Statistics Applications
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 6 November
We conduct a retrospective exploratory pilot analysis of paired PD-L1 and Siglec-15 surface plasmon resonance (SPR) trajectories, examining retained response, dual-marker fusion, and additional early-shape information. Numerical matching links the original exports to previously quality-selected data: 69 files represent 68 distinct source column pairs from 10 cancer and 10 healthy study patients. Existing exclusions and truncations are preserved. All 20 patients contribute to development and exploration; no independent validation cohort is available. Because historical per-scan manual alignment offsets are unavailable, the numerical comparisons are conditional on separately documented candidate response-transition intervals rather than verified injection-event times. Patient-level paired comparisons, leave-pair-out internal assessment, empirical false-positive budgets, and timing sensitivity are implemented in reproducible Python code. Under candidate alignment, the dual-marker retained-response sum has an apparent AUC of 0.570, with a fixed-score bootstrap 95% interval of [0.300, 0.820]. In the nested 13-patient shape subset, retention alone and the fixed 85:15 retention-shape score both have AUC 0.619; the small paired increment changes direction across the -1,0,+1 minute shift scenarios. At a same-sample budget of two false positives among ten controls, dual-marker retention identifies four cancer cases, compared with two for PD-L1 and three for Siglec-15. Analytical bounds describe feature stability under stated measurement assumptions but do not identify the cause of weak discrimination. The study provides a traceable analysis framework and operating-point hypotheses for independent evaluation; the small development cohort does not establish diagnostic utility or a reliable fusion or shape advantage.
Machine Learning
Plug-and-Play Verifier for Decoupling Process Reward in Stabilizing Long-Horizon Agentic Reasoning
Stabilizing long-horizon reasoning requires intelligent agents to sustain useful progress across successive reasoning steps and tool interactions. Such agents are typically trained with reinforcement learning (RL), where actor-critic RL uses a learnable critic network to estimate step advantages, while critic-free methods such as GRPO eliminate critic models and rely mostly on the group variants of sparse outcome-only rewards. Both reward types leave partially useful failed trajectories either implicit or indistinguishable from entirely unproductive ones, and as reasoning horizons grow, it becomes difficult to maintain consistent reasoning trajectories. In this work, we propose a plug-and-play verifier framework for stabilizing long-horizon reasoning through explicit verification of intermediate reasoning artifacts. Specifically, the framework decouples process verification through a unified interface that extracts intermediate reasoning artifacts from agent running trajectories, evaluates them using domain-specific verifiers, and aggregates the verification scores as bounded process reward. Then, we combine our dense verified process rewards and the sparse outcome rewards through a two-stage post-training schedule. We showed that with plug-and-play verifiers, we further stabilize the reasoning steps as last-mile self-reflection and self-consistency. Typical evaluation on HotpotQA using Qwen3.5-4B achieves a superior accuracy (60.18%), outperforming the standard GRPO baseline (57.73%). Our results suggest that plug-and-play verifiers provide a modular approach to learn more stable and reliable reasoning behaviors over long horizons.
Read the full abstract
Software Engineering
Longtail-SWE: High-Quality SWE Tasks for Underrepresented Languages
Challenging repository-level benchmarks remain limited for many programming language ecosystems, where filtering scarce candidates with ambiguous descriptions or defective tests can discard valuable engineering problems. We introduce Longtail-SWE, a benchmark of 202 issue-resolution tasks across 13 languages with limited coverage in existing repository-level benchmarks: Clojure, Dart, Elixir, Erlang, F#, Haskell, Julia, Kotlin, Lua, OCaml, R, Scala, and Swift. Its construction uses a solver-judge-repairer pipeline: a solver attempts each candidate task, and a judge reviews the task together with the solver's trajectory and test results to guide revisions to tests and task descriptions. These revisions aim to correct tests that reject valid alternative implementations or accept incorrect solutions while retaining challenging tasks and preserving their intended requirements. We evaluate nine model configurations, including frontier models, and analyze task-solving performance, consistency across repeated attempts, and resource usage. We release all benchmark instances, executable environments, complete construction and evaluation tooling, and agent trajectories, together with an additional set of easier tasks to support evaluation of smaller models.
