Peer Revenue Returns subject to peer review

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Tuesday 10 November

Optimization and Control

Adaptive Multi-Fidelity Benders Decomposition for Security-constrained Transmission Expansion Planning

Energy grids worldwide rely on efficient, effective and resilient transmission networks to deliver power from generators to geographically distant destinations. The optimisation problem of designing these networks is made very difficult by the non-linearity and non-convexity of the equations which govern AC power flow. In this paper we propose the Adaptive Multi-Fidelity Benders Decomposition (AMFBD) algorithm, which is capable of solving to global optimality the Security-Constrained Transmission Expansion Planning problems with N-1 redundancy, using the non-convex AC power flow equations. We demonstrate the practical capability of this algorithm by solving security-constrained expansion problems based on the Garver 6-bus problem and the 14-bus IEEE test case. Given the large costs generally associated with energy grids, globally optimal investment decisions that can be found using AMFBD have the potential to provide significant practical benefits over suboptimal solutions produced by approximations or heuristics.
Read the full abstract

Systems and Control

An Accurate Modulation-Assisted Method for Phasor Estimation Under Decaying DC Offsets

Decaying DC (DDC) offsets severely compromise the accurate estimation of current phasors immediately following power grid fault occurrence. Consequently, conventional full-cycle discrete Fourier transform (FCDFT) algorithms experience decaying oscillatory magnitude errors, potentially causing distance relays to malfunction. This paper introduces a modulation-assisted technique designed to precisely eliminate the DDC components from fault current signals. By multiplying the fault signal with a high-frequency carrier and applying consecutive sliding Fourier projections, the proposed method effectively isolates the DDC signal while entirely rejecting higher-order harmonic interference. It utilizes a direct algebraic solution derived from a robust forward-difference operation across just three sequential projections to estimate and extract the DDC parameters. Furthermore, the algorithm seamlessly processes both single and multiple DDC components by dynamically tracking them as one equivalent first-order mode. Extensive evaluations are conducted to validate the proposed method accuracy. The proposed algorithm limits the maximum overshoot (MO) to 0.62% (0.59% for $k \geq 8$) for the evaluated DDC cases, and to 2.47% (0.95% at 20 dB for $k = 8$) under noisy conditions.
Read the full abstract

All authors Sina Mohammadi, Apurv Shukla, Junho Hong, Wencong Su

Artificial Intelligence

Darmok: Championship Cryptic Crossword Solving via Constraint Tightening and Resilient Orchestration

We present Darmok, the first automated system to reach superhuman performance on British cryptic crosswords. Unlike American-style crosswords, cryptic clues disguise an unmarked definition and compositional wordplay beneath a misleading surface reading, sparse grids leave nearly half of each word's letters unchecked, and tournament rules rank a grid with even a single wrong or blank square behind all error-free solutions regardless of speed. Competing alongside an elite human field at the 55th Times Crossword Championship in London, Darmok solved all seven championship puzzles error-free and finished faster than every human competitor in every round. It solved the Final puzzle in 6m 52s, a full 3m 13s ahead of the human champion. Today's frontier AI models can on average solve a whole puzzle in a single try with comparable accuracy and speed. However, cloud API calls still stall, time out, trigger safety refusals, or get stuck in reasoning loops. To simultaneously achieve tournament-grade accuracy, speed, and reliability, Darmok cross-checks clues across three independent models in rounds; uses verified crossing letters to unlock harder clues; and uses a whole-puzzle model and dictionary lookups to fill any remaining gaps without overwriting confirmed words. This design maintained near-zero-error performance across a five-month daily evaluation period. Under normal benchmark conditions, removing backup models, fallback steps, or elaborate prompt instructions barely changes accuracy, making the architecture appear over-engineered. Yet during both the worst days of daily operation and a severe slowdown of our primary model in the live Championship Final, the backup paths carried the solve, showing that as foundation models master the average case, the central challenge in building complex AI systems shifts to resilience in the worst case.
Read the full abstract

Systems and Control

Incremental Input-to-State Stabilization of Partially Unknown Control-Affine Systems within Prescribed Safe Sets

Incremental input-to-state stability ($\delta$-ISS) guarantees that the trajectories of a system converge towards one another, up to a gain on the difference of their inputs. Some existing learning-based designs that combine incremental stability with safety achieve only incremental input-to-state practical stability, whose non-vanishing offset can make the guarantee conservative. We consider control-affine systems with a known, full-row-rank input matrix and an unknown drift learned using Gaussian process regression. The prescribed safe set is a superlevel set of a smooth function satisfying a radial condition that ensures a diffeomorphism onto $\mathbb R^n$ and it need not contain any equilibrium of the uncontrolled system. Using this transformation, we design a smooth feedback law and derive sufficient conditions that, under the stated learning assumptions, guarantee with high probability both forward invariance of a compact subset of the safe set and $\delta$-ISS on that subset. The incremental estimate uses the distance induced by the transformation and contains no practical offset. Simulations on a Van der Pol oscillator and a Lotka-Volterra system illustrate the proposed approach.
Read the full abstract

