Peer Revenue Equity in science

Papers

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

All of arXiv

Sunday 30 August

Software Engineering

Drive the Thoughts: Runtime Monitoring of VLA Reasoning-Trajectory Consistency

While machine learning models help autonomous vehicles (AVs) navigate complex environments, their black-box nature complicates validation and verification (V&V). Vision-language-action (VLA) models offer a unique opportunity: alongside trajectories, they generate an explicit chain-of-thought (CoT) explaining their decisions. We posit that this CoT can serve as a candidate specification to cross-check trajectories and expose unsafe behavior. To assess this, we curate DriveAlignBench, a dataset of 150 CoT-trajectory pairs from NVIDIA's Alpamayo 1.5 VLA, manually annotated for CoT reliability, CoT-trajectory consistency, and trajectory safety. We find that 33.3% of CoTs are unreliable, and among reliable CoTs, 26% have inconsistent trajectories. To detect these cases, we propose integrating a CoT-trajectory consistency check into a runtime monitor. This check is nontrivial because CoTs express open-vocabulary, scene-relative maneuvers, while trajectories are low-level ego-motion sequences whose semantics depend on road geometry and motion context. To bridge this gap, we develop a family of automated consistency monitors. The strongest, the lane-aware monitor, adds lane-relative evidence to the input of the judge model scoring consistency. This raises F1 across all four evaluated judges from 0.56-0.62 with raw waypoints to 0.65-0.82, showing that evidence representation improves detection across judges. Moreover, for reliable CoTs, inconsistency is associated with unsafe behavior: 27% of human-labeled inconsistent trajectories are unsafe compared to 3% of consistent ones, and the lane-aware monitor flags most of these unsafe cases. We release DriveAlignBench, the monitor implementations, and per-run results at https://github.com/less-lab-uva/drive-the-thoughts.
Read the full abstract

Computation and Language

SUP-MIMIC: A Multi-Task Clinical Diagnosis Benchmark for Evaluating LLMs' Robustness to Contradictory Evidence

Current evaluations of large language models (LLMs) primarily focus on factual knowledge retrieval, overlooking the fundamental challenge of navigating the complex, non-bijective mappings between clinical indicators and diagnoses. Existing benchmarks fail to assess whether large language models truly possess the reasoning capability required for diagnostic ambiguity scenarios, where identical clinical presentations may correspond to different etiologies, and diagnostic convergence scenarios, where heterogeneous symptoms ultimately indicate the same disease. To address this issue, we propose SUP-MIMIC, a multi-task framework utilizing MIMIC-IV-v3.1 that comprises Basic Assessment (BA), Diagnostic Divergence Task (DDT), and Diagnostic Convergence Task (DCT). Specifically, DDT is designed to evaluate the model's "one-to-many" disambiguation capability among phenotypically similar cases, while DCT assesses the model's ability to identify "many-to-one" diagnostic patterns across different pathophysiological pathways. Comprehensive evaluation of state-of-the-art LLMs reveals substantial performance degradation on DDT and DCT compared to baseline tasks, exposing a systemic reliance on statistical shortcuts over genuine causal reasoning. Our findings further highlight a conservative bias toward "healthy" predictions, implying non-trivial risks for missed diagnoses in realistic medical settings. This work establishes a rigorous methodology for quantifying clinical reasoning robustness and provides a roadmap for enhancing the safety of language models in clinical medicine.
Read the full abstract

All authors Yi Yu, Bo Wang, Chong Feng, Ge Shi, Xia Liu, Ziyi Yang, Xuewen Shi

Distributed Computing

Bridging Agent Semantics with Spot Capacity: An Elastic and Recoverable Service Model

LLM agents increasingly drive long-running cloud inference workloads in which model calls differ in urgency, redundancy, completion semantics, and replay cost. Model-as-a-Service (MaaS) platforms expose several service models for trading cost against latency, availability, and capacity commitment. These models operate primarily at request, job, or endpoint scopes and provide limited support for combining transient platform supply with the evolving semantics of an agent task. We present SemSpot, a semantics-aware service model that allows agent applications to leverage the spot capacity of LLM inference platforms. At the request level, SemSpot lets a provider publish short-lived offers over successful price, completion probability, and failure-notification deadline; the agent runtime selects among these offers using the current task state and completion rule. An audit of 1,535 cases from six agent benchmarks identifies four recurring workflow structures and shows how this service model may produce different cost, service-time, and fallback behavior. With specialized MaaS support, token-level SemSpot further preserves provider inference state and runtime-verified semantic segments inside a long request. We develop the service model, economic boundary, and the cross-layer research agenda required to realize SemSpot.
Read the full abstract

