Papers

Advances in multimodal large language models (multimodal LLMs) allow data systems to answer analytical queries over text and images alongside relational tables, but at a significant cost for large-scale data. Existing systems reduce this cost through rewrite rules written in advance, which cannot use the logic of an individual query or the semantics of the data it processes. To address this challenge, we design, implement, and evaluate KathDB-FAO, a query evaluation system over multimodal data. KathDB-FAO decomposes a complex query into an execution plan of fine-grained steps, which we call Function-as-Operators (FAOs). To reduce execution cost, it searches for subsets of these FAOs to implement together as one function, enabling query-specific optimizations that make fewer model calls. Synthesizing these functions with a frontier model has its own cost, yet we observe that synthesized FAOs often recur across queries. To further reduce plan generation cost, KathDB-FAO uses pre-built functions and reuses functions synthesized for earlier queries. On SemBench and an additional workload, KathDB-FAO achieves quality comparable to the strongest competing system while reducing end-to-end cost per query by 76% on average.

We propose a hybrid quantum state preparation method based on the Walsh–Hadamard transform within the dense-to-sparse quantum state preparation framework. The method approximately prepares structured classical data in a quantum circuit through an indirect approach. Instead of directly preparing the dense amplitude-encoded state, the method first transforms the classical data into the Walsh domain and prepares only the largest-magnitude coefficients using a sparse state-preparation algorithm. The original state can then be approximately recovered through a parallel layer of Hadamard gates. Since this reconstruction requires no CNOT gates and has circuit depth 1, the proposed method introduces no additional CNOT or circuit-depth overhead, and its quantum preparation cost is determined by the underlying sparse state-preparation algorithm. Numerical results on representative benchmark signals demonstrate that the proposed method enables accurate state preparation for signals admitting sparse or approximately sparse representations in the Walsh domain.

This paper proposes a fuel-optimal boost-back guidance algorithm for reusable launch vehicles using successive convexification (SCvx). The guidance problem is formulated as a free-final-time optimal control problem with a terminal instantaneous impact point (IIP) constraint. This constraint depends only on the burnout position and velocity and enforces that the predicted ballistic impact point coincides with the target under a spherical-Earth, central-gravity model. By handling the coast analytically, the formulation confines trajectory discretization to the powered phase, reduces the problem size, and explicitly represents the bang-off structure. It also avoids the accumulation of discretization defects over the long coast and the need to resolve the burn-coast transition on a single-phase grid. Two terminal constraint formulations are considered: the closed-form Keplerian IIP and the eccentric-anomaly-based F&G solution. Their Jacobians are evaluated using complex-step differentiation. The algorithm is validated in a return-to-launch-site case study based on the Falcon 9 CRS-10 mission. Comparisons with a single-phase full-trajectory formulation, a closed-form guidance law incorporating the flight path angle rate, and an offline trajectory optimization benchmark assess the trade-off between computational cost and optimality. The results demonstrate near-optimal propellant consumption with reduced computational cost.

We describe our submission to the Unsupervised Speech in the Wild (UPS) Challenge at Interspeech 2026, a bidirectional Mamba-2 (BiMamba2) encoder trained with masked discrete-unit prediction following the HuBERT-style paradigm. The 47.88M-parameter model is trained on 250 hours of speech across 67 languages from the MLCommons Unsupervised People's Speech dataset, with no labeled data. The objective combines masked k-means pseudo-label prediction with language identification supervision and VICReg regularization. On official evaluation, the system achieves an Adjusted Rand Index of 0.735, exceeding four baselines on speaker clustering. Language identification macro-F1 (0.073) and character error rate (0.870) remain below supervised baselines. We analyze a local-official discrepancy in metric scale and checkpoint ranking, highlighting limitations of in-distribution diagnostics for predicting Dynabench probe outcomes.