Sound

Navigating relational agency in AI feedback musicianship: a diffractive dialogue

Musical AI systems behave in non-linear, dynamic ways that exceed the tool-oriented framings inherited from earlier instrument design. Existing frames of control and ensemble interaction do not articulate the entangled, relational agency that emerges when performers work with AI feedback instruments over time. We examine a duet practice with two AI-enabled feedback instruments, Sù and Agentier, in which feedback operates at the inner level of the AI's recurrent processes and the outer loop of musical interaction between performer, instrument, and co-performer. Our work draws on agential realism and a diffractive dialogue methodology to read our independent reflections through one another across five performances, examining how agency is enacted between performers and AI instruments. We find that control and agency are relational and shift across performances; musical textures enact feedback moments of control with performers; reaching a productive feedback environment is itself a shared exploratory act; and performers maintain their instruments through funnelling and injection, developing the skill of navigating agency along a continuous spectrum. Feedback musicianship with AI instruments demands navigation of this relational agency, displacing the leading-versus-responding binary that initially framed our reflections. This work contributes a materially-grounded vocabulary for AI feedback musicianship and a model of human-AI creative collaboration in which mutual agency is worked for through practice rather than designed in.
Read the full abstract

All authors Yichen Wang, Charles Patrick Martin, David Gunawan, David Cooper

High Energy Physics (Theory)

Bions and infrared renormalons in Kähler sigma models with $U(1)$ isotropy

We study semiclassical bion contributions to the vacuum energy of Kähler sigma models with two-dimensional target spaces on $\mathbb{R} \times S^1$ with twisted boundary conditions. This is first illustrated in the deformed $\mathbb{CP}^1$ model, where explicit real and complex bion solutions are constructed in a manner analogous to that used in its quantum-mechanical reduction [1]. Evaluating the bosonic and fermionic fluctuations at one loop, we obtain the effective action with both the coupling constant and the elongation parameter properly renormalized. Integration over the bion quasi-moduli space using Lefschetz thimbles then gives the leading nonperturbative contribution, which the real and complex bion contributions cancel at the supersymmetric point, while the first imaginary ambiguity appears when deviating from the supersymmetric point. Requiring its cancellation against the perturbative ambiguity allows us to infer the corresponding infrared-renormalon pole. This is an explicit example generalizing the resurgent structure of the $\mathbb{CP}^1$ model in [2] from one coupling to two renormalized parameters. The calculation further extends to Kähler targets of one complex dimension with a $U(1)$ symmetry. At leading dilute one-loop order, the target dependence enters through the moment-map difference and the metric normalizations at the two fixed points. The ambiguity structure and infrared scaling of the bion weight are common to these targets, while the inferred pole position depends on the target metric.
Read the full abstract

Quantum Gases

Pair localization in one-dimensional quasicrystals with spin-dependent hopping

We investigate the quantum localization of two interacting particles moving in a one-dimensional quasicrystal with unequal hopping amplitudes. Using a two-body Green's function approach, we determine the localization phase diagram of attractively and repulsively bound pairs, showing that increasing the hopping amplitude of one constituent while keeping the other fixed raises the critical quasiperiodic strength, promoting pair delocalization. The localization threshold nevertheless remains bounded above by the Aubry-André critical point of the slower constituent. Effective analytical descriptions capture pair localization in the limits of strong hopping imbalance and strong interactions. The dynamics of an initially on-site pair provides signatures of delocalization induced by unequal hopping amplitudes. In particular, the effective spreading exponent of the root-mean-square width of a pair distribution reveals a progression from localized to subdiffusive and eventually superdiffusive behavior as the hopping ratio increases. Our results establish unequal hopping amplitudes as a means of tuning the transport properties of interacting pairs in quasiperiodic lattices, with direct relevance to ultracold atoms in optical lattices.
Read the full abstract

All authors Carlos Rosas, U. A. Díaz-Reynoso, Francisco Mireles, G. A. Domínguez-Castro

Computation and Language

How Post-Training Influences Verbalized Algorithmic Computation in Language Models

Post-training improves language-model reasoning, but how it changes the use of internal information and verbalized chain of thought (CoT) remains unclear. We study successive SFT, DPO, and RLVR checkpoints of Olmo-3-7B-Instruct on multi-digit addition, using carry values as known intermediate state. Probes recover carry information from internal activations with similar accuracy in SFT and RLVR, although manipulating this information does not reliably control model outputs. In contrast, the usefulness of verbalized reasoning differs: CoTs produced by RLVR yield higher accuracy across receiving checkpoints than those produced by SFT or DPO. We also show that carry-related statements receive higher importance scores after RLVR than matched random controls. DPO shows similar sensitivity to carry-related statements but without comparable accuracy gains. These results suggest that post-training changes the usefulness of written reasoning and sensitivity to its content in meaningful ways, although greater sensitivity does not necessarily coincide with better problem performance.
Read the full abstract

