High Energy Physics (Phenomenology)
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Saturday 5 September
As a unique form of matter composed entirely of gauge bosons, glueballs are an important low-energy prediction of QCD. After decades of searches, the BESIII Collaboration recently suggested that $X(2370)$ may contain a dominant glueball component, based largely on evidence that it is approximately a flavor singlet. Recognizing that flavor-singlet character does not uniquely identify a glueball, we perform a comprehensive analysis using available mass, flavor-singlet, and decay constraints. We find that three flavor-singlet configurations-hybrid meson, tetraquark state, and trigluon glueball-can satisfactorily reproduce the existing experimental data. Among these possibilities, the hybrid structure best describes the measured three-pseudoscalar decay ratios. To ultimately pin down the dominant structure of $X(2370)$, we propose measuring the decay-width ratios $a_0(1450)\pi/[K_0^*(1430)\bar K+\mathrm{c.c.}]$ and $\phi\phi/[b_1(1235)\rho]$, which can exclusively distinguish these three scenarios.
Optics
Optoelectronic Reservoir Computing with an On-Chip True-Time-Delay Element
Compact delay elements remain a central challenge in photonic neuromorphic processors. Here, we demonstrate an optoelectronic delayed-feedback reservoir computer that incorporates a foundry-fabricated silicon nitride (Si$_3$N$_4$) true-time-delay circuit within its feedback loop. Eight cascaded Archimedean spirals provide a \SI{1.76}{\metre} on-chip optical path and a calculated passive group delay of \SI{11.63}{\nano\second}. At a feedback gain of $G=0.0223$, the system classifies sinusoidal and square waveforms without error (word error rate, WER\,$=$\,0 across all cross-validation folds), predicts the Mackey–Glass chaotic series with a best-fold normalized mean-square error (NMSE) of $0.0082$, and performs nine-class Japanese Vowels speaker classification with WER\,$=$\,0.0898. We also introduce a subcarrier phase-encoding method that maps the calculated $0.232\pi$ spiral-to-reference phase contrast onto the measured reservoir states through deliberate aliasing, achieving NMSE\,$=$\,0.0767 and WER\,$=$\,0 without increasing the insertion-loss-limited feedback gain. These results show that on-chip propagation delay in Si$_3$N$_4$ can operate as a functional component of an optoelectronic reservoir computer and motivate lower-loss, more-integrated implementations.
Read the full abstract
General Relativity and Quantum Cosmology
Observational properties of magnetized quantum Oppenheimer-Snyder black holes
We consider a thin accretion disk around the quantum Oppenheimer-Snyder black hole immersed in an asymptotically uniform magnetic field, together with the gravitational capture of test particles by the black hole. The field is described as a Wald test field and acts only on the weakly charged matter of the disk. We find that the quantum correction affects the disk only slightly, while the magnetic coupling strongly increases its radiative efficiency and luminosity and suppresses the Doppler blue peak of the iron K$\alpha$ line, which becomes a single peak below the rest frame energy. On the other hand, the quantum correction reduces all capture cross sections, while the magnetic field does not affect the capture of photons and confines charged particles to a finite region. Therefore, in principle, the two effects may be distinguished by combining spectral and shadow observations.
Read the full abstract
Computation and Language
From Narrative to Auditable Forecasts: A Structured Scaffold for Agentic Forecasting
LLM agents are increasingly used for live forecasting, where they retrieve up-to-date information and produce estimates for unresolved future events. However, current agentic forecasting often relies on implicit narrative aggregation: agents collect evidence, discuss it in prose, and often assign a probability without an explicit update path from evidence to forecast. This limits both forecasting accuracy and auditability. We propose AuditForecast, an agentic scaffold for structured probabilistic forecasting. AuditForecast first anchors the forecast with a suitable quantitative baseline model, uses model-guided data retrieval to derive a base probability, and then applies situational factor updates outside the model's scope through mechanical aggregation in odds space. This turns forecasting from a prose-based judgment into a structured process with explicit intermediate objects. Across multiple live forecasting benchmarks, AuditForecast improves forecasting accuracy and calibration relative to strong agentic baselines, surpasses market-implied references in several settings, and outperforms substantially more expensive deep-research agents while remaining Pareto-dominant in the cost–accuracy tradeoff. Beyond performance gains, AuditForecast produces an auditable forecasting report that makes forecast construction explicit and supports systematic post hoc analysis.