Read the full abstract
High Energy Physics (Theory)
The Gribov problem on the horizon: A short guide to the Gribov-Zwanziger framework and its refinement
Gauge fixing in Yang-Mills theories encounters a non-perturbative obstruction: distinct gauge-equivalent configurations can satisfy the same gauge condition. These are the so-called Gribov copies and their existence challenges the global validity of the Faddeev-Popov prescription. Addressing this ambiguity motivates a framework that connects the geometry of gauge-field space to infrared dynamics. This mini-review introduces the Gribov problem in Euclidean Yang-Mills theories and traces the construction of the Gribov-Zwanziger action and its refinement through the introduction of dimension-two condensates. We discuss the resulting infrared propagators, their comparison to lattice results, and the role of BRST symmetry, before surveying selected applications and open questions.
Read the full abstract
Optics
Intensity limits for phase-only holographic optical-tweezer arrays
Optical tweezers use tightly focused laser beams to trap and manipulate individual particles, enabling advances in quantum computing, drug discovery, and biological research. Scaling these technologies to large arrays requires delivering sufficient light to every trapping site while minimizing the total laser power required. Here, we investigate the fundamental limits of how efficiently a phase-only hologram can distribute light across an array of optical traps. We derive upper bounds on the minimum intensity across trapping sites, accounting for illumination, target geometry, and diffraction from spatial light modulator pixels. For arrays arranged in a single plane or across two axial planes, we show how these limits scale with the number of traps and SLM pixels, and establish conditions for focusing light in three dimensions. Numerical experiments demonstrate that optimized holograms can closely approach these theoretical limits, suggesting that further improvements in phase optimization alone offer limited power savings. Instead, our results reveal how the choice of illumination and trap geometry can improve light efficiency, providing practical guidelines for designing larger, more power-efficient optical-tweezer arrays.
Read the full abstract
Instrumentation and Methods for Astrophysics
Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact
We present a spectroscopic database and analysis of satellite streaks in Public Data Release 1 (PDR1) from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX), covering 6778 observations and $\sim$90 deg$^2$ of sky from January 2017 to July 2024. We detect 527 streaks across 492 observations, a contamination fraction of 7.3% that rises to monthly values above 20% by 2023–2024, although the streaks mask only 0.09% of the survey area. Of the 527 streaks, 475 (90.1%) are identified with a cataloged object using archival two-line elements: 258 (54%) lie in low or Earth orbit and 217 (46%) are geostationary or highly elliptical. Only 35% were launched since 2010; the rest are spent rocket bodies, debris, and older spacecraft. This reflects HETDEX's sensitivity to faint objects and its 2017–2024 observing baseline. The Hobby-Eberly Telescope's large 10\,m aperture and HETDEX's $\sim$6\,minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median $g_{\rm AB}=16.2\pm1.1$ for LEO, $16.7\pm1.1$ for GEO). However, once transit time is corrected for, instantaneous magnitudes differ widely: $g_{\rm inst}=7.4$ for LEO versus $g_{\rm inst}=13.0$ for GEO objects, a difference of 5.6\,mag, while the exposure-averaged magnitudes differ by only 0.5 mag. Normalized to 550 km, the 34 Starlink streaks give $g(550 {\rm km})=5.95$, consistent with the mean $V(550\,{\rm km})=5.92\pm0.04$ reported for VisorSat-design spacecraft. The spectra are consistent with reflected sunlight, with modest, signal-to-noise-driven differences in slope. The catalog, spectra, and code are archived at https://doi.org/10.5281/zenodo.22962479.
Read the full abstract
Theoretical Economics
When Is Mediated Communication Possible?
We study when players can truthfully share private information through a mediator. Our analysis focuses on symmetric linear-quadratic games with strategic complements and independent private values. We characterize when mediated communication is informative, meaning that it can induce an outcome different from the unique Bayes-Nash outcome without communication. A sharp dichotomy emerges. When types that prefer higher own actions benefit less from higher opponent actions, informative communication is impossible regardless of the type distribution. By contrast, when those types benefit more from higher opponent actions, informative communication is possible if and only if the relevant extreme type lies sufficiently far from the mean. In a linear-demand Bertrand oligopoly, informative communication is impossible under private costs but may arise under private demand. Finally, informative communication weakly benefits every type. But whenever information sharing can raise aggregate welfare, the maximum aggregate payoff under communication is strictly below the maximum payoff achievable without truth-telling constraints.
Read the full abstract
Quantum Algebra
A metric characterization of free unitary quantum groups
For a compact matrix quantum group $G$ with a fundamental unitary representation $u$ of dimension at least two, the reduced operator norm of the self-adjoint main character satisfies $\left \|\chi_u+\chi_u^*\right \|_{C_r(G)}\geq2\sqrt2$. Within the category of free quantum groups of Kac type, it is known that the minimum $2\sqrt2$ is attained only by free unitary quantum groups. We establish this characterization in full generality, including the non-Kac setting: this minimum is attained if and only if $G$ is isomorphic to a free unitary quantum group $U_Q^+$ as compact matrix quantum groups.