An estimated 1 billion people worldwide live with vision impairment, yet current vision-language models (VLMs) produce descriptions too vague for safe navigation by blind and low-vision (BLV) users. Large VLMs can generate high-quality audio-description-compliant narrations but cannot run on mobile devices; small VLMs offer competitive latency but lack spatial detail, directional cues, and hazard awareness for navigational assistance. We present Smol-VL-BLV, a compact VLM for blind and low-vision users that closes this gap using a 500M decoder transformer model and two post-training mechanisms: (1) teacher-student distillation and (2) Group Relative Policy Optimization (GRPO) with a composite BLV reward targeting directional language, metric distances, and hazard detection. Because multi-stage post-training can induce catastrophic forgetting, we add a lightweight finetuning stage after the last stage GRPO finetuning to recover general descriptive quality while preserving BLV-specific spatial grounding. Our best model substantially outperforms the baseline across various benchmarks, including tasks: VQA, BLV captioning, OCR, and latency. Compared with the baseline for relative improvement, it improves the Spatial score gain of 19.3%, and the Social score gain of 14.8%. It also increases OCR-Bench by 101.5%, and raises TextVQA accuracy by 44.2%. These results show that BLV-focused post-training improves both accessibility-specific spatial grounding and general visual-text reasoning. Deployed on a mid-range Android smartphone via Mixed-Precision Quantization, the model remains approx. 450 MB and runs entirely on-device, offline and without network dependency, generating descriptions with latency dependent on host hardware capabilities. Our model, dataset, and code is publicly released at https://smol-vl-blv.github.io/Smol-VL-BLV-website/

Deep clustering is a powerful approach for discovering meaningful structures in high-dimensional data by learning a low-dimensional latent representation prior to clustering. Despite its empirical success, assessing the statistical reliability of the resulting clusters remains challenging. Testing discovered clusters on the same data induces selection bias and invalidates classical $p$-values. Selective inference (SI) provides a principled framework for correcting this bias, but existing methods focus on clustering performed directly on the observed features. In this work, we develop an SI framework for deep clustering with a fixed pretrained encoder. The key challenge is that cluster assignments are determined through a nonlinear transformation from the original data space to the latent space, resulting in a substantially more complex selection process than in conventional clustering. Our method provides a computationally tractable way to account for this process and enables valid statistical testing of differences between clusters identified in the latent space. Synthetic experiments demonstrate that the proposed method controls the Type I error rate while achieving higher power than valid but conservative baselines, and genomic applications show that it can identify significant cluster differences while appropriately accounting for selection bias. Our framework provides a principled approach to quantifying the statistical reliability of structures discovered by deep clustering.

In preparation for the commissioning of Linac2 (formerly referred to as the {PIP-II} linac) at Fermilab, we are developing high-level applications that automate Orbit Response Matrix (ORM) measurement and beam-based localization of quadrupole magnetic centers. Both procedures are designed to run in a low-amplitude mode compatible with otherwise regular operation. The ORM is measured by exciting multiple dipole correctors simultaneously at low amplitude, each at a distinct frequency, and applying a Discrete Fourier Transform to the recorded BPM signals to extract per-element transfer function with quantified noise. One application of the measured ORM is to predict the corrector coefficients that form local orbit perturbations (bumps). To center the trajectory in a quadrupole, such a bump is created around it, the quadrupole current is oscillated, and the responses of the downstream BPMs are recorded and analyzed. Repeating this at several beam positions inside the quadrupole – each set by the bump – the dependence of the BPM signal on beam position provides information about the quadrupole center. The applications are being tested at the Fermilab 400\,MeV Linac during regular production runs, and the lessons learned are fed back into their improvement.

A graph is integral if the spectrum of its adjacency matrix consists entirely of integers. We prove that every simple graph having a pendant path with at least three edges has an eigenvalue in $(1,2\cos(\pi/9)]$ and one in $[-2\cos(\pi/9),-1)$, and hence is not integral. This settles a conjecture of Braga, Del-Vecchio and Rodrigues (2021) on integral generalized sun graphs. The argument is matrix-theoretic: adjoining a terminal path on three new coordinates to an arbitrary real symmetric matrix produces the same spectral obstruction, and the positive interval above is optimal in this generality. We then disprove a second conjecture of the same authors, which asserts that the cycle of an integral generalized sun graph other than a cycle has length divisible by four. The graph obtained from a hexagon by attaching $6,6,12,6,6$ pendant vertices to five of its six vertices is integral and has $42$ vertices. We show that it is the smallest member of an infinite family governed by the Pell equation $x^{2}-2k^{2}=-7$, and we compute in closed form the characteristic polynomial of the analogous graphs on an arbitrary even cycle. Integrality within this family forces the cycle to be a square or a hexagon, and the square case yields a second infinite family governed by the Pell equation $k^{2}-2c^{2}=1$.