Molecular Networks

Ayurvedic Rasayana Restores Lipid Homeostasis in Liver: Insights from Systems Biology Studies

MASLD and obesity characterized by hepatic lipid accumulation are known risk factors to manifest into Type 2 diabetes (T2D). Increased hepatic lipids due to interorgan cross talk leads de novo lipid synthesis which induce insulin resistance in multiple organs resulting in hyperglycemia. We evaluated two known lipid reducing Ayurvedic formulations, Arogyavardhini Rasa (AVR) and Nishamalaki Rasayana (NA), to understand their mechanistic basis. Network medicine-based analyses identified 128 phytochemicals in AVR and 182 in NA that met the predefined filtering criteria. The reported targets of both formulations showed significant network proximity to genes associated with T2D, fatty liver disease and obesity. An integrated knowledge graph linking formulations, phytochemicals, protein targets, FDA-approved drugs and disease genes enabling systems-level exploration of these relationships revealed eight targets, namely PPARA, PPARG, AKT1, SIRT1, TNF, PPARD, SERPINE1, and ACE which were common to both drugs. In the animal study, both formulations reduced high-fat-diet-associated increases in blood glucose, cholesterol, liver enzymes, macrovesicular and microvesicular hepatic steatosis. Untargeted liver metabolomics showed AVR and NA treatment restored sugar and nucleotide, pentose phosphate, bile and lipid metabolism pathways that were perturbed in response to high fat diet. Taken together, AVR and NA restored hepatic lipid homeostasis and inhibited manifestation of T2D.
Read the full abstract

All authors Harshit B Joshi, Anupama Vallazhath, Priyotosh Sil, Rahul Tiwari, Krishna Raghava Hebbar, Swati Sharma, Satish Rao B S, Basavaraj Hadapad, Areejit Samal, Manjunath B Joshi

Machine Learning

Tabular Foundation Models in Financial Predictions

This technical note introduces tabular foundation models (TFMs) to accounting research and evaluates their performance in one common benchmark task: predicting firms' next-year earnings changes. TFMs produce numeric predictions and learn in context, without post-release training or tuning, making them natural and easy to use for forecasting. The TFMs we study are pretrained on synthetic tabular data, so they avoid look-ahead bias from memorizing firms' outcomes. We compare Google Research's TabFM with a fine-tuned open-weight large language model developed in recent accounting research and with several tree-based models. TabFM is more accurate than all comparison models. It is also faster and less computationally intensive than the large language model. Several tree-based specifications produce predictions more quickly than TabFM, so choosing between them involves a trade-off between accuracy and speed. We conclude that TFMs have promise as a new tool for forecasting in accounting and finance. However, our evaluation is based on a single benchmark task, and no class of models is likely to dominate in all settings.
Read the full abstract

Computers and Society

Perspicacious: An Educational Intelligence Platform for Instructional Designers

Designing effective instruction requires instructional designers to create assessment questions aligned with clearly defined knowledge components (KCs). These KCs represent specific knowledge or skills that students must acquire. This process is often time-consuming and relies heavily on expert labor. Although AI for education research has produced a growing body of automated question-generation and KC-discovery methods, few have moved beyond research prototypes into user-friendly tools that instructional designers can readily use. To bridge the research-to-practice gap, we developed and evaluated an educational intelligence platform that unifies question generation and KC discovery behind a single interface to support instructional content development. Named "Perspicacious," our platform aims to help instructional designers "see through" complex data and algorithms to derive instructional insights. A cost model calibrated on 17 production runs revealed a cost-time trade-off: extracting KCs from a 576-question bank cost US $1.98-4.11 and took 36-158 minutes to complete, depending on the budget. A formative, partially Wizard-of-Oz study with eight instructional designers found strong perceived usability (mean SUS 78.8) and surfaced concrete requirements that we treat as prerequisites for deployment.
Read the full abstract

Machine Learning

Massive Activations Confound Sparse Autoencoders

Sparse autoencoders (SAEs) are commonly evaluated by variance explained and ablation damage. We show how massive activations undermine both evaluations, and propose a correction. Massive activations cause up to 99.9% of activation variance to lie along the first principal component (PC1). Because the PC1 component has no unique sparse decomposition, each SAE run rebuilds it from a different set of features. These features dominate importance rankings: an average of 47 of the 50 most important features are unmatched in a separate run, compared with about 27 of a random 50. A dictionary can also report 99.96% of variance explained while leaving 33% of the variance outside PC1 unexplained. We bypass PC1 by training the SAE on activations with PC1 removed, then adding PC1 to the reconstruction. In affected models, the top-50 feature sharing rate rises from 4-39.5% to 73-99.5%. At the same TopK sparsity, the resulting SAE degrades the model up to 44% less and reduces content reconstruction error by up to 33.9%.
Read the full abstract