Read the full abstract
Robotics
Mobility Information Capacity of Multi-Robot Systems: Safety-Aware Maneuver Design
Multiple robots operating in shared workspaces must adjust their speed, direction, and trajectories to coordinate with one another and respond to changing tasks and disturbances. The available maneuver freedom depends on the dynamics, uncertainty, and safety margins, even for a fixed fleet size and workspace. However, robot count, workspace occupancy, and task throughput do not quantify how much the fleet can vary its motion or how reliably these variations can be observed. We develop a mobility information capacity (MIC) framework that jointly characterizes these two aspects. Mobility information is defined as the mutual information between a joint maneuver sequence and its noisy trajectory observation, and MIC maximizes this quantity under maneuver-resource constraints. For a linear Gaussian maneuver-to-observation model with feedback dynamics, we characterize the finite-horizon information benchmark and derive its stationary long-horizon spectral limit, accounting for heterogeneous mobility gains, colored uncertainty, and temporally coupled maneuver costs. We further map workspace-boundary, inter-robot separation, and actuator chance constraints into coupled maneuver-covariance restrictions and derive conditions under which covariance optimization reduces to resource allocation across mobility modes. Finite joint maneuver codebooks are then constructed through individual risk screening and noise-whitened trajectory separation, with bounds linking sequence recovery error and achieved information to the capacity benchmarks. An aerial-robot simulation study evaluates spectral convergence, safety-aware allocation, finite-codebook identification, and fleet-size scaling. The resulting framework complements occupancy and task-throughput measures by quantifying the maximum observable maneuver freedom supported by a mobility system.
Read the full abstract
Computer Vision
Uncertainty, Not Concealment: A Mechanistic Reappraisal of Sycophancy in Vision-Language Models
Vision-language models (VLMs) in medical retrieval-augmented generation must be able to say when a retrieved report contradicts the X-ray, so understanding whether they can detect such conflicts is important for safe deployment. The sycophancy mask hypothesis holds that these models internally detect conflicts but refuse to report them, concealing a detected contradiction. We introduce a three-part diagnostic that separates concealment from uncertainty: representation probing, activation steering, and two-way forced choice. Across four instruction-tuned VLMs on chest X-rays, we confirm behavioral suppression of contradiction and find that a linear probe separates matched from mismatched inputs, beyond what a simple CLIP model achieves. Steering the learned conflict direction never elicits contradiction, and forcing a binary choice does not recover one; instead, cross-modal attention to the image drops on mismatched pairs. We conclude that these models fail to contradict because the mismatch registers as hedged, gated uncertainty rather than concealed knowledge, calling for re-grounding and uncertainty surfacing. Code is available at https://github.com/mohamed-eldagla/Uncertainty-Not-Concealment/.
Read the full abstract
Optimization and Control
Sample complexity and scalability of adaptive-sampling trust-region methods for zeroth-order stochastic optimization — Part I: common random numbers meet supermartingales
We analyze adaptive sampling stochastic model-based trust-region methods that build their models in random subspaces and use common random numbers (CRN) within an iteration, and obtain, within a STORM-type analysis, the first $\tilde{\mathcal{O}}(\varepsilon^{-4})$ bound on the number of function evaluations needed to reach $\varepsilon$-stationarity, both almost surely and in expectation. The subspace model reduces the per-iteration computational cost, which allows scalability. The analysis relies on a general framework for the complexity of adaptive stochastic processes that carries the almost-sure analysis of ASTRO-DF into the STORM setting and adds expected bounds by a direct counting argument, without the renewal-reward argument of Blanchet et al. (2019). The bounds are obtained for two variants of a stochastic model-based trust-region method that differ in how the function evaluations used to measure progress are generated. The first algorithm, ASTRA-F, proceeds in a STORM-like manner: the trial point and the incumbent are re-evaluated on an independent set of replications, so that the estimates entering the acceptance test are independent of the model, as every existing STORM-type analysis requires their accuracy to be assessed conditionally on the model. The second, ASTRA-R, allows a practical approach: the trial point is evaluated on the replications already drawn to build the model, which avoids re-evaluating the incumbent, needs one adaptive sample size per iteration instead of two, and reduces the randomness entering the iteration. To our knowledge, no STORM-type analysis has been able to accommodate the latter approach as the accuracy of an estimate at a point chosen from the samples cannot be assessed conditionally on the model; we overcome this with a probabilistic chaining argument.
