Computation and Language
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Thursday 5 November
LLM-based digital twins aim to predict how individuals would respond in new settings from representations of their prior responses. Existing evidence that compression of survey histories need not reduce predictive accuracy suggests that representation structure, rather than information volume alone, may be an important design choice. We study BDE, a fixed representation that organizes each respondent's survey history into Background, Decision procedure, and Evaluation. We evaluate it on Twin-2K-500 (Toubia et al., 2025) by predicting each respondent's held-out answers on 17 tasks. In a paired evaluation with 100 respondents, a fully inferred BDE profile improves overall accuracy over the raw survey transcript by 1.27 percentage points (pp) on gpt-5.4-nano (p < 0.01) and 1.19 pp on gpt-5.4-mini (p < 0.001). Three analyses then examine the schema, extraction depth, and component composition. Against a length-matched free-form profile written from the same transcript by the same extractor, BDE improves pricing accuracy on both simulators (2.15 pp on nano, p < 0.01; 1.38 pp on mini, p < 0.05), while its overall advantage is significant only on nano (1.05 pp, p < 0.05). Allowing deeper inference during extraction yields no statistically significant gain on either simulator. By contrast, adding Evaluation to a Background–Decision procedure profile improves overall accuracy by 1.44 pp on nano (p < 0.001) and 1.22 pp on mini (p < 0.001). These results indicate that the advantage over raw transcripts is not explained by compression or length alone, and attribute it primarily to profile composition rather than deeper extraction.
Optimization and Control
Vector Lyapunov functions for nonconvex constrained optimization
Nonconvex constrained optimization problems are commonly solved by a Newton method for necessary conditions (sequential linear/quadratic programming) or a barrier reformulation (interior-point methods). We present a vector Lyapunov analysis for Newton-type algorithms based on constraint violation and optimality of the primal solutions to a sequence of approximate optimization problems. We thus prove local stability and convergence by constructing a strict Lyapunov function for the primal solution.
Read the full abstract
Materials Science
Ion Redistribution and Lattice-Distortion-Modulated d-Electron Correlation in Layered Multiferroic ABP$_2$S$_6$ Probed by X-ray Photoelectron Spectroscopy
Ferroelectric phase transitions are driven by ion displacement-induced symmetry breaking, yet their chemical signatures are rarely probed directly at the atomic level. Here, we establish temperature-dependent X-ray photoelectron spectroscopy (XPS) as a powerful and quantitative probe of ferroelectric phase transitions by tracking ion migration-induced changes in local chemical environments. Using van der Waals (vdW) layered ABP$_2$S$_6$ materials as a model system, we systematically compare CuCrP$_2$S$_6$ (CCPS), which undergoes a ferroelectric phase transition, with AgCrP$_2$S$_6$ (ACPS), which remains paraelectric over the same temperature range. While ACPS exhibits only smooth trivial thermal broadening and shifts in core-level spectra, CCPS shows pronounced and enormously large temperature-dependent XPS anomalies across the ferroelectric transition, demonstrating that XPS is intrinsically sensitive to ferroelectric order. Detailed analysis of the Cu 2p spectra reveals that these anomalies originate from temperature-driven Cu$^+$ ion migration associated with polarization switching, whereas concurrent changes in the Cr core levels reflect secondary lattice squeezing and modulation of d-electron correlation induced by Cu displacement. Supported by a ferroelectric structural model, we further correlate XPS binding-energy shifts with the magnitude of Cu-ion migration, enabling a quantitative estimation of ionic displacement directly from spectroscopic observables. This work also highlights the rich information provided by temperature-dependent XPS for detecting and quantifying ferroelectric phase transitions and ion migration, offering a general chemical probe for ferroic materials and ion-displacement-driven phase transitions.
Read the full abstract
Medical Physics
Medical Radioisotopes: Clinical Demand, Production Physics, and Untapped Pathways
Over 40 radioisotopes are currently in use or under development across virtually all fields of modern medicine. Their availability relies on a complex interplay of nuclear physics, radiochemistry, and nuclear medicine. This review provides an isotope-by-isotope map of medically important radioisotopes from the perspective of these disparate yet complementary fields. We address the landscape grouped into three tiers: the established clinical core, including, familiar workhorses such as Tc-99m, F-18, Lu-177; an under-leveraged middle tier including Re-186, Re-188, and Sm-153, and the emerging frontier, radioisotopes with demonstrated but underutilised clinical promise such as Ac-225, At-211, Cu-67, and Sc-47. We discuss the properties of these radioisotopes with respect to clinical applications and their production pathways, as well as the associated nuclear and radiochemical challenges. We synthesise these disparate perspectives to identify situations in which alleviating production-side bottlenecks may improve access and patient outcomes.