Read the full abstract
Quantum Physics
Boundary-Canceled Score-Hamiltonian Sampling
Diffusion sampling is related to imaginary-time quantum annealing through a forward/backward Schrödinger equation pair and its one-sided counterdiabatic connection. For a smooth, gapped Score-Hamiltonian path, the exact connection transports the target density at any evolution time. We apply quantum-annealing performance analysis to residuals from approximating this connection with a trained score. If the first $r$ terminal schedule derivatives vanish, the squared Hellinger error scales as $T^{-2r-2}$, with a prefactor determined by the terminal learning residual and inverse gaps. Direct two-dimensional imaginary-time evolution with one fixed neural score reproduces these powers and prefactors. Terminal schedule design thus suppresses the distributional effect of imperfect score learning without retraining the network.
Read the full abstract
Cryptography and Security
EvoSafeHarness: Evolving Model- and Domain-Specific Harnesses for Securing Agents
Large language model (LLM) agents can translate natural-language requests into real-world actions, creating safety risks from both indirect prompt injections and directly harmful requests. Existing system-level safety harnesses are typically designed by experts and applied across heterogeneous models and domains, despite substantial differences in model behavior, utility tolerance, and domain-specific safety requirements. We present EvoSafeHarness, a framework for automatically synthesizing deployment-specific safety harnesses for frozen LLM agents. Given a target model and domain specification, EvoSafeHarness jointly evolves a natural-language policy and executable enforcement logic using behavioral feedback from the target model. Each candidate is additionally evaluated through fresh-context adversarial review to filter rules that depend on benchmark artifacts. Across four agent benchmark families, EvoSafeHarness achieves a stronger safety-utility frontier than fixed expert-designed defenses. On DecodingTrust-Agent, evaluated across fifteen model-domain deployments, it reduces attack success rate from 50.9% to 12.6% for direct attacks and from 40.4% to 7.4% for indirect attacks, with a 3.3-point utility cost. On AgentDojo, it achieves 82.8% utility with 0.0% attack success rate, twice the utility of CaMeL at the same zero-attack-success operating point, while the same harness transfers unchanged to unseen AgentDyn suites. It also achieves the best score on Agent-SafetyBench for every evaluated victim model and maintains an attack success rate below 18% against an adaptive PAIR-style attacker with sixteen attempts per task. Analysis of the synthesized harnesses shows that domain semantics determine which safety relations and trajectory state must be tracked, while model and runtime behavior determine how and where those relations should be enforced.
Read the full abstract
Computer Vision
GenPuzzle: Benchmarking Visual Reasoning in Image Generation Models
Recent image generation systems increasingly combine multimodal understanding, reasoning, and synthesis, suggesting that they may do more than render plausible scenes. Yet existing evaluations emphasize aesthetics, prompt alignment, compositionality, or text-based answers, leaving unclear whether these systems can solve visual problems and faithfully express solutions in pixels. We introduce GenPuzzle, a benchmark for reasoning-centric image generation. GenPuzzle contains 2,005 problems across 12 tracks, spanning pattern completion, spatial construction, mazes, Sudoku, nonograms, tangrams, board games, matchstick puzzles, orthographic projection, and mathematical visual proof. Each task provides a visual puzzle and requires an image output that preserves the input state while executing a logically valid solution. GenPuzzle uses task-specific evaluation protocols: discrete grid outputs are transcribed and verified programmatically, while visually complex outputs are assessed with tiered, multidimensional, or binary multimodal large language model (MLLM) rubrics. We further select the automatic judge by measuring agreement with human reference scores. Across three frontier generators, the strongest model reaches only 40.57 Macro Overall, revealing frequent failures in logic, geometry, state preservation, and instruction execution. GenPuzzle provides a testbed for measuring progress from image rendering toward visual problem solving.
