling the network to account for camera-dependent variations in monocular cues. We address the latter by extending each pixel's context beyond its own image through cross-image attention constrained to geometrically plausible regions, derived from the calibrated rig setup. The model is trained in a fully self-supervised manner based on photometric consistency. Evaluations on DDAD and nuScenes show improved overall depth accuracy and cross-image depth consistency over state-of-the-art self-supervised methods under in-domain and cross-domain evaluation. Code is available at https://abualhanud.github.io/CrossDepthPage/.
Read the full abstract
Artificial Intelligence
A Deep Generative Model for Synthesizing Labeled Wireless Signals
Wireless signals with position-related labels are pivotal for both performance evaluation and model training in the realm of wireless sensing. However, acquiring real-world datasets is often challenged by significant measurement and labeling costs. Traditional methods for synthesizing labeled wireless signals typically rely on environmental models, leading to extensive hyper-parameter tuning and inadequate realism for comprehensive model training purposes. To address these limitations, we introduce a novel deep learning (DL)-based method, namely Inter-Instance Generative Adversarial Networks (IIns-GAN), to generate realistic labeled wireless signals. The generated signals are particularly adaptive to different environment scenarios and well-suited for various model training tasks, including distance estimation and environment identification. We have conducted extensive experiments on public Ultra-Wideband (UWB) datasets to evaluate the realism and utility of the generated signals. The results demonstrate that the signals generated by IIns-GAN mirror the physical characteristics of real-world measurements, and significantly contribute to the improvement of model training in diverse wireless sensing tasks.
Read the full abstract
Artificial Intelligence
Multi-Step Tool-Calling over Korean Open Public APIs: A Benchmark and a Data-Synthesis Recipe
Data-sovereignty regulations increasingly require public institutions to deploy open-source, on-premise LLM agents that chain multiple tool-calls across live government APIs. However, open-source models consistently underperform in this multi-step setting, and no existing benchmark measures the gap. We introduce the Korean Open Public API Benchmark (KOPA-Bench), comprising 145 real-world tasks. To close this gap, we present EDGE, an Execution-grounded Dynamic Graph for tool-calling data synthEsis driven by live execution. EDGE builds a graph of how each tool's output can feed another's input, keeps only the links that succeed when actually called against the live APIs, and traverses these verified links to synthesize executable multi-step trajectories. Fine-tuned via GRPO on the resulting dataset, our 9B model nearly matches the untuned 27B model from the same family, improving substantially not only on KOPA-Bench but also on the BFCL benchmark.