Read the full abstract
Statistics Applications
False Discovery Rate Control via Prediction-Based, Model-Agnostic Data Splitting: An Application to Gut Microbiome Data in Liver Function
Many methods have been proposed to control the false discovery rate (FDR) in feature selection, yet challenges remain when the relationship between predictors and the outcome is complex. We propose a prediction-based data-splitting procedure that can leverage flexible predictive models while controlling the FDR. The data are divided into three parts: one for model training and two for constructing the mirror statistic. For each feature, we construct a negative control using the remaining features without incorporating outcome information and define a test statistic based on the resulting change in predictive performance. Under the null, such a perturbation is approximately equally likely to increase or decrease predictive performance, yielding a statistic symmetric about zero. In contrast, perturbing an informative feature tends to reduce predictive performance, resulting in a large positive statistic. Features are selected by thresholding these statistics, with false discoveries estimated from the corresponding negative tail to control the FDR. We apply the proposed method to liver function microbiome data, where high-dimensional, correlated features and potentially complex feature–outcome relationships make reliable type I error control particularly important for reproducible scientific discovery. Through simulation studies and the real-data application, we demonstrate that the proposed framework provides robust FDR control and can be integrated with state-of-the-art machine-learning models to accommodate complex relationships between features and the outcome.
Read the full abstract
Nuclear Experiment
Probing the initial geometry and the onset of collectivity with anisotropic flow in light-ion collisions with ALICE at the LHC
Collective behavior, traditionally associated with heavy-ion collisions, has been observed in small collision systems at the LHC. The recent oxygen-oxygen and neon-neon collisions allow to bridge the gap between small and large collision systems and to study the collectivity in well-constrained initial states. Measurements of azimuthal anisotropies are sensitive to the initial-state geometry and enable detailed investigations of the nuclear structure of $^{16}$O and $^{20}$Ne, highlighting the interplay between nuclear structure and collective dynamics in ultrarelativistic collisions. This contribution presents recent anisotropic flow measurements performed by the ALICE Collaboration in OO and Ne-Ne collisions at $\sqrt{s_{\mathrm{NN}}}$ = 5.36 TeV using two- and multiparticle correlation methods. The results are compared to hydrodynamic calculations incorporating the possible nuclear structures of these light nuclei, providing stringent constraints on the modeling of the initial state. In addition, prospects for measuring long-range correlations with the forward ALICE detectors are discussed, with the aim of studying the longitudinal dependence of anisotropic flow and thereby constraining the longitudinal structure of the initial state.
Read the full abstract
Mesoscale and Nanoscale Physics
Analytical mean-field model for electrochemical metallization memristors with a nanoporous matrix: application to Ag/anodic alumina/Al
Memristors based on electrochemical metallization with many filaments are usually modeled by microscopic simulations with many parameters or by phenomenological laws whose parameters are hard to relate to physical quantities. We present an analytical mean-field model for a population of Ag filaments confined in the pores of a nanoporous matrix, built on Ohm's law, Faraday's law and a power-based rupture criterion. The filaments have a truncated normal distribution of cross-sections and rupture leaving lognormally distributed gaps whose median grows with the dissipated power. Gap closing shifts the distribution of squared gaps rigidly, and the model therefore admits closed-form expressions for the switching voltage, the read current, the rupture branch and the high-voltage current. We apply it to Ag/nanoporous anodic alumina/Al structures fabricated from industrial-grade aluminum, with a 2.4$\mu$m alumina layer, and measured under ambient conditions. The parameters were fitted to the bipolar sweeps, with the gap dispersion taken from the temporal shape of the read after a pulse. They reproduce the bipolar response, including voltage-controlled negative differential resistance, and predict without refitting the unipolar sweeps and the amplitude of the pulse response, where state memory appears. A second sample, with a larger contact deposited chemically, is described by the same model. In vacuum the devices neither switch nor show negative differential resistance, as expected if adsorbed water supplies the counter reaction; the residual high-voltage current follows ionic hopping conduction.
Read the full abstract
Multiagent Systems
Divide, Diagnose, and Compose: Optimizing Agentic Pipelines and Harnesses
Building agentic pipelines that achieve high task performance while keeping cost and latency low requires jointly optimizing heterogeneous and coupled design choices, including prompts, topology, model selection, and tool policies. Manually searching this combinatorial design space is expensive and time-consuming, while existing automated optimizers typically operate within a predefined subset of this space, modify the pipeline monolithically, or provide little structured evidence for pipeline changes. To mitigate these challenges, we introduce the Decomposed Credit-Assignment Framework (DeCAF), which adapts the design space to the given benchmark and pipeline, optimizes each dimension separately, and generates reasoning traces for every decision. Given a task specification, benchmark examples, and a seed pipeline, DeCAF identifies relevant optimization dimensions and generates diverse experimental hypotheses for each. Independent coding-agent sessions then optimize each dimension under a restricted edit scope, jointly evaluating performance, inference cost, and latency while recording all hypotheses, edits, and outcomes of each experiment. A final synthesis session combines compatible findings across all dimensions and uses a component-level trajectory evaluator, which derives role-specific rubrics from each agent's instructions, to localize underperforming components and guide cross-dimension revisions. Across a diverse suite of agentic benchmarks, DeCAF improves average task performance by 2.58 percentage points, while reducing inference cost by 4.30% and end-to-end latency by 42.45% compared to the best overall baseline.