Read the full abstract
Cryptography and Security
From Review to Authorization: Key-Isolated Threshold Signing for LLM Agents
Autonomous LLM agents can turn semantic judgments into consequential actions such as payments and permission changes. If the same process makes these judgments and controls a reusable signing credential, a mistaken judgment can reach execution without further approval. To make LLM review govern execution, we present KITA, an authorization architecture that keeps the user's personal signing key and every threshold signing-key share outside all LLM processes. Under threshold-signature unforgeability and our system assumptions, an adversary controlling the proposer and fewer than t reviewer-signer domains cannot produce a valid authorization for a new action without signing contributions from t distinct domains. This execution-bound authorization integrity ensures that each accepted authorization includes a share from an uncompromised domain, bound to the canonical action and released only after authenticated reviewer approval. We implement an in-process prototype of KITA with a structured-output LLM adapter and threshold BLS. Six system tests validate quorum gating and message binding, seven controlled baseline scenarios compare authorization outcomes, and cryptographic microbenchmarks measure online signing cost and its scaling behavior.
Read the full abstract
High Energy Physics (Experiment)
Prospective study of light dark matter search in positron beam mode with DarkSHINE experiment initiative
We perform the prospective study on searching for invisibly decayed dark photons ($A^{\prime}$) in positron-on-fixed-target (POT) scheme with the DarkSHINE experiment simulation setup. In the positron-on-fixed-target scheme, two additional production channels of dark photons via annihilation with extranuclear electrons are present: non-resonant t-channel ($e^+e^-N \to \gamma A^{\prime}N$) and resonant s-channel ($e^+e^-N \to A^{\prime}N$), alongside bremsstrahlung emission similar to electron-on-fixed-target scheme. Based on a GEANT4 full detector simulation of an 8 GeV positron beam with $3 \times 10^{14}$ POT events, we developed an analysis strategy using missing energy and momentum signatures across three orthogonal signal regions optimized for these channels. Our results indicate that the additional annihilation channels of dark photon production enhance the dark photon invisible decay search sensitivity, improving the exclusion limits by approximately two orders of magnitude compared to previous constraints in the electron-on-fixed target (EOT) scheme, particularly near the center-of-mass energy threshold for annihilation production of dark photons.
Read the full abstract
Computation and Language
AlignDiff: Exploiting Model-Intrinsic Information for Better Preference Data Selection
Aligning large language models with human preferences remains a challenge, primarily due to the critical role of preference data quality in effective alignment. Existing datasets are frequently plagued by inherent noise and distribution shifts, which inherently limit model performance. To bridge this gap, we propose AlignDiff, a preference data filtering framework driven by intrinsic model signals. AlignDiff first identifies samples with clear preferences using both positive and inverse signals, then prioritizes the more challenging samples based on the average negative log-likelihood gap, encouraging the model to learn richer information from them. AlignDiff is evaluated on two widely used model families (LLaMA and Qwen) and three benchmarks widely adopted in the alignment community (AlpacaEval 2.0, Arena-Hard, and MT-Bench). Across all settings, it consistently outperforms seven strong baselines. We conduct comprehensive ablation studies to validate the effectiveness of AlignDiff, and further show that difficulty-based curriculum learning improves model performance.
Read the full abstract
Mesoscale and Nanoscale Physics
Large Nernst effect in chemically derived multilayer graphene at millitesla magnetic fields
Simple synthesis of graphitic compounds, their high conductivity, and integrability with other materials motivate the effort toward graphite-based thermoelectric generators. At the same time, semimetallic nature of graphite and graphene results in nearly-zero thermopower due to electron-hole compensation. Here, we observe large transverse thermopower in chemically derived multilayer graphene films with strong fluctuations of thickness and carrier density at low magnetic fields $B$. Using the scanning laser-induced heating of macroscopic film, we find that transverse (Nernst) thermoelectric voltage becomes comparable to the longitudinal thermoelectric voltage at the metal-doped graphene contact at $B^*\approx4$ mT and ambient conditions. Estimates of transverse thermopower $S_{xy}$ based on the known laser-induced temperature show that it is as large as $\sim 10$ $\mu$V/K at $B^*$, and raises in a sub-linear fashion to 250 $\mu$V/K at $B\approx315$ mT, the maximum field we reach with centimeter-sized permanent magnet. Extra increase in thermoelectric signal is achieved upon voltage measurement at Hall probes when the dc field lines are co-directional with local Nernst current. Our results show the promise of large-scale multilayer graphene for thermoelectricity generation.