Optimization and Control

A Variable-metric Inexact Progressive Hedging Algorithm with a Lagged Stopping Criterion for Multistage Stochastic Variational Inequalities

The progressive hedging algorithm (PHA) solves multistage stochastic variational inequalities (SVIs) by independently solving regularized scenario subproblems and enforcing nonanticipativity through primal-dual updates. Since these subproblems are often solved iteratively, practical implementations need stopping criteria that control errors while preserving convergence. We propose iVMPHA-Lag, a variable-metric inexact PHA with a lagged stopping criterion for monotone SVIs. The method allows scenario block metrics to vary across outer iterations while remaining compatible with nonanticipativity. For each scenario, an error budget combines a positive fraction of the previous budget with a progress measure from the preceding outer iteration. This gives a stopping tolerance fixed before the inner iterations begin, so each scenario can check its criterion independently without a prescribed summable tolerance sequence. With positive initial budgets and a positive inexactness parameter, we prove finite inner termination provided the inner solver produces valid approximations whose residual and enlargement errors tend to zero. Under suitable conditions, we establish global convergence of the primal-dual iterates and local $R$-linear convergence under metric subregularity at the limit point. Numerical experiments on synthetic strongly monotone box-constrained SVIs show that iVMPHA-Lag with adaptive metrics uses fewer scenario evaluations and less time on average than exact PHA and the competing inexact methods among the tested configurations in all four baseline regimes. Its average time advantage over the competing inexact methods persists across the tested problem sizes. Further comparisons show that metric adaptation improves efficiency over the fixed scalar configuration, while lagged stopping and positive budget retention reduce inner computational work.
Read the full abstract

Statistics Theory

Statistical Analysis of the Structure from Noise Phenomenon in Generalized Procrustes Problems

In cryo-electron microscopy (cryo-EM), observations are highly noisy, and scientists often use reference templates to guide matching, alignment, and averaging to reconstruct the molecular structure. However, these procedures can produce an output resembling the template even when the observations contain only noise, a phenomenon known as Einstein from noise (EfN). This phenomenon has attracted attention in statistics, where simplified observation models involving group actions have been studied to understand the resulting model bias. Characterizing the bias in the full cryo-EM pipeline remains challenging. We study template matching and averaging in the generalized orthogonal Procrustes problem, which provides an analytically tractable setting involving continuous orthogonal group action and lower-dimensional projections of ordered point clouds. We show that, for pure Gaussian noise, the mean of the aligned-and-averaged estimator admits the same left and right singular-vector bases as the template but its singular values may be distorted. This identifies the subspace structure imposed by the template even in the absence of signal. We characterize the singular values and bound their distortion, establish finite-sample operator-norm bounds, and give sufficient sample sizes for the population mean to dominate random fluctuations. In fact, this bias still exists under weak signals: when the template rank is at least the observation dimension, the change from the pure-noise mean is $O(\eta^2)$ in signal amplitude $\eta$. Our results provide an explicit characterization of model bias in a registration problem motivated by cryo-EM and show why averaging more observations can stabilize structure imposed by the template without removing the bias.
Read the full abstract

Machine Learning

Closing the BF16-FP16 Convergence Speed Gap in RL Post-Training with Stochastic Rounding

BF16 is widely used to pretrain large language models because it has the same dynamic range as FP32 and converges similarly to FP32 training. However, in reinforcement learning with verifiable rewards, where learning rates are typically low, we find that standard mixed-precision BF16 training converges more slowly than FP16 training at the same learning rate. We attribute this gap to weight sticking: when FP32 shadow weights are quantized to BF16 using round-to-nearest, small updates often leave the BF16 weights used for rollout generation and training unchanged. Consequently, several optimizer steps may be needed before updates to the shadow weights are reflected in the BF16 weights. Raising the BF16 learning rate reduces weight sticking and matches FP16's convergence speed, but it also increases the KL divergence from the reference model. We instead apply stochastic rounding when quantizing the FP32 shadow weights to BF16. This approach matches FP16's convergence speed across different models and datasets while reducing the KL divergence relative to BF16 training with a larger learning rate. Unlike FP16, it keeps BF16's wider dynamic range and needs no loss scaling.
Read the full abstract

All authors Mahdi Nazemi, Oleg Rybakov, Tijmen Blankevoort, Asit Mishra, Carlo del Mundo

Multimedia

DASWeave: Weaving Evidence across Speaking Conditions for Multimodal Depression, Anxiety, and Stress Screening