Read the full abstract
Systems and Control
Residential Vehicle-to-Grid Deployment: A United States-Based Stakeholder Perspective
The integration of vehicle-to-grid (V2G), in which electric vehicle (EV) owners can send electricity from their vehicles back to the grid, is an emerging solution to enhance grid resilience. However, V2G deployment in the United States lags behind many other countries. This can be attributed to a broad range of factors, and to investigate them, this study analyzes semi-structured interviews with stakeholders using inductive-deductive qualitative coding. Results from the interviews are coded across financial, technical, social, political, and organizational dimensions in order to understand the three main deployment factors of drivers, benefits, and barriers of V2G. Once coded, the relationships between these factors are used to construct a cognitive map to identify actions that will most effectively drive V2G deployment. The results indicate that grid resilience and decarbonization are seen as the main benefits of V2G. User willingness to participate, a lack of technological readiness and standardization, prohibitive costs, and unsupportive laws and policies are among the most frequently cited barriers. Connections in the cognitive map indicate that the most closely related drivers for mitigating these barriers are user compensation for battery degradation, community outreach, the creation of grid interconnection standards and protocols, and supportive grants and policy frameworks for V2G piloting and adoption.
Read the full abstract
Computers and Society
Remembering Is Not Enough: Evaluating Timing and Omission in Longitudinal AI Mental Health Support
Evaluations of large language models (LLMs) for mental health support commonly focus on the clinical appropriateness of individual responses, potentially overlooking failures that emerge across extended interactions. As AI assistants accumulate memory of users' circumstances and prior disclosures, this raises an accountability question: Do AI chatbots provide appropriate support as users' needs evolve? We investigate by analyzing 14,256 longitudinal simulations grounded in donated chatbot histories from 22 users, examining how a memory-equipped assistant responds to suicidal ideation, self-harm, and anxiety crises across recovery trajectories and communication styles. We introduce the Context-Aware and Readiness-Evaluative (CARE) framework, developed with mental health professionals, to assess the timing and quality of advice, including omissions. Advice was less aligned with users' readiness after a crisis than in matched control stretches of the same conversation, largely because of premature recommendations, which were most common when recovery stalled. Revisiting a crisis reduced premature advice but not lagging advice, and users' communication style changed how often and when the assistant gave recommendations. Although explicitly unproductive advice was rare, the assistant frequently omitted clinician-expected support such as follow-up and planning for next steps. These findings show the limits of single response-level safety evaluation and the need to assess timing, omission, and consistency across users when holding AI systems accountable for longitudinal support.
Read the full abstract
Distributed Computing
The GPU Needs an Operating System
AI chips sit mostly idle while serving models, because replies must be fast and each step waits on memory. On two large open models, the math units were busy only about 6-11% of the time. We argue the GPU needs an operating system that keeps urgent work on schedule and fills idle capacity with other useful work. A feature that ships today recovers a quarter of a GPU with work designed to share, and a tenth with real jobs, at 5% slower decode. Our research system does better: one resident kernel runs a whole model faster than vLLM's best for a single request, and shares the GPU at under 2% slowdown. Scaling beyond one hand-built model needs compilers and a few new chip features, which we specify.
Read the full abstract
Earth and Planetary Astrophysics
Coherent radio emission at 45 MHz from the nearby planet-hosting star GJ 725A
Ultra-low-frequency (<100 MHz) coherent radio emission from exoplanet-hosting stellar systems carries information about planetary and stellar magnetic fields, as well as the space weather environment - information that is difficult to obtain at other wavelengths. We carried out the circularly polarized LOFAR LBA Sky Survey, the most sensitive circularly polarized survey in this frequency range, using the LOFAR low-band antenna system. As part of this untargeted search, we detected a signal in a 1h observation that is coincident with the quiescent, planet-hosting, solar-neighborhood M dwarf GJ 725A at a distance of 3.5 pc. The signal has a high circular polarization fraction of $74\pm16$% and a peak brightness of $B=135\pm17$ mJy/beam between 42 MHz and 50 MHz. It was undetected in the 50-66 MHz range, with a 3$\sigma$ upper limit of $B<48$ mJy/beam, corresponding to a drop of at least a factor 3 across 12 MHz, and indicating a strong spectral feature or cutoff around 50 MHz. Our detection represents the lowest-frequency confirmed detection of a stellar system. Based on the high estimated brightness temperature ($\sim10^{14}$ K), transient character and polarization fraction, the emission is most plausibly attributed to the electron cyclotron maser instability. The available observations are not sufficient to identify the astrophysical origin of the emission. We consider three plausible scenarios: auroral emission of a yet undetected wide-orbit giant planet with a magnetic field $B\geq18$ G, star-planet interaction with the known 11 d orbit terrestrial planet GJ 725Ab, or an origin due to stellar magnetic activity. Ultra-low-frequency monitoring, combined with new, preferably simultaneous observations in the X-rays, UV, and H$\alpha$, and a search for a wide-orbit (thousands of days) giant planet across the full allowed parameter space will allow us to distinguish between these scenarios.