Plasma Physics

Unified Gas-Kinetic Wave-Particle Method for Multiscale Simulation of Vlasov-Poisson-Fokker-Planck System

A unified gas-kinetic wave–particle method with Fokker–Planck collisions (UGKWP-FP) is developed for the Vlasov–Poisson–Fokker–Planck system. The collision operator is modeled by the Lenard–Bernstein operator, whose stochastic representation corresponds to the Ornstein–Uhlenbeck process in velocity space. To extend the UGKWP framework beyond the conventional Bhatnagar–Gross–Krook (BGK) collision model, the Fokker–Planck operator is decomposed into a nonstiff drift–diffusion contribution and a stiff thermalization contribution. The former is retained in the particle dynamics through a modified Ornstein–Uhlenbeck process, whereas the latter is represented by a BGK-type relaxation toward the local Maxwellian. This decomposition enables an adaptive wave–particle representation: the method follows stochastic particle dynamics in rarefied regimes and increasingly represents the rapidly equilibrating distribution by the analytical wave component as the continuum regime is approached. The modified friction coefficient is constructed to recover the original Fokker–Planck dynamics in the rarefied limit while preserving the hydrodynamic limit under strong collisions. Numerical experiments demonstrate that the proposed method captures velocity-space drift and diffusion, recovers the expected kinetic and continuum behavior across a range of Knudsen numbers, and reproduces the characteristic evolution of collisional plasma phenomena.
Read the full abstract

All authors Zhigang Pu, Chang Liu, Yixiao Wang, Kun Xu

Machine Learning

Predicting the Unpredictable: LLM-powered Long-term Chaotic Time Series Forecasting under Short-term Observations

Chaotic time series forecasting is a challenging task due to its sensitivity to initial conditions and long-term unpredictability. Traditional methods typically rely on sufficient temporal trajectories to learn long-term dynamics, which limits their applicability when only short-term observations are available. While recent Large Language Models (LLMs) have shown great potential for time series forecasting, their temporal representations are not explicitly tailored to the phase-space structure and nonlinear evolution of chaotic systems. To address these issues, we propose PAC-LLM, a phase-space-aware adaptive fusion framework for long-term chaotic time series forecasting powered by LLMs. PAC-LLM leverages learned phase-space features and textual information to fully enable LLM's time series forecasting capacity. In particular, we design an auxiliary feature module and a gated weighting mechanism for multivariate coupling information fusion and selection. Extensive experiments on representative chaotic systems demonstrate that our method outperforms existing fine-tuned and zero-shot baselines in both short-term and long-term predictions. Our ablation study further confirms the effectiveness of each key component in PAC-LLM.
Read the full abstract

Information Retrieval

The Edge Spectrum of Choice-Derived Item Graphs: Strong and Weak Edges Encode Different Relations in Collaborative Filtering

Graph collaborative filtering relies on item–item graphs whose edges are used for positive smoothing, under the implicit assumption that stronger edges encode more of the same relation as weaker ones. We show that this assumption fails for a practically important class of graphs: those whose edge weights come from a choice model. On such graphs, strong and weak edges encode qualitatively different relations, which we call an edge spectrum. Specifically, strong edges concentrate on the in-slate competitors of clicked items, exactly the pairs that the within-slate ranking gradient pushes apart, while weak edges do not. We formalize this as a sign mismatch between the smoothing operator and the ranking gradient, and prove that co-click graphs cannot exhibit the same misalignment by construction. This diagnosis explains three empirical observations on MIND and EB-NeRD: (i) drop-in choice-derived operators do not beat co-click, despite indexing structurally distinct neighborhoods; (ii) uniform scalar fixes (sign flip, in-slate margin loss) fail predictably, because the misalignment lives in the graph, not in the loss; (iii) only edge-magnitude-aware operators, with the regime boundary located by the diagnosis rather than by tuning, recover the predicted ordering. The neighbor cutoff $k$ is therefore a semantic switch, not a sparsification hyperparameter. Our claim concerns which interventions fail or succeed and why, not absolute headline gains, which the diagnosis itself predicts to be small under the attenuated propagation channel we observe. We turn the diagnosis into a reusable protocol practitioners can run before deploying any choice-derived item-side operator. Code: https://github.com/kyomusso/Edge-Spectrum-in-CF.
Read the full abstract