Multimodal screening for depression, anxiety, and stress in children and adolescents uses verbal and nonverbal behaviour. Most existing multimodal methods learn from a single recording or interview. Yet a participant's behaviour changes with the speaking condition, and these changes hold evidence that is invisible in a single recording. We propose DASWeave to bring this evidence into a neural network that encodes each response and then combines all responses of a participant. We train supervised trees to find which behavioural statistics matter for screening, both within each response and in how they change across conditions. We feed the tree outputs into the network both before and after this combination. We fine-tune a large language model to relate what a participant says to how they say it across conditions, and feed its representation into the network after the combination. The language model reads behaviour as text descriptions. We therefore derive additional training targets for the network from a multimodal large language model that takes in the audio-visual features directly. On the official hidden test set of the ACM Multimedia 2026 AdoDAS Grand Challenge (Track 1), DASWeave ranks second among 95 eligible teams. Its mean F1 is 27.1% higher than that of the official audio-visual baseline. It also has the highest mean AUROC among the top five teams. Ablation studies on the validation set show that the supervised trees and the fine-tuned language model contribute most to mean F1.
Read the full abstract

All authors Binyan Chen, Tong Pan, Yadong Liu, Xin Yun, Shangfei Wang

Machine Learning

Sanity Checks for Attention Surgery: Null Models and Matched Interventions for Representation-Motivated Design

Transformer architectural modifications frequently measure a statistic on internal representations, observe that it is large, remove it architecturally, and report an improvement. Interpreting that sequence requires evaluating whether the statistic exceeds its noise floor, whether it survives a geometric null, and whether the training gain requires the specific direction removed. We formulate these requirements as three reusable checks and apply them to self-attention removal, which deletes the component of an attention head's output aligned with its own value vector. Across nine multi-head pretrained models (5,408 heads, 124M to 6.9B parameters), an earlier-position causal null accounts for 48% to 63% of the head-value cosine alignment, with no monotone scale trend. Under grouped-query attention, this alignment is specific to each head's own key-value group and reverses across groups (+0.24 to +0.27 within-group, -0.11 to -0.19 across-group). Split-half reliability demonstrates that the per-head intervention effect is resolvable in GPT-2 ($r_\Delta = 0.80$) but attenuated in Pythia-160M (0.42) and Pythia-410M (0.53), whereas structural self-cosine reliability is near-perfect in all three ($r_{\text{stat}} > 0.99$); which statistic predicts its own intervention is model-dependent. In a pre-registered 3-arm by 8-seed paired factorial at $4.0 \times 10^8$ tokens per run (399,900,672 tokens), a fixed arbitrary rank-one direction does not differ from baseline (+0.0011 nats, $p = 0.197$), while self-directed removal lowers validation loss (-0.0029 nats, Holm-corrected $p = 0.0023$) and outperforms the fixed arbitrary control directly (-0.0040 nats, $p = 0.0018$). At 50.9M scale the training benefit requires the own-value direction rather than a fixed arbitrary one, but a large cosine alignment alone does not establish redundancy in frozen models.
Read the full abstract

Cosmology and Nongalactic Astrophysics

OASIS: A Fast Dynamical Halo Finding Algorithm

In recent work (García et al. 2023), we advocated for defining a dynamical halo as the collection of particles that orbit their self-generated potential. Here we introduce the Orbiting-mass ASsIgnment Scheme (OASIS), a fast energy-based algorithm for constructing dynamical halo catalogs. Using single-snapshot information, OASIS achieves $\simeq 9\%$ scatter in orbiting mass estimates relative to the SPARTA particle-tracking algorithm. As an illustration of the applications enabled by OASIS, we use the Quijote and Erebos simulation suites to show that the $z=0$ dynamical halo mass function is nearly universal with respect to cosmology, and that it is well described by a modified Press-Schechter form with an effective collapse threshold $\delta_{\rm sc} = 1.54$ at intermediate masses. However, this description fails at high masses.
Read the full abstract

All authors Bayli Hayes, Edgar M. Salazar, Eduardo Rozo, Benedikt Diemer, Calvin Osinga, John Hugon, Vadim Bernshteyn, Ze'ev Vladimir

Artificial Intelligence

Agent Benchmarks Stop Ranking Before They Run Out of Tasks

Data-science and ML-engineering agents are evaluated on benchmarks such as MLE-bench, ALE-Bench, SWE-fficiency and AlgoTune, whose tasks yield a quantity (a Kaggle metric, contest score or speedup). When one stops separating the best models, the usual remedy is a harder version. Re-scoring released runs of eight benchmarks, including SWE-bench Verified and Terminal-Bench for breadth, we find much of this saturation lies in the measurement, not the tasks. Most agent benchmarks reduce a run to one pass/fail bit, which fails in two ways. It is too coarse: neighbouring top-ten agents on SWE-bench Verified differ by less than its tests' errors, and correcting them reorders neighbours no more than deleting a random sixth of each agent's passes. It is capped at the reference: ALE-Bench's top ten, separated by contest score, tie when graded against the median human, although nearly half the runs beat that human. On fixed tasks, a higher bar only moves the problem; graded credit fixes it. We derive a clipped log-ratio score with dated anchors and cost as a separate axis. In a Terminal-Bench re-run, pass/fail-tied models differ over tenfold on the quantity the tests discard.
Read the full abstract