Read the full abstract
Quantum Physics
Dynamical Light-Matter Coupling with Protected Superconducting Oscillators
Superconducting circuits hosting long-lived bosonic modes are an excellent platform for emulating the dynamics of light and matter and exploiting them for quantum information processing. The space of accessible dynamics is greatly extended by introducing the frequency tunability of artificial atoms, whereby the interactions with the light field of a microwave cavity can be modified on demand. However, using external flux bias to control the frequency of transmons introduces significant losses to coupled cavities, reducing their single-photon lifetimes to a few tens of $\mu$s. These imperfections compromise the usefulness of the bosonic mode, which loses its ability to store and manipulate the quantum information of many-photon states. This thesis investigates the issue and demonstrates how to enhance the flux control of microwave cavities without compromising protection. Several hardware solutions were explored, including off-plane remote flux sources routed by a magnetic hose, and lithographically printed on-chip lines equipped with multiple stages of filtering. These developments equip the systems with broadband flux control and photon lifetimes of several hundred $\mu$s. They enable the real-time manipulation of the two central parameters of the Rabi model: detuning and coupling strength. This flexibility translates into operational primitives of interest, such as the calibration of nonlinearities to different interaction regimes, the activation of parametric resonances with adjustable coupling strengths, and the use of number-selective transmon-cavity transitions for qudit state preparation and control. These results lay the groundwork for exploring richer synthetic light-matter systems, controlling cavities with more complex nonlinear circuits, and engineering driven-dissipative systems with applications in quantum computation and sensing.
Read the full abstract
Game Theory
Robust Analysis of First-Price Auctions with Large Supply
This paper performs a robust analysis of welfare (i.e., the price of anarchy) in multi-unit first-price auctions with independent private values and unit-demand bidders. We employ a smoothness framework (see \citealp{ST-13}) to analyze the price of anarchy when supply is large. The main barrier to analysis in this framework is a worst-case bid distribution that establishes a lower bound on the price of anarchy of $1- \frac{1}{e}$. We show that such worst-case distributions cannot arise in equilibrium when agents' values satisfy a mild density lower bound condition. This characterization leads to novel asymptotic lower bounds. In particular, the price of anarchy for $\epsilon$-approximate equilibrium as the number of units goes to infinity is at least $1-\Theta \left(\epsilon \log \left(\frac{1}{\epsilon}\right)\right)$. Our analysis relies on a key structural property of equilibrium bid distributions: for a welfare analysis in any multi-unit first-price auction, it is without loss to analyze (approximate) equilibria with no atoms. In addition to this, we establish continuity and monotonicity properties of Bayes-Nash equilibria. We believe these results are of independent interest.
Read the full abstract
Performance
Sparse Predict and Demote: Optimal Scheduling with Minimal Size Advice
Knowledge of job sizes is crucial for optimizing the mean response time in single-server scheduling. However, predictions of job sizes are typically inexact and may be costly to obtain. Thus, a long-standing research question is: what is the minimal information about job sizes required to obtain near-optimal mean response time? We introduce a novel policy, Sparse Predict and Demote (SPD), and prove that it achieves order-optimal mean response time in heavy traffic while requiring only one-bit job-size predictions for a vanishing fraction of arrivals. SPD achieves heavy-traffic order-optimality even in situations when predictions are (i) inexact up to any fixed multiplicative error bound, and (ii) costly, i.e., the server is used to compute predictions and is thus unavailable to serve jobs while processing prediction queries. Finally, we use numerical experiments to illustrate that SPD maintains near-optimal performance beyond the heavy-traffic regime.
Read the full abstract
Distributed Computing
Continual Instruction Tuning for LLM-based Management of Continuum Resources
Large language models (LLMs) are increasingly explored for resource management across the Cloud-Edge continuum. However, embedding LLMs directly inside latency-sensitive control loops places inference latency, execution jitter, and hallucination risks on the critical path. This study argues for a different use of LLMs: a continually instruction-tuned intent compiler that runs once per management intent or declared context change, above a fast, deterministic tactical controller that handles all low-latency runtime actions. We formulate intent compilation as a continual instruction tuning problem, because management intents arrive sequentially and accumulate while infrastructure and workload contexts drift. Unlike recent LLM-based management frameworks, our design scores the model's outputs by execution, so that catastrophic forgetting can be measured in the executed objective of the underlying controller. On a continuum simulator with a 2B backbone and parameter-efficient adapters, preliminary results show that naive sequential tuning forgets earlier intents in the executed objective, that replay of stored examples nearly prevents this forgetting, and that the tuned compact compiler performs on par with simpler translators while being more robust than a grammar parser to rephrased intents. Code and data are available at https://github.com/lilanpei/continuum-intent-compiler.