Read the full abstract
Artificial Intelligence
Memory-Aware Multi-Agent Architecture with LLM-based Orchestration for Conversational Recommendation
Conversational recommender systems often treat user sessions in isolation, limiting their ability to capture how user intent evolves over time. While large language models (LLMs) have improved semantic understanding, they are typically used as isolated components, without addressing how memory, retrieval, and decision-making should be integrated at the system level. In this work, we propose a memory-aware multi-agent architecture for conversational recommendation, in which LLMs act as orchestration layers that coordinate intent refinement, query construction, and interaction flow. A key design principle is the integration of long-term memory as a first-class component, influencing both routing decisions and semantic query formation. We evaluate the system through cross-session memory experiments, multi-turn conversational simulations, and supervisor-driven interaction studies on the Amazon Reviews dataset. Results show that memory-enabled interactions recover up to 63% of the gap toward explicit-query scenarios and prevent semantic degradation in extended dialogues. In addition, supervisory orchestration increases user acceptance by +28.9 percentage points, highlighting the importance of structured orchestration in guiding interaction and output quality. We also observe an exposure bottleneck between retrieved candidates and displayed recommendations. While LLM-reranking does not improve Hit@K, it consistently enhances semantic consistency and diversity. To capture these effects, we introduce the Semantic Consistency Score (SCS), a complementary evaluation measure. Overall, the results emphasize the importance of system-level design choices, suggesting that effective conversational recommendation benefits from explicit memory integration, orchestration, and evaluation beyond traditional ranking metrics.
Read the full abstract
Number Theory
Composite witness sieves, Kneser–Plancherel containers, and the exponent $5/7$ for the Erdős–Straus conjecture
For the Erdős–Straus equation $4/n=1/x+1/y+1/z$, Vaughan's unconditional bound on the exceptional set $E(N)$ has stood at $|E(N)|\ll N\exp(-c(\log N)^{2/3})$ since 1970. We harness composite divisor witnesses at growing depth to cross this $2/3$ barrier unconditionally. On the full exceptional set $E(N)$, we prove the unconditional bound $|E(N)|\ll N\exp(-c'(\log N)^{5/7}/(\log\log N)^{2/7})$, improving the exponent to $5/7$ (with an intermediate stage at $7/10$). On the structured component $E_2^{\mathrm{small}}(N;J(N))$, an algebraic and spectral container hierarchy combined with a generating-polynomial large sieve reaches the exponent $1-o(1)$: $|E_2^{\mathrm{small}}(N;J(N))|\ll_\kappa N\exp(-(\frac{1}{40960}+o(1))\frac{\log N}{(\log\log N)(\log\log\log N)^{\kappa}})$, matching the CRT capacity scale. These results resolve the obstructions to composite-witness sieving at growing depth. First, we replace subgroup generation by Kneser's stabilizer $K(M)$, which embeds into a $2$-primary character kernel, reducing container counts from $2^{\Theta(J^{4})}$ to $\exp(\mathcal{O}(J^{2}\log\log J))$. Coupling this with a generating-polynomial large sieve, Markov $1/2$-popular symmetry containers, and Pontryagin–Plancherel character containers yields the exponent $1-o(1)$ on $E_2^{\mathrm{small}}$. Second, to bypass unlocalized diffuse divisor sheets without subset enumeration, we construct a semi-composite congruence sieve $s=pq\equiv-1\pmod{12ua}$ ($p\in(Z^{9/10},Z]$, $q\le Q$ rough). Conditioning on $n\bmod M_{\mathrm{small}}$ yields CRT mutual independence while decoupling the modulus $12ua\le 12Z^{1/4}$ from $q$; a Bonferroni majorant and the Bombieri–Vinogradov theorem then yield the unconditional $5/7$ exponent on all of $E(N)$. Finally, we study bilinear dispersion on $4uavw-4u^{2}v-w=n$ and difference sets $D(K_c)$.