Read the full abstract
Combinatorics
Spectral moments and characteristic polynomials of vertex expansion hypergraphs of graphs
The $s$-vertex expansion hypergraph $G^{[s]}$ is the $2s$-uniform hypergraph obtained by replacing each vertex of a graph $G$ with $s$ new vertices. A closed walk in $G$ is called an $s$-multiple closed walk if the number of times it arrives at each vertex of $G$ is divisible by $s$. In this paper, we obtain an expression for the spectral moments of $G^{[s]}$ in terms of $s$-multiple closed walks in $G$. Using these spectral moments, we give the characteristic polynomial of $G^{[s]}$.
Read the full abstract
Social and Information Networks
Polarity-Asymmetric Structural Calibration for Link Sign Prediction
Link sign prediction (LSP) aims to infer the positive or negative polarity of unobserved links in signed networks. Signed Graph Neural Networks (SGNNs) usually rely on signed-graph structural priors, including structural balance and homophily-like similarity, to guide message passing and prediction. These priors describe population-level tendencies, not guarantees for individual target edges. Their failures are especially costly under severe sign imbalance, where errors on minority and locally conflicting relations are harder to detect and correct. We propose Polarity-Asymmetric Structural Calibration (PASC), a target-edge structural-prior calibration framework for signed link prediction. PASC constructs a structure-only prior representation, estimates a target-edge structural prior score, and compares this score with a local signed-context cue to derive a conflict residual. The residual calibrates signed attention aggregation, target-edge gated fusion, and regime-adaptive optimization. Experiments on five real-world signed network datasets show that PASC consistently achieves the best Macro-F1 among representative baselines, with competitive AUC, Binary-F1, and Micro-F1. Structural-shift experiments further suggest reduced dependence on dense-neighborhood and local-closure shortcuts. Source code is available at https://github.com/iqqGGGGGGG/PASC-for-LSP.
Read the full abstract
Machine Learning
A First-Order Learning Algorithm for Online Resource Allocation with Constant Regret
We study a finite-horizon online resource allocation problem with initial resource capacities proportional to the horizon. In each period, a request type is observed and one action is chosen from a finite menu. Each action earns a reward and consumes a vector of resources. We present a primal first-order learning policy that, in each period, performs one gradient ascent update of the action coordinates associated with one request type, for a fluid relaxation in which resource overuse incurs a quadratic penalty. The policy achieves $O(1)$ expected additive regret relative to the hindsight optimum, with a bound independent of the horizon $T$, under a general arrival model. This model includes independent and identically distributed arrivals with unknown probabilities, independent nonstationary arrivals with known probabilities, independent nonstationary arrivals with unknown probabilities and one historical sample path, and Markov-modulated arrivals with an unknown Markov chain. The policy does not solve any linear program, and the regret bound does not require a nondegeneracy assumption on the fluid linear program.
Read the full abstract
Artificial Intelligence
The End of AI Exponentiation: Fluttering Inside and Outside AI Bubble
The exponentiation of Artificial intelligence (AI) in the recent past has entered a transformative era that has been driven by the growth in large language models (LLMs), large-scale compute infrastructures, and autonomous reasoning systems. However, the rapid acceleration of AI has increasingly shown technological, societal, economic, ethical and infrastructural challenges associated with peak data limitations, rising computational demands, synthetic data recursion, valuation inflation, and societal instability. The traditional scaling paradigms that have powered the modern AI systems are gradually encountering friction in sustaining continuous exponential growth. This paper views “the end of AI exponentiation,” thus exploring how it flutters inside and outside the bubble, where instability emerges within the AI ecosystem through compute and data-center races, speculative investments, and the rat-race toward superintelligence, and outside the ecosystem through labor disruption, governance concerns, public uncertainty, and geopolitical acceleration surrounding future intelligent systems and infrastructures globally.
Read the full abstract
Spectral Theory
Neumann–Dirichlet eigenvalue comparison at the first Dirichlet threshold in the plane
Let $\Omega\subset\mathbb R^2$ be a bounded connected Lipschitz domain, and let $\{\mu_j(\Omega)\}_{j\geq1}$ and $\{\lambda_j(\Omega)\}_{j\geq1}$ denote the Neumann and Dirichlet Laplacian eigenvalues, respectively, counted with multiplicity. We prove that $$ \mu_3(\Omega)<\lambda_1(\Omega), $$ thereby removing the simple-connectivity assumption from the previously known planar result at the first Dirichlet threshold. We also establish a three-spectrum counting inequality relating the Dirichlet, Neumann, and conductivity spectra.