Reliable stellar-cluster membership is essential for deriving cluster ages, distances, reddenings, metallicities, mass functions, dynamical parameters, and Galactic-structure tracers. The Gaia era has produced a proliferation of membership and cluster-detection methods, but the resulting catalogues are not interchangeable. We review the historical development of membership assignment, compare the statistical assumptions of major methods, and explicitly examine the practical consequences of methodological choices on real Gaia data. We consider spatial, proper-motion, maximum-likelihood, photometric, UPMASK, density-based, machine-learning, Bayesian, and bootstrap membership approaches. We add three case studies: a direct comparison of \citet{CantatGaudinAnders2020}, \citet{HuntReffert2023}, and \citet{Perren2023} for common clusters; a Pleiades astrometric comparison; and a critical examination of low-trust candidates in the Unified Cluster Catalogue (UCC), including CWNU, CWWDL, and CKCWDM objects. The literature shows that disagreement is not simply a consequence of Gaia measurement errors. Different definitions of a cluster, different magnitude limits, different treatment of uncertainties, different spatial priors, and different tolerances for extended or low-density populations can lead to substantially different member lists. \citet{Perren2023} reported an average member-list overlap of roughly 75–80% with \citet{CantatGaudinAnders2020} and 70–75% with \citet{HuntReffert2023} at bright magnitudes, with the HUNT23 overlap decreasing to about 35% at $G=20$. The UCC also flags many newly reported candidates as low-trust objects when their member distribution is sparse, their literature support is weak, or they have strong duplicate/non-cluster indicators.

We present a global HQE analysis of inclusive semileptonic $\bar{B} \to X_c l \bar \nu_\ell$ moments, partial branching fractions, and $B^+$, $B^0$ lifetime observables. Our fit setup follows a new approach where perturbative scales are treated as correlated nuisance parameters, while missing higher-order power corrections are assessed through order-by-order fits up to $O(1/m_b^5)$. In the default scenario, we find $|V_{cb}| = (42.32 \pm 0.28|_{\rm fit} \pm 0.10|_{\rm h.o.})\times 10^{-3} = (42.32 \pm 0.30) \times 10^{-3}$ and $\mathcal{B}(\bar B \to X_c \ell \bar \nu_\ell) = (10.91 \pm 0.11|_{\rm fit} \pm 0.21|_{\rm h.o.})\%$. For the lifetimes we obtain $\tau_{B^+} = 1.89^{+0.04}_{-0.22}$ ps and $\tau_{B^0} = 1.74^{+0.03}_{-0.23}$ ps, where the strongly asymmetric uncertainty is dominated by the perturbative scale variation of the non-leptonic width. Our predictions are in agreement with previous predictions and the corresponding widths lie $1.1$-$1.2,\sigma$ below the measurements. For the lifetime ratio, we find $\tau_{B^+}/\tau_{B^0} = 1.082^{+0.031}_{-0.013}$ which agrees with experiment. For the full dataset of $q^2, E_\ell$ and $M_X$ moments, we obtain a poor fit quality, which we investigate using a leave-one-out study. In light of the persisting $|V_{cb}|$ puzzle, new correlated moment measurements as well as updated branching ratio measurements are highly encouraged.

We study nonpreemptive scheduling on a single machine with release dates, due dates, positive job weights, and a common processing time. The objective is to minimize total weighted tardiness. Although closely related equal-processing-time problems admit polynomial-time algorithms, the complexity of this problem has remained open in the literature since 2010. We prove that its decision version is strongly NP-complete, even when every job can meet its due date if processed immediately upon release. The reduction is from unweighted MAX-CUT and uses a quadratic number of jobs with polynomially bounded numerical data. Its main ingredient is a constructive normalization theorem that converts every sufficiently inexpensive feasible schedule into a binary choice for each graph vertex; after normalization, total weighted tardiness equals a constant minus a scaled cut value. We also give a deterministic polynomial-time phase-grid assignment algorithm for the shifted objective $\Phi=F+p\sum_jw_j$, where $F$ is total weighted tardiness. The algorithm enumerates at most $N$ release-date residues modulo $p$, solves one minimum-cost assignment problem for each residue, and returns the best phase-grid schedule. It runs in $O(N^5)$ arithmetic operations and achieves the tight ratio $3/2-1/(2N)$ for this algorithm. Because the added term $p\sum_jw_j$ is independent of how the jobs are scheduled, the shifted and original objectives have exactly the same optimal schedules. However, the approximation guarantee applies to the shifted objective; for the original objective, the analysis provides an additive bound. Thus, the paper both resolves the long-standing complexity question and provides a complementary worst-case guarantee for the phase-grid assignment algorithm.