Read the full abstract
General Mathematics
Calculation and Recurrence Relations of the Determinants of the Hessenberg - Toeplitz Matrices
Optics
Adaptive wavefront shaping through flexible multicore photonic lantern fibres
Holographic micro-endoscopy enables images to be funnelled all-optically through hair-thin strands of optical fibre. Future applications of this technology encompass biomedical imaging within mobile organisms, LiDAR, and industrial inspection. Yet holographic imaging through dynamically flexible optical fibres, called for in many real-world scenarios, represents an ongoing challenge. In this work we demonstrate how a bespoke fibre architecture that we term a 'photonic lantern fibre' offers a way forward. This device benefits from the negligible modal cross-talk of a polarisation-maintaining multicore fibre throughout the majority of its length. A tapered region at one end - known as a photonic lantern - guides light to a short segment of multimode optical fibre. We show how this fibre enables the projection of high-fidelity structured light to the multimode distal fibre facet via wavefront shaping, providing a level of optical control beyond that possible using conventional multicore fibres which suffer from aliasing due to their spatially separated cores. Furthermore, we demonstrate how high-contrast distal focussing can be re-established, anywhere across the output facet, after the multicore fibre is arbitrarily contorted, without access to the distal end. To achieve this we introduce two in-situ bend compensation strategies, that rely on a distally mounted guide-star or analysis of light reflected from the distal facet - offering the future potential for rapid single-shot bend calibration. Our work paves a new route towards the controlled delivery of high-dimensional structured light fields through dynamically flexible optical fibres.
Read the full abstract
Quantum Physics
Practical Ellipse Fitting for Atom Interferometry
In differential atom interferometry, it is necessary to extract the differential phase shift between two interferometers that are individually noisy. A variety of techniques can be used for this purpose, under the general label of ellipse fitting. We compare several methods for ellipse fitting, considering parameters and performance relevant to atom interferometry. Statistical, algebraic, geometric and Bayesian methods are discussed, and we describe a novel "reciprocal" geometric method that offers some benefits. Each method has advantages and disadvantages that depend on the use case required. Supplemental material includes basic Python and Matlab code for each method.
Read the full abstract
Machine Learning
A Clinical Evaluation of Conformal Prediction for Chest Radiograph Report Generation
While multimodal large language models (MLLMs) have improved automated radiology report generation, they remain prone to hallucinations, necessitating clinically meaningful risk-control methods. Conformal prediction (CP) provides statistical guarantees, yet existing CP approaches rely on uncalibrated automated extractors and fail to account for the clinical relevance of errors. We introduce clinical admissibility criteria (CLARITY), a framework for evaluating radiology reports that explicitly distinguishes clinically consequential errors from minor deviations. To eliminate reliance on automatic judges, we collected radiologist annotations of MLLM-generated reports using CLARITY. By benchmarking CP strategies and automatic evaluation metrics against these expert annotations, we demonstrate that commonly used admissibility proxies systematically fail to detect clinically relevant errors. Our findings reveal that automated admissibility judgments are largely decoupled from true clinical admissibility.
Read the full abstract
High Energy Physics (Phenomenology)
Quantum Open Effective Field Theory: anomalous decoherence
We study decoherence of a scalar field coupled to “anomalous” environments with critical behavior with the Schwinger-Keldysh including the influence functional, and the quantum master equation (QME) up to second order in the coupling. The Schwinger-Keldysh formulation yields a Langevin equation and the Lindblad (QME) includes time dependent rates allowing to study transient dynamics. We focus on three examples featuring non-Markov dynamics that cannot be described with S-matrix/Fermi's golden rule rates i) Unparticle physics: a conformal sector beyond the Standard Model at an infrared fixed point featuring correlation functions with anomalous scaling dimensions, ii) Threshold and infrared enhancement: when the mass shell of the field coincides with or is near to a kinematic threshold of the spectral density of the bath, iii) Critical environments: the bath is at a critical point, correlation functions scale with anomalous dimensions and long-wavelength fluctuations feature critical slowing down. In all cases the reduced density matrix undergoes decoherence in the energy basis of the probe, and off-diagonal matrix elements decay as $e^{-\Phi(t)}$ with $\Phi(t)>0 $ the decoherence function and $\Gamma(t) = \dot{\Phi}(t)$ the decoherence rate. In all cases we find that either transient or asymptotically $\Phi(t) \propto t^{-\mu}~;~ \mu>0$ with $\mu$, determined by the behavior of the spectral density at threshold and the anomalous scaling dimensions of the correlation functions, or $\Phi(t) \propto \ln(t)$ for the infrared case. For a critical environment there is a new dynamical scaling exponent. A probe of frequency $\omega_q$ is resonant with a critically slow quasiparticle for $q_*\propto \omega^{1/z}_q$ at which decoherence is strongly enhanced.