Read the full abstract
Representation Theory
Long-root $\SL_2$ dual pairs in split exceptional groups: Compact types, cone geometry, and theta correspondence
The paper establishes a case of Howe duality for an exceptional theta correspondence using the ping-pong of types method. To that end, one needs to compute lifts of types of maximal compact subgroups. A brand new ingredient enters here: The graded module of the minimal representations is the coordinate ring of an affine cone whose projectivization is a Grassmannian. Via the filtered Nakayama lemma the problem of lifting types becomes a problem in algebraic geometry: compute certain linear sections of the Grassmannian.
Read the full abstract
Emerging Technologies
Efficient Spintronic Stochastic In-Sensor Computing Modules
In-sensor processing (ISP) is used in various fields, including autonomous driving automotive systems, wearable healthcare devices, environmental monitoring, wireless sensor networks, smart cities, and industrial control systems. One of the main features of most of these applications is energy and hardware restriction. This restriction amplifies the importance of using high-performance and low-energy consumption computing systems and methods. This paper proposes efficient spintronic stochastic computing modules for in-sensor computing systems to address this challenge. The proposed design utilized the stochastic behavior of the magnetic tunnel junction in the subcritical switching regime to generate a probability-adjustable random stream. Probability-adjustable random stream generation is the most important step in any stochastic system because it determines the quality of the final output. The simulation results indicate that the proposed design offers almost the same peak signal-to-noise ratio and structural similarity while occupying at least 62% lower area and consuming 60% lower energy. These advantages made the proposed design a good choice for ISP applications with limited energy.
Read the full abstract
Machine Learning
Patching the Blind Spots: Reversible LLM Rule Overlay for Frozen Fraud Detectors
Gradient-boosted trees dominate industrial tabular fraud classification, but even a well-trained model has blind spots, and retraining a deployed model is not always safe. We bring LLM reasoning to patching such a model without modifying it: LLMs read its tree digest and SHAP attributions, reason about the fraud it misses, and propose rules anchored on features it underweights, screened and calibrated into a reversible log-lift overlay at the production cut-off. Single-LLM benefit is high-variance and inverts between datasets, so we pool three diverse backbones by logit averaging, trading peak gain for a bounded worst case. On an experimental detector trained on sampled, selection-biased real-world e-commerce data, tuned continued training regresses, whereas the overlay gains on average +2.26% recall at equal precision ($\Delta$R@P) and +1.60% precision at equal recall ($\Delta$P@R). Replayed offline on a legacy production detector, the overlay, calibrated on one month and frozen, stays more precise at equal recall for seven months (+5.64% $\Delta$P@R). On seven bias-shift variants of a public benchmark it raises $\Delta$P@R on all and keeps $\Delta$R@P near neutral. LLM reasoning over a model's diagnostics can improve a deployed tabular classifier through readable predicates naming the fraud they cover.
Read the full abstract
Distributed Computing
ResServe: Accurate and Efficient FP4 LLM Inference with Residue Compensation
Block-scaled 4-bit floating-point (FP4) quantization reduces large language model inference costs but can degrade task quality. Higher-precision computation can mitigate this loss at the expense of some of the efficiency gains from FP4 inference. Narrow matrix computations common in decoding often leave spare hardware capacity for low-cost precision recovery. ResServe combines an FP4 activation base with an FP4-quantized residue representing information lost during quantization. It uses this spare capacity to compute the residue's contribution and combines results on-chip, avoiding extra result transfers. We apply this principle to linear layers and decode attention, while selective compensation limits added work when processing input prompts. On NVIDIA B200, compensated linear kernels using the NVFP4 format improve numerical precision while retaining latency close to uncompensated execution for the evaluated operations with few rows. In the evaluated NVFP4 inference configurations, ResServe adds at most 3.4% request-latency overhead over W4A4 and reduces request latency by 22.4-53.3% relative to W4A8. Across five models, code-generation accuracy improves over same-format uncompensated inference, with mean relative gains of 15.04% for NVFP4 and 53.57% for MXFP4.
Read the full abstract
Machine Learning
An Educated Guess: Deriving Statistically Aligned Gradient Estimates for Zeroth-Order Optimisation
As novel hardware architectures are explored for the purposes of machine learning, including neuromorphic chips, asynchronous systems, and ASICs, effective learning algorithms are highly desirable. While backpropagation is the standard for differentiation on standard hardware, its requirement for global synchrony and exact differentiability limits the exploration of exotic algorithms on non-traditional substrates and devices. Zeroth-order (ZO) optimisation methods, such as SPSA and weight perturbation, offer a compelling alternative by estimating gradients through inference-only passes; however, these methods have historically suffered due to the `curse of dimensionality', where random perturbations become increasingly misaligned with the true gradient in high-dimensional spaces. In this work, we move beyond blind random noise by deriving zeroth-order search directions from activation geometry available during the forward pass. This leads to a family of Spectral Zeroth-Order methods, SZO-$\kappa$, whose directions solve local response-based gradient-guessing objectives under explicit stochastic assumptions. These methods explore feature-supported directions that are more likely to produce informative loss responses. Using multilayer and convolutional networks, we analyse gradient alignment, bias-variance structure, and training stability of these methods relative to standard perturbation baselines and backpropagation.