Existing state-of-the-art static taint flow analyses for imperative programming languages can scale to large applications by using precise user-provided taint flow models of library methods. However, manually and precisely modeling a method's taint flows is tedious and potentially unsound. Furthermore, automatically modeling the method via an inter- procedural taint analysis can be inefficient. To solve this problem, we propose a guess-and-check approach: (1) an LLM agent that generates a precise taint flow model of a method and (2) a symbolic algorithm to check the soundness of the model. The algorithm deduces which taint flows must not occur in the method for the LLM's taint flow model to be sound, and uses lightweight static analyses (e.g., type system and pointer analysis) to prove these must-not-flows. When these analyses are insufficient, the algorithm deduces maximally-general callee models and recursively verifies their soundness, avoiding a full inter-procedural taint analysis in most cases. Since a more precise model requires fewer must-not-flows to be verified, the precision of the LLM's model directly determines the efficiency of our approach. We evaluate our approach on 97 LLM-generated taint flow models for methods in 6 large Go codebases and prove the models sound for 93% of the methods they cover. The proven-sound LLM-generated models are also precise, resulting in no new false-positives when proving taint flow properties.

Diffusion models have shown strong potential for kilometer-scale precipitation downscaling, but their performance in geographically unseen regions and event regimes remains insufficiently understood. Building on the wavelet diffusion model (WDM) framework, this study evaluates cross-region and cross-event generalization. Six 3 x 3 deg U.S. regions represent convective, winter, tropical, and atmospheric-river precipitation regimes. Low-resolution inputs are generated by block averaging NOAA Multi-Radar/Multi-Sensor (MRMS) composite reflectivity fields. A WDM trained only on Oklahoma (OK) samples and a WDM trained on all six regions are compared with nearest-neighbor and Bicubic interpolation. Model performance is evaluated using three metric families that measure image-domain reconstruction, spectral and distributional fidelity, and bin-wise precipitation detection. The OK-trained WDM remains competitive outside OK. Although the all-region WDM delivers the best and most consistent overall image-domain and detection performance, its gains are uneven across precipitation intensities. Bin-wise critical success index (CSI) over 5-dBZ reflectivity bins shows that WDM improvements concentrate in localized higher-reflectivity structures, which image-domain metrics partly obscure. In addition, the performance differences among samples are strongly associated with the spatial organization of the precipitation field, quantified by Moran's I as the spatial autocorrelation of each reflectivity bin. The sample-level Moran's I-CSI correlation stratified by sample intensity reaches 0.901 in all six regions, including regions unseen during training. Overall, these findings support future efforts to transfer downscaling models to regions with limited local training data and to generate globally consistent, high-resolution precipitation products.

The sterile insect technique (SIT) is a biological control method that reduces insect pest and disease vector populations through the release of sterile insects (in general only males). In this work, spatial aspects are accounted for through an explicit metapopulation model of $n$ interconnected patches. Each patch includes a non-adult stage and adult stages divided into females and males, with distinct dispersal networks for females, wild males, and sterile males. This framework allows to adopt an area-wide integrated pest management perspective. Focusing on the oriental fruit fly, Bactrocera dorsalis, we consider two types of additional control methods, with life stage and sex specificities. These correspond to entomopathogenic fungi soil treatment (EPFST), with a differentiated effect at every life stage; and to the male annihilation technique (MAT), which attracts and kills wild and sterile males. We derive a sufficient condition ensuring elimination of the wild population in all patches, considering either continuous or impulsive periodic release strategies. Under constraints such as the impossibility of releasing sterile males in some patches, we then study an optimisation problem to minimise a weighted sum of the release rates over all release patches to eliminate the wild population. For illustrative purposes, numerical simulations are used to investigate how dispersal network configurations, local conditions, and the simultaneous use of SIT with EPFST or MAT affect optimal release strategies. The main contribution of this work is to identify efficient spatial release strategies for population elimination under practical constraints, in combination with other control methods.