Computer Vision

TRINITY: A Multi-Perspective Benchmark for Personal-Style Video Highlight Detection

Traditional video highlight detection relies on a narrow, event-centric definition of saliency, which often fails to generalize to unconstrained personal videos where highlights are heterogeneous and perspective-dependent. To address this, we introduce TRINITY, a multi-perspective benchmark that decomposes highlight saliency into three complementary dimensions, Event, Emotion, and Nature, within a unified temporal framework. Leveraging this multi-faceted view, we propose a shared-backbone multi-branch architecture designed for parallel multi-perspective prediction via view-specific experts. Comprehensive experiments demonstrate that our method significantly outperforms state-of-the-art baselines, achieving gains of +7.15/+3.62 mAP (rho=15%/50%) on Mr. HiSum and +10.82 mAP on YouTube Highlights. These results validate that multi-perspective modeling provides a more robust and comprehensive formulation of video saliency, especially for complex real-world scenarios. The benchmark and relevant codes will be released upon acceptance. The benchmark is available at https://huggingface.co/datasets/vanilladucky/TRINITY and the code is available at https://github.com/vanilladucky/TRINITY.
Read the full abstract

All authors Qianqian Chen, Hyun Bin Kim, Denzel Elden Wijaya, Yang Yi, Bo Liu, Yangkai Ding

Machine Learning

Event-triggered Control and Online Learning for Networked Systems under Computational Delays

Online learning-based control is a promising approach to control uncertain systems, where unknown components are identified during operation to improve control performance. However, resource-intensive online learning algorithms introduce non-negligible computational delays, especially when executed on systems with limited local computational resources. To mitigate this, an in-network online learning-based control structure is employed by deploying the learning-based controller on a remote computation node and connecting it via a communication channel. In this paper, control performance guarantee is first established by deriving tracking error bound for the in-network control architecture, while accounting for computational delays. The derived tracking error bound allows for diverse communication and computation strategies under a specific condition, including time-/event-triggered mechanisms. Additionally, the trade-off between communication and computation performances is shown for a given desired control performance. Furthermore, to enhance the efficiency in both communication and computation, an efficient control framework with an asynchronous event-triggered mechanism in both control and online learning is devised under the existence of computational delay. The proposed event-triggered strategy is proven to achieve the same control performance as time-triggered scenario while excluding Zeno behavior. Finally, we derive an explicit expression of the proposed event-trigger condition for exponentially stabilizable systems, and demonstrate its effectiveness through simulations.
Read the full abstract

All authors Xiaobing Dai, Armin Lederer, Zewen Yang, Sihua Zhang, Lu Wan, Yang Tang, Sandra Hirche

Computation and Language

SemTrace: Source-Grounded Semantic Signatures for Tracing LLM Exposure to Protected Documents

Large language models are increasingly used to read documents and produce downstream text, creating a provenance problem when the document owner cannot control or inspect the model that performs the generation. We introduce SemTrace, a source-grounded semantic watermark for detecting whether a generated review was influenced by a known protected manuscript copy. Rather than biasing token probabilities or imposing surface-form patterns, SemTrace constructs a document-specific binary signature from factual propositions that are directly supported by the manuscript itself. A protected PDF invisibly carries a content contract that selects one fact from each binary pair and asks an instruction-following reviewer to express those facts in fixed review slots without changing its independent evaluation. A frozen natural language inference model then decodes the resulting semantic evidence with explicit erasures and scores the recovered bits against the codeword assigned to that copy. This design targets model-agnostic, assigned-copy exposure detection while keeping the watermark semantically tied to the source document.
Read the full abstract

All authors Junyan Zhang, Yudong Zeng, Yongwei Huang, Zuhao Ouyang, Hong Chen, Xuming Hu