Computer Vision

AgenticConflictLens: From Long Traffic Videos to Verifiable Conflict Evidence

Traffic cameras continuously record urban road-user behavior, providing proactive safety evidence beyond crash reports. Yet safety-critical events - conflicts, near misses, unsafe turns, sudden braking, and crosswalk interactions - are rare and brief, buried in hours of routine footage. Existing traffic-video pipelines extract detections, trajectories, and risk scores but process footage uniformly, without adapting to an analyst's query. Vision-language methods improve scene understanding but often lack traffic-specific verification of participant-level conflicts. We introduce AgenticConflictLens, an agentic, query-conditioned framework for efficient, auditable conflict extraction from long fixed-camera traffic videos. Given a natural-language safety query, a Traffic-Aware Controller combines structured traffic memory with specialized conflict-analysis tools to choose where to search, which road users to examine, and which evidence to verify or reject. A coarse-to-fine search retrieves candidate intervals and applies detailed verification selectively. The framework produces conflict records with refined time intervals, participant pairs, semantic regions, trajectories, motion evidence, uncertainty, rejection reasons, and supporting clips. On 96 annotated hours from Bellevue and NYC DOT, AgenticConflictLens achieves the highest conflict-existence F1 and temporal IoU on both datasets: 0.84 F1/0.77 tIoU on Bellevue and 0.80 F1/0.73 tIoU on NYC DOT, while inspecting only 8.7% and 11.9% of video, respectively. These results support scalable, evidence-grounded urban traffic-safety analysis.
Read the full abstract

All authors Boshra Khalili, Sonia Moshfeghi, Jinwoo Jang, Andrew W Smyth

Information Retrieval

Exploiting Feature-invariant Scaling Laws for Efficient A/B Experimentation in Recommendation Systems

Researchers and engineers improve recommendation systems by performing A/B experiments with new model architectures, optimization methods, and data features. However, running full-scale experiments (e.g., full-model training with online data) for every feature change is prohibitively computationally intensive, while the result from a single small-scale experiment is often misleading. To address this challenge, we show that modern recommendation models exhibit feature-invariant scaling: the change of a model's performance follows a predictable scaling-law even when different feature changes are applied to the baseline model. We introduce an efficient method that exploits scaling-laws and uses statistical inference to predict (with confidence intervals) the full-scale effect of an A/B experiment in recommendation systems. This enables practitioners to efficiently filter out neutral or regressive feature changes with only small-scale experimentation. We validate our method on three distinct recommendation model families—Wukong, HSTU, and LLM-based recommendation models—across more than 40 architectural, optimization, and data changes. Our method accurately distinguishes beneficial feature changes from neutral and regressive ones while using only a fraction of the compute required for full-scale evaluation. It also identifies several changes that improve HSTU's state-of-the-art performance on MovieLens-20M. Our results demonstrate that scaling laws can be practically exploited to make compute-efficient decisions for improving industrial-scale recommendation systems.
Read the full abstract

All authors Zhiliang Chen, Sammy Nammari, Jiahao Su, Xiaoqiang Luo

Systems and Control

Data-Driven Robust Dynamic Wireless Charging Lane Planning with Tunable Profit–Risk Trade-off and Modified User Equilibrium

Dynamic wireless charging (DWC) has attracted increasing attention as a complement to conventional plug-in charging for electric vehicles (EVs). However, existing planning methods for dynamic wireless charging lanes (DWCLs) overlook realistic EV routing and charging behavior and lack explicit out-of-sample risk control under traffic-demand and renewable-energy uncertainties. To address these limitations, this paper proposes a data-driven two-stage robust planning framework that coordinates DWCL deployment with the operation of coupled traffic–power networks (CTPNs). A data-driven uncertainty set is constructed from joint historical scenarios to preserve spatio-temporal correlations without imposing parametric distributional assumptions. A modified user-equilibrium (MUE) model is proposed to capture realistic EV routing and charging behavior by incorporating state-of-charge (SoC) dynamics and endogenous charging decisions. The resulting two-stage robust model with mixed-integer recourse is reformulated as an equivalent single-stage scenario-approach problem and solved using a customized nested column-and-constraint generation algorithm. An a posteriori feasibility guarantee is established to quantify out-of-sample risk. Based on this guarantee, the scenario set is iteratively enriched to meet a prescribed risk tolerance. Case studies on a 69-node power distribution network and a 24-node traffic network demonstrate that the proposed MUE model prevents overinvestment in DWCLs, while the proposed risk-quantification procedure enables a tunable profit–risk trade-off.
Read the full abstract

All authors Xin Cui, Yujun Lin, Ziang Liu, Francis Yunhe Hou, Wei Liu

High Energy Physics (Phenomenology)

Helicity Decay to Baryons: The Cosmological Evolution of $L_\mu-L_\tau$ Gauge Fields and Baryogenesis