This manual documents version 2.0.0 of an open toolkit for fine-tuning small causal language models on fabricated and rule-governed arithmetic corpora and measuring what they take up from them. The fact domain is the 81 additions of two single-digit natural numbers, small enough to be enumerated exhaustively. The toolkit fine-tunes a model on the correct sums, on one fixed fabricated answer for every addition, and back on the correct sums of a subset of the additions; it fine-tunes copies of these models on simple rules (the sum plus a constant) and on a conditional rule (a shift that depends on the order of the addends), each paired with a control that has the same answers but no rule; and it measures every model on every candidate answer of every addition with one unchanged procedure, reporting results separately for additions seen in fine-tuning and additions held out. We describe and justify each stage of the pipeline: the confidence index (the probability of a complete answer, closed by an end marker), the single candidate set, the answer-only training loss, the lineage of fourteen measured models, the held-out split, the controls, the exclusion of additions that would count as hits by coincidence, and the exact and resampled intervals attached to every result. We then explain every figure and table a run produces and how each is read. This manuscript is a methodological and implementation reference: it documents the instrument, and it neither states nor tests hypotheses, nor reports or interprets the outcome of any specific run. Those are the subject of work that uses the toolkit. The toolkit and its pinned dependency environment are archived separately (Section 10) under a persistent identifier, to be cited as an instrument.

Galactic archaeology relies on stellar kinematics and chemical abundances to identify various stellar populations and associations. With JWST, Galactic archaeology of ancient metal-poor brown dwarfs is now possible through high-sensitivity, medium-resolution spectroscopy. We present the first metal-poor brown dwarf radial velocity (RV) survey using JWST/NIRSpec G395H spectra ($\lambda/\Delta\lambda \sim$ 3,000) for a sample of 23 LTY dwarfs across a wide range of temperatures and metallicities. We introduce a forward-modeling framework to measure robust RVs for brown dwarfs by focusing on the CO fundamental band at $\sim$4.5 $\mu$m. By comparing to sources with existing high-resolution RV measurements, we demonstrate that G395H RVs can achieve a systematic uncertainty of 5 km s$^{-1}$, sufficient to examine their Galactic space motions. We identify kinematic members of the Milky Way's thin disk, thick disk, and halo populations in this sample. We also confirm WISE J155349.98$+$693355.2 as likely associated with the Gaia-Enceladus merger remnant, and 2MASS J05325346$+$8246465 is likely associated with the Thamnos stream. Our study systematically demonstrates that JWST G395H spectroscopy can provide robust RVs in addition to detailed chemical abundances, necessary for Galactic archaeology studies. Future discoveries and characterization of brown dwarfs associated with Galactic halo substructures and streams will enable a stringent test of stellar/substellar formation and evolutionary models for diverse metallicities.

We present a survey of electron distributions measured by THEMIS within Earth's equatorial magnetosphere at $L$ values $\le$ 14. Particular attention is focused on the spatial distribution and occurrence of magnetic field-aligned and perpendicular anisotropies in electron distribution functions and how these depend on solar wind dynamic pressure and magnetospheric activity. Using 161,300 3-s resolution electron measurements from THEMIS A and D, we find that field-aligned anisotropies are most prominent at low energies (30–200 eV) and occur over nearly all $L$ and MLT, with the largest values concentrated in the dayside/postnoon plasmasphere, while perpendicular anisotropies dominate at higher energies (1–30 keV) and are largest and most probable in a narrower band from $L \sim$ 5–10 with a pronounced dayside/prenoon bias. A localized region near dawn at $4\lesssim L\lesssim 8$–9 is found to be typically devoid of field-aligned electrons over the entire 30–200~eV band. Both anisotropy types shift toward lower $L$ shell and become more probable under elevated solar wind dynamic pressure and Kp, and their spatial and energy dependence is largely complementary: perpendicular anisotropies coincide spatially with regions of quasi-parallel chorus generation, while field-aligned anisotropies coincide with oblique chorus and time domain structure (TDS) generation regions. In addition to being a useful diagnostic for electron population types, our results provide a useful basis for comparison with global distributions of wave modes which arise from these anisotropies, and offer an anisotropy-based observational framework for interpreting their dependence on magnetospheric activity.

Recently, a population of free-floating planets has been discovered in the Galaxy. This paper examines the influence of the planets trapped within a protostellar hydrogen clouds on the processes of gravitational instability in the gas. A sufficiently massive planet can become an alternative dynamic center for the contracting cloud. This occurs due to rapid mass growth from accretion onto the planet within the cloud. Owing to residual velocity and density gradients in the cloud, the planet that triggers star formation is likely to end up in a distant orbit around the newborn star. Some second- and subsequent-generation stars in the Galaxy may have formed via this mechanism.