Read the full abstract
Robotics
A4A: Cross-Embodiment Transfer of Action-Oriented 4D Affordances from Human Demonstrations
Human demonstrations contain rich manipulation knowledge, but it remains unclear what information can be transferred effectively to robot control. Existing affordance representations are typically formulated as 2D masks, 3D regions, contact points, or actionability scores, and therefore primarily identify where interaction may occur. However, effective manipulation also requires modeling how interaction-relevant geometry evolves during task execution. To bridge this gap, we introduce action-oriented 4D affordances, which represent the language-conditioned future trajectories of interaction-relevant 3D points. These trajectories capture task-conditioned geometric evolution rather than embodiment-specific actions, enabling transferable interaction priors across humans and robots. Based on this representation, we construct a large-scale action-oriented 4D affordance dataset from existing human–object interaction video data and complementary RGB-D demonstrations, and introduce A4A, an affordance-to-action framework that uses 4D affordance trajectory prediction to pretrain robot policies before manipulation fine-tuning. Experiments in both simulation and the real world validate the effectiveness of A4A, showing that pretraining with action-oriented 4D affordance data consistently improves the manipulation performance of diverse VLA policies. These results establish action-oriented 4D affordances as an effective cross-embodiment representation for transferring manipulation knowledge from human demonstrations to robot control.
Read the full abstract
Commutative Algebra
Weak Polynomial Completeness Does Not Imply Almost Polynomial Completeness
We show that weak polynomial completeness does not imply almost polynomial completeness by constructing an explicit extension of domains \(D\subseteq A\) such that \(\Int(D)\subseteq\Int(A)\) but \(\Int(D^2)\nsubseteq\Int(A^2)\).
General Relativity and Quantum Cosmology
Self-supervised reconstruction of transients in data from space-borne gravitational-wave detectors
Space-based gravitational-wave (GW) data may contain transient signals whose waveform morphologies are not known in advance. Extracting these signals is important for characterizing new sources and mitigating instrumental anomalies. However, existing deep neural network (DNN)-based extraction approaches rely on clean training targets and waveform-class-specific examples, which may limit their applicability when the transient morphology is not specified in advance. This work develops a Noise2Noise (N2N)-inspired self-supervised framework that learns from noisy observations without clean training targets and requires no transient-specific waveform templates at inference. A single model trained on noisy massive black-hole binary (MBHB) observations provides high-overlap MBHB recovery and recovers the dominant morphologies of instrumental glitches and other GW transient signals such as cosmic string bursts in source-confused test data. Beyond waveform recovery, when independent information identifies a candidate transient as instrumental, its extracted waveform can be subtracted from the data without an anomaly-specific template. In a simulated continuous data stream, this procedure substantially suppresses the injected-anomaly power within the glitch-dominated frequency band. These results support the use of self-supervised extraction for initial waveform estimation of candidate transients with unknown morphologies, enabling subsequent characterization and, where appropriate, conditional subtraction of instrumental anomalies.
Read the full abstract
Artificial Intelligence
Multimodal Resource-Exhaustion Attacks on Vision-Language Models via Joint Pixel-Prompt Optimization
Resource-exhaustion attacks against autoregressive vision-language models (VLMs) typically assume unimodal threat models, treating the image branch as the primary optimization surface while holding user-visible prompts fixed. Even recent loop-centric variants remain confined to this single-channel paradigm, leaving the exploitation of availability unexplored as a cross-modal optimization problem over jointly controllable input surfaces. We introduce Joint Pixel-Prompt Optimization (JPPO), the first compound adversarial framework elevating the visible prompt to a first-class adversarial variable alongside image perturbations. Under a restricted joint-input threat model, JPPO performs coupled, stagewise optimization over both the pixel and prompt surfaces. This produces synergistic cost amplification, mechanistically distinct from loop-dependent failures, exhibiting negligible loop incidence in our experiments. Evaluating five open-source VLM families on MS COCO and ImageNet under an 8/255 infinity-norm budget, JPPO achieves over 4.6x latency and 5.3x energy amplification on Qwen2.5-VL-7B, and over 36.6x latency with 32.7x energy amplification on BLIP-2. This represents the strongest cost amplification among directly compared baselines while requiring substantially fewer optimization iterations. Ablations confirm this amplification arises from multimodal coordination rather than prompt length or isolated modalities. These findings reveal structural blind spots in current VLM serving defenses, motivating cost-aware robustness evaluation as a first-class security requirement for multimodal deployments.