We study baryogenesis from the post-inflationary decay of helical U(1)$_{L_\mu-L_\tau}$ gauge fields generated during axion inflation. Assuming that Schwinger pair production is suppressed during inflation, a large helicity-to-entropy ratio $\eta_H$ can be generated while keeping the initial lepton asymmetry just after inflation negligible. After reheating, the dissipation of magnetic helicity sources a lepton asymmetry through the chiral anomaly, which is partially converted into a baryon asymmetry by electroweak sphaleron processes. We find that there are parameter spaces where the observed baryon asymmetry can be generated in both the turbulent and dissipation-dominated regimes of the U(1)$_{L_\mu-L_\tau}$ magnetic field evolution. The baryon asymmetry generation in the turbulent regime is, however, dominated by the pre-turbulent nearly frozen regime. We further find that the requirement of successful baryogenesis imposes stronger constraints on the inflationary Hubble parameter and the helicity density generated during axion inflation than the requirement for the gauge-field energy not to backreact on the inflaton dynamics.
Read the full abstract

Machine Learning

Adaptive catalyst discovery emerges from composable scientific intelligence

Catalysis underpins chemical manufacturing, yet knowledge for catalyst design remains dispersed across experimental observations, computational data and catalysis subdisciplines. To overcome this challenge, we present Composable Catalytic Intelligence (CatAI), a physics-grounded modular AI framework that learns 1.8 million structure-property relations and organizes them into 30 physicochemical encoders. This reusable intermediate layer enables task-specific models to guide catalyst discovery for targeted reactions and objectives. Across benchmark tasks, CatAI achieves a 7.8% lower task-averaged prediction error than the next-best models. In experimental validation, CatAI identified a new confined-passable pocket catalyst applicable in industrial homogeneous polymer synthesis with 9 times the activity of the commercial benchmark. In parallel, it guided autonomous closed-loop optimization of a highly active Cu-based geminal-atom photocatalyst for C-O coupling reactions surpassing the traditional Ni-based systems. This approach connects property-level data to catalytic objectives, supporting physicochemical knowledge reuse across mechanistically distinct systems through computational screening and iterative model-experiment feedback.
Read the full abstract

All authors Wentao Li, Qilong Cai, Qi Lei, Jinxing Chen, Ruixuan Chen, Jiangjie Qiu, Botian Wang, Kai Zhao, Honghao Chen, Leyi Zhao, Zemeng Wang, Yijun Li, Jun Yin, Fangfang Li, Hongxuan Liu, Yingdong Lu, Yawen Ouyang, Manu Suvarna, Wei-Ying Ma, Ya-Qin Zhang, Kostya S. Novoselov, Hao Zhou, Jiong Lu, Xiaonan Wang

Combinatorics

Covering the Triangular Lattice with Non-Overlapping Unit Disks

For $d>0$, let $\mathcal{T}_d$ be the triangular lattice with minimum distance $d$. We study for which $d$ the set $\mathcal{T}_d$ can be covered by closed unit disks with pairwise disjoint interiors, a question left open by Phetmak, Nuntasombat and Fakcharoenphol for $2-\sqrt3\le d<2/\sqrt{37}$ and on several gaps between their constructions. Writing $\mathcal{N}(d)$ for the largest number of points of $\mathcal{T}_d$ in a closed unit disk, we show that $\mathcal{T}_d$ cannot be covered whenever $\mathcal{N}(d)<4/d^2$, and we determine this set of spacings exactly as a union of fifteen intervals; combined with their constructions, this shows that coverability is not monotone in $d$. At the right endpoints $d=2/\sqrt k$ of these intervals the density bound is attained, and we show that such a spacing is coverable if and only if $k$ is the norm of an Eisenstein integer. We also give two new coverings, at $d=1/\sqrt{13}$ and $d=2/\sqrt{21}$, in which the disks form a hexagonal packing. Consequently, together with an earlier result of Okayama, Kiyomi and Uehara, the smallest coverable spacing $d_{\min}$ exists and satisfies $2/\sqrt{55}<d_{\min}\le1/\sqrt{13}$, which answers a question of Phetmak, Nuntasombat and Fakcharoenphol.
Read the full abstract

Machine Learning

Beyond the Log-Linear Frontier: Pre-Asymptotic Analysis of Finite-Budget Neural Scaling Laws