We introduce a generalized classical long-range stochastic cellular automaton that exactly captures the entanglement dynamics and transitions of a measurement-only monitored quantum system. The corresponding quantum model consists of a one-dimensional qubit chain subject to competing single-site and long-range Bell measurements, with separations drawn from a power-law distribution. At every step of each trajectory, the state remains a tensor product of Bell pairs and unpaired qubits, with bipartite entanglement entropy fully encoded in a classical matching of the sites. This exact representation allows us to solve the dynamics analytically in several limits and to determine the steady-state scaling of the entanglement entropy. We find volume-law, fractal, and area-law regimes. Numerical results show that these regimes persist beyond the analytically solvable limit. Because Bell measurement outcomes change only the Bell-state labels and leave the matching unchanged, all subsystem entropies are outcome independent, avoiding the exponential cost of trajectory postselection often encountered in experimental studies of measurement-induced entanglement dynamics.

We discuss in detail Neutral Measures, which are measures which believably generate any real in Cantor space. We intuitively build up the classical notions of randomness, and show that neutral measures do exist, using the language of continuous semimeasures. Finally, we construct some specific randomness tests which allow us to enforce de- sirable properties on neutral measures. As an application, we show the existence of Gibbs measures for a large class of finite range Hamilto- nians on lattice models, and establish their relationship with neutral measures.

Multi-modal multitasking models have shown strong performance on remote sensing datasets. However, because these models are trained on heterogeneous data and vary across tasks, designing a unified model that performs well in captioning, visual question answering (VQA), and visual grounding remains challenging. In this work, we evaluate several models on the VRS Bench and NWPU-VHR-10 datasets. The EarthMind model demonstrates strong results in both captioning and VQA. For grounding, we propose multiple pipelines - RemoteSAM-SAM-v1, RemoteSAM-SAM-v2, and DiffuSAM - and ultimately adopt a majority-voting ensemble across EarthMind, RemoteSAM, SAM3, Falcon, RemoteSAM-SAM3-v1, RemoteSAM-SAM3-v2, and DiffuSAM predictions. Our unified, modular pipeline integrates advanced SAM variants with multimodal LLMs to jointly perform captioning, VQA, and grounding. It achieves 82% accuracy on captioning and 83.32% on VQA, with 90.94%, 52.04%, and 92.06% for binary, numeric, and semantic question types respectively. For grounding, it attains 64.94% accuracy. By combining diverse VLMs with our custom RemoteSAM-SAM3 models through ensemble majority voting, the system delivers more accurate and consistent remote-sensing understanding than task-specific approaches.

The Standard Model Higgs quartic coupling runs to zero around $10^9-10^{12}$ GeV at $2\sigma$, suggestive of an ultraviolet boundary condition $\lambda\simeq0$ at this scale. In a well-motivated class of high-scale supersymmetric (SUSY) models, the quartic is automatically set to zero at tree level at the superpartner scale. The Higgsino can naturally emerge as the lightest SUSY particle, with its two neutral Weyl components sufficiently degenerate to form an effectively Dirac fermion with an unsuppressed vector coupling to the $Z$. For such Higgsino dark matter, direct detection requires a mass above $\sim10^{11}$ GeV, leading to a coincidence with the scale independently selected by the vanishing quartic. Such a heavy Higgsino cannot freeze out with the observed relic density. However, if its mass exceeds the reheating temperature, the observed dark matter abundance can be reproduced through Boltzmann suppressed freeze-in. This scenario connects the dark matter mass to the measured Higgs mass, with a parameter space already constrained by LUX-ZEPLIN (LZ) and which will be decisively tested at DARWIN/XLZD. Finally, we consider PeV Higgsinos with Majorana gauginos of order $10^7$ GeV, permitting an inelastic Higgsino interpretation of the tentative LZ event while retaining the connection between the scalar superpartner scale and the vanishing Higgs quartic scale.