Read the full abstract
Computation and Language
RL-ATTACK: A Generalized Black-Box Adversarial Attack on Language Models via Reinforcement Learning
Large language models (LLMs) exhibit vulnerabilities to adversarial perturbations that can induce unintended responses. These vulnerabilities can be identified via intentional attacks and subsequently used to improve model safety. In black-box settings where gradients are inaccessible, existing methods tend to guess adversarial transformations using external knowledge, ignoring the information provided by the target model. To mitigate this, we propose a trainable black-box framework based on reinforcement learning, which learns from the target model to generate tailored adversarial transformations. We demonstrate that our approach achieves lower post-attack accuracy, greater query efficiency and better semantic preservation on both natural language inference and generative Q&A tasks. On natural language inference, it drives post-attack accuracy to 2% within 20 queries. Against generative language models, our trained attacker consistently achieves lower post-attack accuracy than the untrained attacker and wins in 4 out of 6 settings when compared to a strong optimization-based baseline. We further experiment with multiple reward designs and training settings, and we find that no design or setting is consistently dominant. Our code is available at https://github.com/taikunudel/RL-ATTACK.
Read the full abstract
Social and Information Networks
"A Space Where I Can Be Myself": Exploring Safe Visibility for Graduate Student Mothers in Mom Cafes
Graduate student mothers navigate intertwined demands of graduate study and caregiving, yet little is known about how they seek support through online communities. We examine this through Mom Cafes, locally based online parenting communities in South Korea. Drawing on focus group interviews and design workshops with 22 graduate student mothers, we identify a pattern of dual invisibility: caregiving circumstances were difficult to make legible in academic contexts, while academic circumstances were difficult to make legible in parenting contexts. Although Mom Cafes offered useful local information and emotional reassurance, concerns specific to graduate student motherhood were harder to articulate or safely disclose. Participants sought ways to make graduate student motherhood recognizable when relevant, but their designs highlighted tensions around identifiability, judgment, relational burden and accountability. We conceptualize these tensions as safe visibility and discuss infrastructures for selective recognition, low-obligation connection, and accountable pathways to practical care.
Read the full abstract
Information Theory
Levenshtein-type bounds for $(a,b)$-avoiding spherical codes
We develop a Levenshtein-type linear programming framework (for maximal $s$-codes and for minimal energy) for $(a,b)$-avoiding spherical codes, that is, codes whose inner products between distinct points avoid $(a,b)$. We introduce and investigate signed measures that are positive definite up to an appropriate degree and yield stationary auxiliary linear programming polynomials and exact quadrature formulas. We give explicit conditions for positivity of the quadrature weights and sufficient Krein-type conditions for nonnegative Gegenbauer coefficients (positive definiteness of the candidate optimal linear programming polynomials). Under these hypotheses and suitable node-location assumptions, the resulting cardinality bounds are optimal within the corresponding degree restriction and unique under natural nondegeneracy conditions. Under proper conditions on the quadrature nodes, we obtain universal lower energy bounds by utilizing Hermite interpolants as polynomial solutions of the respective linear program. Common test functions characterize optimality in the full polynomial linear programs under mild additional assumptions.
Read the full abstract
Optimization and Control
Optimal Feedback Control of the Fokker-Planck Equation for an SIRS Epidemic Model
We develop a Fokker–Planck feedback control framework for a stochastic susceptible–infected–recovered–susceptible (SIRS) epidemic model with three intervention mechanisms targeting disease transmission, loss of immunity, and recovery. The interventions are formulated as space- and time-dependent feedback controls that influence both the drift and diffusion coefficients of the underlying stochastic differential equations, allowing control of the probability distribution of epidemic states. We investigate the existence of optimal controls and derive first-order necessary optimality conditions involving a forward Fokker–Planck equation, a backward adjoint equation, and a Hamiltonian minimization condition. For numerical approximation, we combine a Chang–Cooper discretization of the diagonal drift–diffusion terms with a finite-difference treatment of cross-diffusion and a sequential quadratic Hamiltonian (SQH) algorithm for computing the optimal feedback policies. We also examine sufficient-decrease and convergence properties of the SQH iteration. Numerical experiments consider four epidemiological objectives involving cumulative infection burden, infection-threshold exceedance, terminal susceptible population, and a combination of running and terminal costs. The resulting feedback policies exhibit distinct spatial and temporal structures depending on the epidemiological objective. Monte Carlo simulations of the underlying stochastic system illustrate that the optimized interventions can substantially reduce infection growth and undesirable epidemic outcomes relative to uncontrolled dynamics. The proposed framework provides a distribution-based approach to stochastic epidemic control that integrates mathematical analysis, numerical optimization, and state-dependent intervention design.