Read the full abstract
Computer Vision
UniFusion: Sparse-View 4D Reconstruction via Unified Spatio-temporal Depth Alignment
In this paper, we address the challenging problem of 4D reconstruction from sparse-view videos. This setup usually relies on monocular depth estimation to provide priors for the reconstruction model. A key challenge arises from limited cross-view overlap and temporal variation, making monocular depth predictions inconsistent across views and time. Existing methods align spatial and temporal dimensions in separate stages, requiring foreground segmentation masks while failing to leverage temporal cues for cross-view alignment. Contrary to these methods, we propose a unified spatial-temporal depth alignment framework that jointly resolves cross-view and cross-time inconsistencies without distinguishing foreground/background. Our method represents depth maps across views and time as a set of spatio-temporal neural fields. This representation not only yields fast convergence, but also captures spatio-temporal correlation among depth maps implicitly, without dependence on external segmentation/tracking models. We also propose a multi-view depth-order loss while leveraging the classic scale-and-shift-invariant loss to further improve the final depth quality. The aligned depths initialize and supervise Gaussian splatting models for 4D reconstruction. Experiments on Ego-Exo4D and EgoHuman demonstrate that our improved depth alignment substantially benefits dynamic Gaussian-splatting-based reconstruction methods for novel-time/view synthesis and geometry accuracy/consistency.
Read the full abstract
Nuclear Theory
Improving the accuracy of the Lorentz Integral Transform method using complex kernels
The Lorentz integral transform (LIT) method is a powerful tool for calculating quantum response functions; however, it requires an ill-posed inversion. Here we show that solutions of the LIT equation can also determine two additional integral transforms: a complex Stieltjes transform and a double-pole transform, at little additional computational cost. A Fourier analysis of the associated deconvolution problem shows that both alternative kernels are better conditioned than the Lorentzian kernel at the same width parameter $\Gamma $. We benchmark the resulting inversions for deuteron photodisintegration by adding controlled noise to the LIT solutions. Relative to the standard LIT, the alternative kernels yield response functions with reduced noise-induced scatter: by a factor of 2-3 for the complex Stieltjes kernel and 3-4 for the double-pole kernel.
Read the full abstract
Optics
Rapid and high-sensitive NV-based microwave field imaging via digital lock-in amplification for on-chip microstrip diagnostics
High-resolution, high-sensitivity microwave (MW) magnetic field imaging is indispensable for non-destructive integrated circuit (IC) testing, radio-frequency device characterization, and spintronic research. Yet, the practical utility of these techniques is severely constrained by the pervasive challenge of isolating weak magnetic signatures from intense optical and electronic noise, which fundamentally limits both acquisition speed and detection sensitivity. Here, we overcome this barrier by introducing a wide-field imaging scheme based on an ensemble of diamond nitrogen-vacancy (NV) centers, synergistically combined with digital lock-in amplification (DLA). By exploiting digital demodulation, the DLA precisely extracts the MW-field response at a specific modulation frequency from background noise (e.g., laser intensity fluctuations), dramatically improving the signal-to-noise ratio (SNR). Consequently, our system attains a magnetic field sensitivity of 126 nT/$\sqrt(Hz)$. Critically, the unprecedented SNR permits a pixel dwell time of under one millisecond, allowing full-field images to be acquired within seconds-more than an order of magnitude faster than state-of-the-art NV-based wide-field techniques. This combination of speed, sensitivity, and micron-scale spatial resolution (1.6 $\mu$m) paves the way for quasi-real-time, non-invasive diagnostics of dynamic MW devices and integrated circuits.