Automatic speech recognition models are audited for demographic fairness at full precision, yet the models that ship to production have been quantized, pruned, and distilled. We ask whether post-training weight compression, which alters model weights rather than the audio signal or its feature representation, redistributes error burden across demographic groups. Across the Whisper family on Fair-Speech, Common Voice 25, and AfriSpeech-200, 50% Wanda pruning of Whisper-large-v3 sharply widens the Black/AA-vs-Asian temporal-taxation differential on Fair-Speech: the absolute word-error-rate gap between the worst- and best-served groups more than doubles; at an assumed cost of five seconds of correction effort per transcription error this is a rise from 30 to 64 seconds of correction time per minute of speech. This +111% relative increase is invariant to the assumed per-error cost, survives an audio-quality control, and is only partly mitigated by beam-search decoding, which still leaves an +86% increase. At edge model size, INT4 HQQ quantization compounds catastrophic transcript loops on West African accents by factors of five to seven. Distillation, by contrast, narrows demographic gaps in 21 of 27 evaluated settings (teacher-student pair, precision, and dataset), with the exceptions concentrated on a single model pair. We cast the temporal-taxation construct of Choi and Choi (2025) as a quantitative metric, and show that single-snapshot fairness audits on full-precision models do not capture the deployment-time burden that compression places on already-marginalized speakers.

In conventional insulators, electronic charge is localized in Wannier orbitals centered on the atoms. In an obstructed atomic insulator (OAI), the Wannier orbitals sit instead in the empty voids between atoms. These displaced orbitals give rise to an analog of the topological bulk-boundary correspondence, in which a boundary mode occurs only where the lattice termination severs an orbital. However, experimentally identifying OAIs has remained challenging, because the Wannier orbital of a dispersive electronic band is not an energy eigenstate and cannot be measured directly. Here we engineer an acoustic metamaterial hosting an isolated flat band, in which the Wannier orbitals become energy eigenstates. By measuring the amplitude and phase of the acoustic pressure field, we map the orbital's $f$-like structure and locate its Wannier center in a void between lattice sites—providing the first phase-resolved image of an obstructed Wannier orbital. By altering the lattice termination, we observe the predicted boundary states. Such acoustic flat-band metamaterials offer a direct probe of single-particle ingredients such as quantum geometry, which govern correlated phenomena in electronic flat bands, including proposed routes to unconventional superconductivity.

Biological agents do not learn under conditions of unlimited computation. For humans, learning and choice are shaped by constraints on perception, attention, and working memory, which limit how much state information guides behavior and therefore bound policy complexity. Standard reinforcement learning models typically optimize reward without explicitly representing these internal costs, making them less suitable as models of biological intelligence. We derive MI-SARSA, an on-policy temporal-difference algorithm that incorporates mutual-information regularization through a learned marginal action prior and a penalty on state-specific deviations from that prior. This yields a sequential learning model in which state information is used selectively when its expected return benefit justifies the added informational cost. Critically, the same state-specific information cost that governs policy compression also generates trial-level predictions for reaction time, distinguishing MI-SARSA from most reinforcement learning models, which predict choices or returns but not latency. Empirically, MI-SARSA produces a reward-complexity tradeoff, and stronger information penalties produce simpler policies with lower control costs and faster reaction times. Under environment shift, increasing regularization reduces post-switch performance degradation but also lowers asymptotic return, revealing a robustness-capacity tradeoff. Together, these results position MI-SARSA as a model of bounded sequential learning under cognitive constraints.

The ongoing digital transformation of work, administration, health, mobility, and social interaction is profoundly reshaping everyday life, steadily shifting control from individuals to large platform providers. Although data is often labeled the "gold of the 21st century", its real value is realized through services that access, combine, and exploit it. Today, individuals have little sovereignty: life events (e.g., changing an address, insurance, job, or marital status) require fragmented, repetitive interactions across numerous systems, leaving users overwhelmed rather than empowered. We argue for a fundamental rethinking of this situation and propose a virtual, trustworthy platform, called "MyVirtualME" that acts on behalf of the human individual towards companies, administrations, and other actors. This platform centers services, data, permissions, and data usage around humans, not foreign actors, allowing genuine control and transparency. Our vision goes beyond a pure data focus: supported by trusted services ranging from basic notifications to intelligent assistants, MyVirtualME aims to reduce digital bureaucracy, improve and even automate interactions, and support informed oversight of an individual's health, financial, and administrative situation. We firmly believe that a human-centered MyVirtualME is essential for restoring personal sovereignty in our digital future.