Read the full abstract
Networking
Measuring Bounded Autonomy for an LLM Decomposition Planner on a Live 5G-Core Testbed
Large language models (LLMs) are entering network-management control loops, yet their plans can fall outside the executable action space. Whether hard deterministic bounds around such a planner have real effect, or merely restate a definition, is an empirical question. We measure it on a live 5G-core testbed, where an LLM planner decides when to decompose control-plane procedures into stateless handlers inside a bounded loop: a JSON schema constrains the output, a projection-repair validator re-checks it, a deterministic fallback guarantees liveness, and every decision is persisted. Three measurements anchor the paper. First, the two tested models, run unbounded with the allowed action sets stated in the prompt, still violated intent-handler rules in 13 of 44 standard calls. Every violation broke an intent-handler combination rule that the deployed flat schema cannot encode (schema enforcement as deployed caught 0/13); a union schema encoding the rule removes the class by construction. Second, with the bounds in place, all 33 decoded live plans were feasible without repair, and a matched-condition probe put the in-distribution unbounded rate at one violation in 20 windows. Third, exploratory: on the design set the planner matched an author-specified selection reference that a threshold policy and three a priori codings each missed. Transfer to held-out scenarios remained partial (5-6 of 10), and the planner's two-clause distillate could replace the runtime planner itself. Together, the results suggest elicitation and distillation, not runtime control, as the LLM's role.
Read the full abstract
Machine Learning
Dual-Stream LLM-Guided Reward Design for Reject-Inference-Aware Reinforcement Learning in E-Commerce Payment Fraud Detection
Production fraud-detection models learn from a censored slice of their own traffic: transactions a legacy system blocked never receive fraud labels, so models train on a biased, approved-only distribution. We extend LLM-guided reward design, the framework of Qu et al. (2025) in which an LLM evolves the reward of a two-step REINFORCE fraud-blocking policy, to attack this reject-inference problem inside the reward. A single LLM designs a dual-stream reward: stream 1 applies the standard dollar-weighted precision objective to labeled transactions; stream 2 scores legacy-blocked, unlabeled transactions by contrasting the newest gradient-boosted risk signal against the legacy scores that drove the original block. Rather than prescribe which signals matter, we expose a superset of evidence and let the LLM decide for itself which to use, guided by evaluation feedback alone; because it writes the reward as code, that choice is directly auditable. On a held-out four-month out-of-time test, the dual-stream agent beats the prior labeled-only agent at every target recall, by +4.7, +2.4 and +3.0 points of dollar-weighted precision at 80%, 85% and 90% recall, and the production baseline by +6.8 points on average. The resulting policy is architecturally identical to that of Qu et al. (2025) and adds no inference cost.
Read the full abstract
Computational Complexity
Faster Treasure Hunt via Strongly Universal Exploration with Polylogarithmic Overhead
We study deterministic exploration and delayed-activation treasure hunt in anonymous undirected port-labeled graphs with incoming-port information. From polynomial-time constructible ordinary universal exploration sequences of length \(O(n^c)\) for a fixed regular degree, we construct a single infinite strongly universal exploration sequence with cover time \(O(n^c\log^2n)\). Summable dyadic insertion schedules yield the sharper bound \(O_\eps(n^c\log n(\log\log n)^{1+\eps})\) for every fixed \(\eps>0\), together with refinements at every fixed iterated-logarithm depth. The proof combines reversible excursions, neutral guards and an arbitrary-window argument that treats both orientations simultaneously. For this recurrence, uniform linear expansion is equivalent to convergence of the normalized period sum. Exact counting identities also permit direct instruction access. With a logspace ordinary generator, table-free enumeration of decreasing scale paths computes instruction \(t\) using \(O(\log(t+2))\) working bits and time polynomial in \(t\). The earlier polynomial-slack bounds follow as corollaries. A local cubic simulation extends the treasure-hunt application to arbitrary-degree graphs, with the \(2m\)-vertex replacement size included explicitly in the bound.