Read the full abstract
Machine Learning
One Rate Is Not Enough: Adaptive Anisotropic Learning Rates for LoRA Fine-Tuning
Low-rank adaptation (LoRA) has become the standard for parameter-efficient fine-tuning of large language models. Most LoRA variants follow a uniform-LR convention, applying a single global learning rate across every rank-one component of every adapter. We show that this convention overlooks substantial within-module heterogeneity, where the rank-one components of a LoRA adapter update at highly uneven rates and low-velocity modules converge to concentrated singular spectra that underutilize the nominal rank budget. To address this, we propose an adaptive anisotropic learning-rate model that assigns each rank-one component its own effective learning rate, computed online from training-time signals and mean-normalized per module to preserve the global LR budget. AnLR-LoRA instantiates this model with two signals available during AdamW optimization, namely function-space velocity and Adam SNR, as a lightweight scheme with no extra trainable parameters. Across commonsense reasoning, natural language generation and visual instruction-tuning benchmarks, AnLR-LoRA consistently improves over LoRA while encouraging broader use of rank capacity, with gains that remain robust across a wide range of global learning rates and transfer cleanly to other LoRA variants.
Read the full abstract
Machine Learning
CALM: Class-wise Agreement and Label-gated Disagreement Modulation for Decentralized Federated Learning
Conventional federated learning relies on parameter averaging, which forces clients to be doubly homogeneous: all must run an identical architecture, and accuracy degrades when local data are non-IID. Decentralized federated distillation sidesteps both: each client runs its peers' model snapshots as teachers on its own local data and distills from their soft predictions, with no server, no public data, and no shared architecture. Under severe non-IID skew, however, the trustworthiness of the aggregated teacher target is a matter of degree, yet existing pipelines make hard, all-or-nothing decisions: outlier teachers are discarded by threshold, and whatever target survives is trusted in full. We propose CALM, which replaces every hard decision with a smooth trust gate at three levels: per class, teachers are weighted by agreement with the peer consensus; per sample, distillation is scaled by the teachers' divergence from that target; and a label gate scales it by how strongly the target supports the sample's true label. None of this adds communication or auxiliary data. On CIFAR-10, SVHN, OrganAMNIST, and Google Speech Commands with heterogeneous client architectures under Dirichlet label skew, CALM consistently outperforms uniform and hard-filtered distillation and matches or exceeds competing heterogeneous-FL methods.
Read the full abstract
Econometrics
Fixed-smoothing Uniform Inference for Quantile Regression
This paper develops fixed-smoothing inference methods for time-series quantile regression that are robust to heteroskedasticity and autocorrelation. Our approach is uniformly valid over quantile levels and accounts for dependence over time and across quantiles. It enables uniform confidence bands, Wald and Sup-t tests for joint hypotheses, and tests of shape restrictions, providing a unified framework for assessing heterogeneity in quantile effects. Under weak dependence, uniform inference for quantile regression processes is non-pivotal because the limiting distributions depend on the long-run covariance structure across quantiles. To address this issue, we develop two complementary approaches. The uniform-in-$\tau$ method estimates this structure and simulates the non-pivotal limiting distribution. For certain tests involving finitely many quantiles, the stack-Wald method delivers pivotal fixed-smoothing inference. We establish the asymptotic validity of both approaches. Simulations show improved size control over existing HAC-based procedures while maintaining good power. An application to stock-return prediction reveals substantial heterogeneity in predictive effects.
Read the full abstract
Computation and Language
What if LLMs Ate Their Words: Causal History Effects in Multi-Turn Interaction
Multi-turn interaction creates a feedback process in which an LLM's previous responses become context for later behavior. Prior work shows substantial multi-turn degradation and that assistant-generated history can affect later behavior. However, it remains unclear how these effects manifest across models, tasks, turns, and inside a model. We study these gaps across six task families and five models. Degradation from fully specified single-turn input (FULL) to progressively revealed multi-turn interaction (SHARDED) is clearly task- and model-dependent, and stronger one-shot performance does not imply greater interaction robustness. We then retrospectively analyze completed SHARDED conversations by replaying the user messages already observed in each trajectory while editing only assistant-generated history. Replacing prior assistant responses with neutral content (termed neutralization) changes downstream min-max normalized performance by +.027 across 2,973 trajectories. On a prespecified length-controlled subset, short and length-matched neutralization yield nearly identical effects (+.069 versus +.068), showing that simple context shortening is insufficient to explain the effect of history editing. Turn Surgery further intervenes on one assistant turn at a time. Among 237 selected degraded trajectories, 63.7% contain at least one beneficial intervention, while most tested positions remain unchanged; for binary tasks, 48.4% admit a fail-to-success reversal. An open-weight case study links behaviorally consequential history changes to measurable downstream state differences, but finds task-dependent rather than universal internal signatures. Overall, assistant-generated history has active but selective effects on multi-turn performance, motivating selective rather than uniform history management.