Classical neural scaling laws model excess risk as a separable power law in model capacity $N$ and dataset size $D$. At finite budgets, however, empirical learning curves and compute-optimal allocations depart from this log-linear frontier. This paper analyzes how spectral and domain heterogeneity govern these finite-scale departures. Along the capacity axis, Mellin–Barnes analysis of covariance resolvents yields a three-term expansion where competing spectral and regularization modes produce finite-scale curvature and logarithmic transitions. Because fitting multiple power laws on narrow pilot grids is ill-conditioned, we eliminate the loss floor via first-order Aitken differencing to obtain a well-conditioned closed-form initializer. Along the data axis, KKT duality reduces $K$-domain mixing to scalar water-filling, yielding finite-budget bottleneck bounds, an online sampling curriculum, and a simplex Hessian metric for cross-domain transfer. Along the iso-compute frontier $C=6ND$, we prove that the optimal allocation exponent drifts in proportion to capacity and data exponent variances, while Schur-complement reduction separates intrinsic drift from frozen-hyperparameter bias. Across 45 operator spectra, Pythia (70M–12B), Chinchilla (245 runs), and RegMix (17 domains), our estimators achieve $5.89\%$ holdout MAPE, cut high-compute allocation regret by up to $76.4\%$, and reduce cross-scale mixture RMSE by $59.4\%$ with zero selection regret.
Read the full abstract

Strongly Correlated Electrons

Field-induced magnetic quantum phase transitions in a bilayer triangular lattice antiferromagnet K$_2$Co$_2$(HPO$_3$)$_2$(C$_2$O$_4$)(H$_2$O)$_2$

We present a $^{31}$P nuclear magnetic resonance (NMR) study on the spin-$1/2$ bilayer triangular-lattice antiferromagnet K$_2$Co$_2$(HPO$_3$)$_2$(C$_2$O$_4$)(H$_2$O)$_2$. Measurements were performed at temperatures down to 0.06 K and magnetic fields up to 26 T. In the field range of 0.5 T to 4 T, the low-temperature spin-lattice relaxation rate $1/T_1$ exhibits a peak, signaling the onset of a magnetically ordered state. Within this ordered phase, $1/T_1$ shows gapless behavior, suggesting a spin supersolid phase. This ordering is suppressed around 4 T. Between 6 T and 18 T, clear splittings of the NMR spectra are observed, indicating the emergence of a second ordered magnetic phase. The field derivative of the low-temperature Knight shift $K_{\rm n}$ reveals changes in slope at approximately 10 T, 15.5 T, and 18 T. The first two features correspond to bulk magnetizations of $1/3$ and $2/3$ of the fully saturated moment, respectively, while the last one marks a quantum phase transition from the second ordered phase to the fully polarized state. These results establish the studied compound as a platform for investigating field-induced quantum phase transitions and emergent phases in frustrated bilayer triangular-lattice antiferromagnets.
Read the full abstract

All authors Yiqiu Liu, Chengcheng Yin, Xinyu Shi, Xiaohui Bo, Kefan Du, Zhanlong Wu, Rui Bian, Shuo Li, Jun Luo, Jie Yang, Yanhong Wang, Weiqiang Yu, Rui Zhou, Jinshan Zhang, Yi Cui, Hongcheng Lu

Computation and Language

Framing-Induced Vulnerability in Large Language Models: Evidence from Chinese News Headlines

Large language models increasingly evaluate, route, and moderate news, so a model's judgment of a headline has become a gatekeeping decision. We ask whether that judgment follows the fact being reported or the way the fact is framed. We combine an ecological audit of 63,926 Chinese news headlines with a controlled experiment. The experiment covers 30 real events, 30 synthetic events, and 29 natural headline pairs, tested on four Chinese models. Changing only the headline frame shifted both perceived misleadingness and event severity, and the direction and ordering of the shifts were the same for DeepSeek, MiMo, Dots, and LongCat. Emotionally intensified headlines raised severity ratings by 0.65 to 0.88 points, whereas positively reframed headlines lowered them by 0.34 to 0.52 points. Supplying the article body reduced this severity bias by 54 to 80 percent, and all four reductions were significant at p < .01. A small human reference study showed the same direction of effect. Headline-only LLM evaluation therefore carries a measurable robustness gap, and framing sensitivity belongs among the reliability properties that deployment and governance check.
Read the full abstract

Computer Vision

UniWeaver: Harnessing Unified Multimodal Models for Interleaved Text-Image Generation

Unified multimodal models (UMMs) have opened up new possibilities for generating rich interleaved text-image content. However, they generally lack native support for coordinating the two modalities, and translating coarse user requests into coherent multimodal compositions also requires explicit planning. We propose UniWeaver, a structured, model-agnostic harness framework that integrates multimodal planning, generation, and visual feedback to produce coherent interleaved outputs. To train UMMs for this task, we construct UniWeaver-10K, a dataset of 10,000 tasks across 20 task families, paired with execution trajectories guided by UniWeaver. We also introduce UniWeaver-Bench, a task-oriented benchmark systematically assessing text coverage, writing quality, visual coverage, and image-text alignment. Experiments show that UniWeaver achieves state-of-the-art performance and improves interleaved text-image generation across multiple UMM backbones.
Read the full abstract

All authors Hongxu Chen, Zixuan Zhong, Donghao Zhou, Haodong Li, Xiaomin Yu, Tianhao Li, Shiyao Yu, Zhongwei Wan, Pheng-Ann Heng