Read the full abstract
Optimization and Control
Costate-Conditioned Dynamic Programming for Parallel-in-Time Solution of Equality-Constrained Optimal Control Problems
Gradient based trajectory optimisation algorithms spend most of their per-iteration time on a structured Newton–KKT system. The system corresponds to an equality-constrained linear-quadratic optimal control problem and dedicated Riccati-type solvers exhibit linear complexity in the horizon length. Parallel-in-time algorithms exhibit the potential to reduce the complexity to logarithmic in the horizon by distributing several subtrajectories over multiple cores and then stitch together the results by means of an associative scan. Existing exact formulations condition each subtrajectory on its terminal state. Each subproblem then carries the reachability to that terminal state explicitly as constraint rows. The associated reachability decision however is ill-conditioned for weakly actuated systems. In this work we condition each subtrajectory on the terminal costate instead. The resulting subproblems can now be computed by a free-end Riccati recursion without reachability carrying rows. This approach effectively compensates for the ill-conditioning of weakly actuated systems but requires additional bookkeeping with Lagrangian multipliers when stitching together the various subproblems. Numerically the method reproduces a direct solution of the primal-dual Newton-KKT system to $10^{-12}$ in relative terms, on a weakly controllable chain of masses. The accurracy rivals that of a sequential recursion, whereas terminal-state conditioning loses up to six digits or fails. Operation counts on a problem with 64 subtrajectories and a horizon of 4096 stages gives a speed-up of 26 compared with a sequential solve.
Read the full abstract
Artificial Intelligence
FAIDER: Fairness-Aware Item-Level Decoding for Large Language Model Recommenders
Large language model (LLM) recommenders can produce semantically different recommendations when a user's protected attribute changes, even if the interaction history remains fixed. Controlling this disparity is challenging when model weights and full-vocabulary logits are inaccessible. We propose FAIDER (Fairness-Aware Item-level DEcoding for Recommendation), which incorporates fairness into list construction by selecting complete item candidates rather than manipulating token-level logits or repeatedly repairing finished outputs. The central idea is to separate the operational criterion for a violation from how strongly generation is constrained. Conformal-inspired empirical calibration fixes a cross-group semantic-disparity threshold, while online decoding applies either hard filtering or a soft penalty adjusted through dual-inspired feedback to target a prescribed violation rate. Separately, we establish a pathwise guarantee: under a uniform enforceability assumption, the soft decoder's empirical violation rate exceeds its target by at most $O(1/T)$. Experiments across recommendation benchmarks and LLM backbones show that Soft FAIDER improves both measured semantic fairness and ranking quality over the evaluated prompt-repair baselines. It also generally achieves a more favorable fairness-utility balance than the hard variant, indicating that stricter per-recommendation constraints do not necessarily yield lower measured disparity.
Read the full abstract
Optimization and Control
Memoryless BFGS on a nonsmooth convex function: two new results
We extend the analysis of Asl and Overton~\cite{AO2021} for memoryless BFGS (also known as L-BFGS-1) using an Armijo-Wolfe line search and applied to the nonsmooth convex function $f(x)=a|x^{(1)}|+\sum_{i=2}^n x^{(i)}$, where $n\ge2$ and $a \geq \sqrt{n-1}$. This function is unbounded below, so the method is successful if it generates function values diverging to $-\infty$, and fails otherwise. For scaled L-BFGS-1, we sharpen the sufficient condition for % az{convergence of the absolute normalized search directions} in Theorem~3.1 of failure in \cite{AO2021} from $a\ge2\sqrt{n-1}$ to $a\ge\sqrt{3(n-1)}$. For the unscaled version, we prove success for every $a$: either the method generates a direction along which no Armijo-Wolfe step exists, certifying unboundedness below, or its function values diverge to $-\infty$. Both results are consistent with the numerical experiments in \cite{AO2021}. Both proofs given here were derived with help from GPT 6 in discussions directed by the first author and corrected and rephrased afterward.