The power-law profile of hopping and superconductivity introduces rich critical phenomena that are not possible to explore in the nearest-neighbor models. Having this in mind, we construct the Su-Schrieffer-Heeger (SSH)-Kitaev model with mono-, bi-, and tri-partite sub-lattices where superconductivity [hopping] is of power-law Kitaev [SSH] type. The long-range (LR) superconductivity can alone gap out the zero-momentum critical line while LR hopping alters the finite-momentum critical line. There exists open-ended critical line at vanishing superconducting gap on the phase diagram for all versions of the SSH-Kitaev model. Interestingly, LR superconductivity (hopping) promotes massive Dirac (Majorana zero) modes, while in a finite-size system, their emergence in open boundary conditions is caused by the bulk gap of the anti-periodic Hamiltonian in momentum space. This indicates an unconventional bulk-boundary correspondence for power-law models as compared to what is seen in the short-range models. Importantly, the thermodynamic phase boundaries are correctly reproduced by the above unconventional correspondence. Intriguingly, an even (odd) number of sub-lattices yields an integer (half-integer) winding number while the massive Dirac modes continue to survive irrespective of the sub-lattice profile as long as there exists LR pairing only.

The thermal history of the melt pool in laser powder bed fusion (LPBF) additive manufacturing processes governs the solidification microstructure and the mechanical properties of the resulting 3D-printed parts. Dual-laser systems offer additional degrees of freedom to control the cooling profile by reheating material behind the melt pool, but calibrating process parameters is challenging due to the complex physics of the process. We present an optimization-based iterative learning controller that determines optimal power, velocity, and offset settings by judiciously combining simulations and experiments: the model supplies search directions while feedback obtained from experiments on the real plant corrects for parameter errors, enabling convergence despite model inaccuracies. The approach is validated in simulation using a high-fidelity thermal model as a plant surrogate, with deliberate mismatches in absorption coefficient, latent heat treatment, and powder-bed effective conductivity between plant and model. Results show that the controller drives the plant cost down by over an order of magnitude and reaches a tight band of low-cost solutions across seeds, while model-only feedforward optimization stalls at a substantially higher plant cost despite appearing to converge on the surrogate.

Chimera states, characterized by the coexistence of coherent and incoherent dynamics in networks of coupled oscillators, are among the most intriguing collective phenomena in nonlinear systems and are strongly shaped by the underlying topology and initial conditions. Their analysis in large-scale networks remains computationally demanding and motivates the development of coarse-graining strategies that reduce network size while retaining the essential dynamical features of chimera behavior. In this Letter, we investigate whether equitable-partition-based quotient graphs can preserve chimera patterns under substantial reductions of network size. Using nonlocally coupled FitzHugh-Nagumo (FHN) oscillators, we exploit the fact that, for an equitable partition, the full-network dynamics restricted to the associated cluster-synchronous subspace coincide exactly with the quotient dynamics. We ask whether this structural exactness is sufficient to preserve chimera signatures after the network size reduction. Our results show that this is not necessarily the case: agreement between quotient and full-network chimera dynamics depends on the equitable partition, and exact quotient dynamics do not necessarily guarantee the transverse stability of the corresponding full-network trajectory. Importantly, when the cluster-synchronous solution becomes transversely unstable, the full network may depart from the exact quotient trajectory while remaining in a macroscopic chimera regime. Therefore, structural exactness and dynamical preservation are distinct requirements, and transverse stability provides a complementary criterion for assessing quotient-based coarse-graining of chimera states.

Primary bone tumors are rare but clinically aggressive neoplasms whose diagnosis from radiographs is challenged by heterogeneous morphology, subtle lesion margins, and overlapping bone structures. To address the limitations of existing single-view models, we present a dual-input, multi-task learning framework that, to our knowledge, is the first to apply bidirectional cross-modal attention between a lesion crop and the full radiograph for joint segmentation and subtype classification. Using the multi-institutional Bone Tumor X-ray Radiograph Dataset (BTXRD, n=3,746), we employ a YOLO-based detector to generate regions of interest, which are paired with full images as inputs to a dual-stream DenseNet121 architecture. Features are integrated via a novel cross-modal attention fusion strategy, refined by Hierarchical Multi-scale Feature Fusion, effectively balancing fine-grained lesion detail with global anatomical context. Evaluated on a held-out patient-level test split, the model demonstrates superior performance over single-input baselines, achieving an overall Dice Similarity Coefficient of 0.896 and a macro-averaged classification F1-score of 0.928. Notably, the system exhibits exceptional sensitivity for malignant osteosarcoma (AUC 0.999), validating the potential of dual-stream context modeling to support radiologists in accurate, early decision-making.