Read the full abstract
Software Engineering
Broken on Arrival: Non-Functional Quantized LLM Artifacts in Public Model Registries and How to Catch Them
Developers download quantized large language models (LLMs) from public registries to run them locally. None of the registries we examined documented functional testing of their files. We tested all 305 eligible code-capable GGUF tags (the llama.cpp format) across 15 model lines in the official Ollama library, plus 22 popular Hugging Face community files. A 15-task code screen identified suspects, which we retested on all 164 HumanEval+ tasks. Seven official tags (2.3%) solved at most six of them. All were low-bit quantizations of models with at most four billion parameters. We rebuilt five suspect files from the original weights. All five registry files matched our uncalibrated rebuilds tensor for tensor, so none was corrupted. Exact reproducibility did not imply function. With the weights, toolchain and tensor mix held fixed, adding an importance matrix raised Qwen2.5-Coder-3B scores at three quantization levels from zero to 76-113 of 164 tasks. The same uncalibrated recipe left the smaller Qwen2.5-Coder-1.5B functional, showing that the recipe alone does not predict failure. None of 66 sampled tags that passed the screen was non-functional. We release the audit data and the quantcheck tool and recommend screening low-bit tags before publication, then calibrating or withholding failures.
Read the full abstract
Machine Learning
Grounded and Faithful P&ID Reasoning: Constraining Vision-Language Models with Recovered Evidence Graphs
Piping and Instrumentation Diagrams (P&IDs) are the authoritative maps of process plants: isolation, maintenance, and HAZOP decisions depend on what connects to what. Vision-language models describe these sheets fluently, yet they often invent or miss process connections—and an invented or missed link can reverse an isolation or reachability call, so a plant decision cannot trust a fluent answer that was never checked against the linework. We instead recover an explicit graph of the drawing—its symbols, the process connections between them, and the tags that name them—and then require the model to answer only by querying that graph through seven read-only operators, so a topology claim is returned only when it cites the query results that support it. On TopoPID-VQA, a new suite of 3000 topology questions over these sheets, Graph-Grounded Harness (Ours) raises exact match accuracy from 36.7–41.3% under image-only prompting to 74.3–76.0% for Qwen3-VL-4B, Qwen3-VL-8B, and Gemma-4-E4B. It does so on an imperfect substrate: on Digitize-PID dataset the recovered graph scores F1 0.742 on exact process connections, and 0.801 once symbols and tags are pooled in. The residual errors track that gap—grounding pays off where the recovered graph is right, and perception error still breaks topology questions where it is not.
Read the full abstract
Software Engineering
Correct Tests Are Not Enough: Measuring and Training Oracle Conversion in Specification-Based Test Generation
Generating tests from a natural-language specification requires both an input that exposes faulty behavior and a correct expected output. These requirements need not improve together: a model can increase test correctness by choosing easier inputs, or discover useful inputs whose expected outputs it cannot predict. We study this interaction through executable reward decomposition and suite-level oracle-conversion measurement. Our generator jointly emits five input–output tests in one response. During training, audited reference programs provide correctness feedback, while a fixed bank of faulty programs provides two utility signals: potential input kill and effective kill after checking the generated output. An additive GRPO objective preserves both signals without requiring execution at inference time. On an audited TC-Bench split with 506 training and 142 evaluation tasks, three independently trained Qwen3.5-9B runs at step 75 increase full-test correctness from 28.59% to 42.54%, input kill from 24.06% to 25.27%, and effective full kill from 12.23% to 14.15%. Matched 50-step ablations reveal a trade-off: removing kill rewards yields higher correctness and slightly higher full kill, but lowers input kill to 21.60%. A fixed-input source–oracle crossover on 64 training-pool tasks attributes the principal NoKill-to-FullKill difference to harder input selection rather than worse output prediction on identical inputs. These results identify oracle conversion as a measurable bottleneck and show the benefits and limits of preserving input-utility feedback in joint test generation.