Read the full abstract
Quantum Physics
Regev's Reduction Reaches Reconstruction
Regev's reduction uses a quantum Fourier transform to carry out reductions between various problems related to lattices and codes. We apply it to the problem of sampling from Gibbs distributions of certain classical spin glass Hamiltonians, which can equivalently be thought of as cost functions for average-case instances of an optimization problem called $D$-regular max-$k$-LINSAT. This problem becomes hard at low temperature. Here, we use a quantum decoding algorithm called Belief Propagation with Quantum Messages (BPQM) to solve the decoding problem arising from Regev's reduction. Our main contribution is a proof that the minimum temperature reachable using Regev+BPQM is exactly equal to the reconstruction threshold temperature for noisy channels transmitting signals on a tree network. This equality can be understood via Renes' duality between classical and quantum channels, which we use to show that quantum decoding is possible exactly when classical reconstruction on the dual problem is impossible. Mézard and Montanari identified this reconstruction temperature with a dynamical glass transition predicted by the one-step replica-symmetry breaking (1RSB) cavity equation. It is conjectured that local samplers such as Metropolis Monte Carlo and Glauber dynamics cannot equilibrate efficiently below this reconstruction temperature but may do so above it; our large numerical experiments show close agreement with this prediction across hundreds of different problem parameters. Therefore, our results rigorously show that Regev+BPQM reaches a temperature for which, to our knowledge, there is no matching classical guarantee, and brings Regev+BPQM to the conjectured limit of efficient Gibbs sampling by classical local dynamics.
Read the full abstract
Quantum Physics
Experimental visual quantum state discrimination
Almost all solutions to vision tasks $\unicode{x2014}$ offered by nature and technology alike $\unicode{x2014}$ rely on capturing an image using a lens or a similar optical instrument, followed by computational analysis by neurons of a brain or a digital computer. This approach is inherently suboptimal in photon-limited and noisy conditions, where critical information is irreversibly lost at the measurement stage. We address this gap by introducing AI-driven optimization of quantum measurements on the incoming photons, learning how to extract the maximum possible information for a given computer vision task from each incoming photon. The trained measurement consists of a diffractive optical neural network (DONN) followed by a detector array. We train the DONN with a quantum information-theoretic objective that maximizes the rate at which the discrimination error decays as photons accumulate. The resulting measurement achieves exponentially lower error than both direct imaging and a DONN optimized for single-photon discrimination. We validate the advantage in a noiseless setting as well as a photon-counting experiment with an electron-multiplying CCD with the noise corresponding to $10^4$ photons per single signal photon, where the measurement also learns to mitigate the detector noise.
Read the full abstract
Machine Learning
Transit-Oriented Land Use Planning Using Deep Reinforcement Learning
Transit-Oriented Development (TOD) is a key strategy for advancing sustainable urban development by promoting high-density, mixed-use land development within walking distance of public transit stations. However, traditional TOD planning often relies on heuristic methods, making it difficult to efficiently identify quantitatively optimized transit-oriented land use patterns. This paper presents DRL4TOD, a deep reinforcement learning framework that automatically optimizes land use types and floor area ratios (FARs) around transit stations in alignment with TOD principles. DRL4TOD incorporates a graph neural network to capture spatial interdependencies among land parcels, road networks, and metro stations, and uses a DRL-based optimization strategy to assign land-use types and FARs to selected parcels. It jointly optimizes multiple objectives, including land use density, land use diversity, and transit capacity utilization, while minimizing reconstruction costs to account for development feasibility. Applied to two real-world station areas in Shanghai, DRL4TOD consistently outperforms rule-based approaches and genetic algorithms, achieving more effective TOD outcomes with lower reconstruction costs. By adjusting objective weights, the framework enables flexible prioritization, such as placing greater emphasis on cost minimization when financial feasibility is a concern. These results demonstrate DRL4TOD's potential as a decision-support tool for generating coordinated land-use and FAR configurations under planner-specified priorities, rather than prescribing a single fixed design. For policymakers, the framework provides parcel-level, constraint-aware alternatives that make trade-offs among competing objectives explicit and support targeted decisions about development intensity and feasibility.