Read the full abstract
Computation and Language
Contextuality and Contradictions: Exploring Real-World Identifications of AI-Generated Writing
As AI-generated content spreads into nearly every domain of human interaction, practices and discourses of AI detection, both formal and informal, have become an everyday part of how people interact with text and with each other. In this paper, we explore the real-world practices and discourses of AI identification that have emerged in response to a shifting discursive landscape where the human provenance of a text can no longer be taken for granted. We pair qualitative coding with computational analysis of lexical change, grounded in linguistic anthropological perspectives on language and communication, to investigate AI identifications among English Wikipedia editors and students. We find that these identifications draw on a diversity of evidentiary types, from discrete textual features and judgments about content to circular sense-making claims and presumptions about the author's moral character, and that they are contradictory, dynamic, and contextually dependent. In practice, what often motivates an AI identification is not the presence of any specific textual element, but pragmatic infelicity: a perceived mismatch between the elements present and those deemed appropriate within the text's particular context. Consequently, real-world identifications of AI-generated writing do not merely neutrally point out the perceived presence of "AI-like" features, but can be leveraged as a discursive resource to maintain and define sociolinguistic norms: what can be said in a given context, what is appropriate, and, ultimately, what is human. Further, the design and implementation of systems and tools that treat AI-likeness as a context-free property of text may therefore inherit, and enforce at scale, the norms of whichever contexts they were trained on.
Read the full abstract
Machine Learning
Foundation-Model Residuals as Signals of Urban Bus Irregularity
Pretrained trajectory models encode mobility patterns learned across cities, but their errors when transferred to a new city are usually treated only as performance limitations. We instead ask whether these residuals reveal systematic local conditions. Architectures such as UniTraj and TrajFM learn transferable trajectory representations, although cross-context transfer remains challenging. Using the NetMob26 bus location, ticketing, weather, network, and socioeconomic data for Niteroi, we analyze spatial and travel-time residuals of these recent trajectory foundation models. We test whether high residuals concentrate spatially, whether spatial and travel-time residuals overlap, and which urban-context variables explain their distribution. We interpret high residuals as candidate proxy signals of transit irregularity rather than direct evidence of service disruption, and analyze their contextual associations using Explainable AI (XAI). Insights into such varied environments are increasingly valuable for evaluating new urban paradigms; for instance, the 15-minute-city perspective serves as a key motivation for our approach, guiding our methods and expanding the analysis beyond standard mobility metrics.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Modified Newtonian Dynamics and galaxy clusters
First proposed by Milgrom in 1983 as an alternative to the standard Cold Dark Matter (CDM) paradigm, Modified Newtonian Dynamics (MOND) has gained renewed momentum in light of persistent non-detections of exotic dark matter particles and growing tensions facing CDM on galactic scales. While MOND perfectly reproduces galactic rotation curves, it encounters notable challenges in explaining the equilibrium of galaxy clusters. Although it reduces the overall dark matter budget in these massive systems, an unidentified mass component remains required. This paper investigates the hypothesis that this missing mass is baryonic: composed of non-radiating ordinary matter such as stellar remnants, gas too diffuse to initiate star formation, or compact, ultra-dense clouds of cold molecular gas with low volume filling factors. Using a sample of 106 clusters mapped in X-rays, emitted by the hot intra-cluster medium (ICM) composing most of the visible mass, we derive first the total mass within the Newtonian dynamics, assuming hydrostatic equilibrium and spherical symmetry. Then, with the modified Poisson equation, we reconstruct the spatial distribution of the invisible missing matter within the MOND framework. In the sample, the relaxed systems harboring cooling flows, known as cool-core clusters, have the highest peaks of dark missing mass. Its density profile reveals a pronounced central concentration bounded by a steep truncation immediately beyond a core radius of 20-30 kpc, confirming previous studies. MOND reduces the necessary dynamical mass by a factor of 4 to 10. The residual missing mass is proportional to the total gas mass, being roughly twice lower.
Read the full abstract
Machine Learning
Does a Trading Policy Need to Know What It Trades? Microstructure Invariance and Zero-Shot Transfer of Execution Policies
Execution algorithms are calibrated asset by asset, because the queues, trade rates and bid-ask spreads they face differ by orders of magnitude across assets. For reinforcement learning this is zero-shot generalisation across contexts of very different scales, and microstructure invariance proposes a normalisation fixed by theory: measured in each asset's own units of time, size and price, markets look alike. In an exact order-by-order replay of public Nasdaq data, we train policies once, on 954 stocks in 2019, to choose between waiting in the queue and crossing the spread, and evaluate them unchanged on 1,964 unseen stocks in 2020-21, 2,629 stocks in 2025-26, crypto pairs and a second exchange. First, a policy that sees a stock only through its recent trade rate, trade size, spread and price increment (tick) costs within 0.015 spreads per share (a fifth of what learning saves) of a policy whose training data included the stock, and the stock's own history does not help it; it comes within a similar margin of policies trained on the second exchange when every episode counts equally, and within a wider one on crypto. Second, in mean cost before fees, invariant units beat raw units, z-scores and asset descriptors, but by under a basis point: most of their advantage is information and a common cost scale that a learner can also take as plain inputs, and none of the representations tested extrapolates to unseen spread regimes. Third, a near-zero transfer gap to a pooled comparator is weak evidence that a representation is right: we show that such a comparator detects only a small share of what a poor representation loses, and confirm it in synthetic markets and on pendulums.