Read the full abstract
Robotics
Who Taught the Robot That? Removing One User's Corrections from a Robot Policy That Learns Online
A robot that keeps learning after deployment is corrected by several people in turn, and one of them may later ask to be removed. We show that whether the robot can forget that person depends on how people correct it. The goal of removal is the robot that would exist if the person had never been there, which we compute exactly in two simulated tasks with scripted users. When users correct the robot by taking over control, an ordered log of training updates is enough: retraining on the remaining sessions in their original order gives a robot as close to the goal as training noise allows. Deleting the person's data and retraining in one pass does not, even when the other users' corrections do not depend on the removed person, because it loses the order of training. When users only nudge the robot's action, the other users' corrections carry the removed person's influence, and no method we tested recovers the goal robot. Continuing to train without the person removes most of their habits, but not what their data in the training buffer keep alive. The log can identify one clearly harmful user, but not rank users by influence better than counting their corrections.
Read the full abstract
Artificial Intelligence
Restoring General and Improving Domain-Specific Safety in Fine-Tuned LLMs Without Retraining
Safety-aligned language models are known to suffer degradation in refusal behavior following domain-specific fine-tuning, even when the fine-tuning data is entirely benign. The predominant mitigation strategy, preference optimization on the fine-tuned model, is computationally expensive and requires domain-specific preference data that is often unavailable in practice. We propose SPECTRA, a training-free, weight-level method that restores refusal behavior by transferring the base model's refusal direction into the fine-tuned model's weights via an activation-aware, low-rank update, requiring no gradient-based optimization or domain-specific safety data. The low-rank design restricts the intervention to the subspace most causally active on harmful inputs, leaving benign behavior largely undisturbed by construction. Empirically, SPECTRA substantially improves refusal behavior across a comprehensive safety evaluation, while preserving general coherence, domain-specific knowledge, and producing negligible change in false refusal rates.
Read the full abstract
Numerical Analysis
Convergence of Adaptive Least Squares Finite Element Methods
We consider a general class of operator equations $\mathcal B u=f$ where $\mathcal B:\mathbb U\to \mathbb L_2$ is a linear isomorphis from a Hilbert space $\mathbb U$ onto an “$L_2$-type” space $\mathbb L_2 = L_2(\Omega;\mathbb R^q)$. This covers weak first order formulations of stationary elliptic boundary value problems, the Stokes system or systems for linear elasticity as well as space-time formulations of heat or wave equations. Since an $L_2$-residual is uniformly equivalent to the corresponding error in the trial space norm reliable and efficient computable a posteriori error bounds are for free. In contrast to a matured theory of classical adaptive finite element methods, estblishing linear convergence - and eventually optimal complexity - in a least squares setting seems to have remained open. To a large extent this is due to the fact that least squares error indicators do not exhibit any explicit dependence on local mesh-sizes. In this paper we propose a rather different approach to establish under which circumstances standard Dörfler-marking mesh refinement gives rise to linear convergence and rate optimality of adaptive least squares finite element methods as well. A central role is played by a new notion of data-visibility which is a structural property of the data rather than a smoothness condition.
Read the full abstract
Artificial Intelligence
Adversarial Evaluation of Large Language Models for Patient-Friendly Radiology Report Summarization
Purpose: To develop and validate a scalable adversarial evaluation framework for assessing the robustness of large language models (LLMs) that generate patient-friendly summaries of radiology reports. Methods: We introduced a clinically grounded adversarial framework that uses controlled term substitutions and impression removals across intensity levels. Using a radiologist-validated adversarial dictionary, we generated adversarial variants of CT and MRI reports. Six LLMs summarized original and adversarial impressions using 2 prompt styles, yielding 18,000 evaluation pairs. Robustness was evaluated with an automated LLM-as-a-judge framework scoring factual faithfulness, sensitivity to adversarial edits, certainty fidelity, and hallucination rate. Failure mode analysis quantified vulnerabilities in incorporating clinical modifiers. Agreement with two radiologists was assessed on 105 cases. Results: GPT-5-mini achieved the highest factual faithfulness and certainty fidelity under both adversarial edits. Under term substitution, hallucinations occurred in 28.1-30.1% of GPT-5-mini summaries, compared with 35.3-36.4% for Claude Sonnet 4.5 and 41.4-41.9% for MedGemma. Impression-removal edits were handled more robustly than term substitutions. Even for the highest-scoring summarizer, substitutions of anatomical terms were not reflected in 23.7% of summaries, diagnostic certainty in 22.3%, and descriptor terms (e.g., "substantial") in 21.2%. Judge-radiologist agreement (macro-F1: 0.55-0.56) was comparable to inter-radiologist agreement (0.53). Conclusions: Adversarial evaluation exposes clinically meaningful failure modes, including missed anatomical modifiers, altered diagnostic certainty, and omitted severity descriptors.