Functional Analysis

Spectral Rigidity of Commutators: Dynamics, Resonance, and Nilpotency

Let $A, T \in M_n(\mathbb{C})$ and let $\Delta_A(T) = AT - TA$ denote the inner derivation induced by $A$. We determine when $T$ and $\Delta_A(T)$ are nilpotent under the second-order relation $$ \Delta_A^2(T) + \alpha\, \Delta_A(T) + \beta\, T = 0, \qquad \alpha, \beta \in \mathbb{R}, $$ according to the location of the roots of $z^2 + \alpha z + \beta$. If the roots have nonzero real parts of the same sign, then both $T$ and $\Delta_A(T)$ are nilpotent. If the roots are purely imaginary and nonzero, no nilpotency conclusion holds in general, whereas if one root is zero and the other is nonzero, $\Delta_A(T)$ is nilpotent but $T$ need not be. The double zero root gives the Kleinecke–Shirokov theorem. For real roots of opposite signs, the answer is governed by an arithmetic threshold: writing $z_1/z_2=-p/q$ in lowest terms, every solution is nilpotent when $p+q>n$, while for $p+q\le n$ an explicit cyclic construction yields solutions for which both $T$ and $\Delta_A(T)$ are invertible. Finally, for a relation $$ \sum_{k=0}^{m} c_k\,\Delta_A^k(T)=0 $$ of arbitrary order, every iterated commutator $\Delta_A^j(T)$ is nilpotent whenever the roots of the associated polynomial lie in an open half-plane whose boundary passes through the origin.
Read the full abstract

Information Theory

A General Plotkin-type Bound on Function-Correcting Codes with Wyner-Graham Distance

Function correcting codes (FCCs) are designed to protect a specified function evaluation of messages at a higher level than the level of protection for messages, against errors while reducing the redundancy required for reliable communication. FCCs have thus far been studied for channels matched to various distances, including the Hamming and Lee distances. Every function partitions the message space into preimage sets corresponding to its distinct function values. Existing Plotkin-type bounds on the optimal redundancy of FCCs under the studied distances, applicable to arbitrary functions on the message space, depend on the pairwise distances among all the message vectors. This makes these bounds difficult to compute. We derive a general Plotkin-type bound on the optimal redundancy of FCCs under Wyner Graham distances, which include the Hamming and Lee distances as special cases. Our bound depends only on the cardinalities of the preimage sets and the sum of pairwise distances only among vectors within each preimage set. This approach significantly reduces the computations required to evaluate the existing Plotkin-type bounds and yields simplified bounds that are easier to compute for specific functions. We obtain simplified Plotkin-type bound for linear functions under the Wyner-Graham distance. Furthermore, the existing simplified bounds for linear functions under the Hamming and Lee distances are recovered as special cases of the proposed bound. We also obtain simplified bounds for several important classes of functions, including the Hamming weight function, the Hamming weight distribution function, the monomial functions under the Hamming distance, and the modular sum function, the Lee weight function, and the Lee weight distribution function under the Lee distance.
Read the full abstract

All authors Kanchana Lokshmii Jagatti, K. Hareesh, N.T. Rashid Ummer, B. Sundar Rajan

Instrumentation and Methods for Astrophysics

AO3k + SCExAO: on-sky wavefront quality and demonstration of novel WFS techniques with the double XAO system

The Subaru Coronagraphic Extreme Adaptive Optics (SCExAO) system, fed by its upstream 3000-actuator "woofer" (AO3k), serves both as a platform for high contrast imaging (HCI) technology maturation and as a science instrument for imaging, spectroscopy, and polarimetry of exoplanets and disks. SCExAO operates in the visible and near-IR and offers a wide choice of instrument configurations. Over the last year, AO3k/SCExAO underwent significant upgrades to bring improved capabilities and support new developments, all while easing science operations. The new configuration features a beam switcher so that light can be shared between several instrument modules. The system is evolving toward a tighter integration between multiple WFSs and AO stages of correction, with the first stage (AO3k) providing visible and nearIR WFSing, as well as laser tomography. AO3k+SCExAO has been fully operational since October 2025, demonstrating very high stability on-sky, even in bad seeing conditions up to 2". Having two XAO in series allows us to deploy advanced wavefront control techniques optimized for high-contrast imaging (e.g. speckle nulling, EFC, Coronagraphic LOWFS, Fast and Furious) on the second-stage XAO loop, as AO3k by itself delivers high-contrast PSFs already. Areas of active ongoing research include use of photonic devices for spectrally dispersed interferometric sensing, PSF reconstruction from WFS telemetry, and non-linear sensors (focal plane and curvature). Recent upgrades to the computer infrastructure are aimed at supporting these R&D efforts and providing a rich collaborative environment for experimentation. In this paper, we will present on-sky high-contrast performance characterization of AO3k, AO3k+SCExAO, and on-sky demonstrations of novel wavefront control techniques to improve the contrast behind the coronagraph.
Read the full abstract

All authors Julien Lozi, Kyohoon Ahn, Vincent Deo, Olivier Guyon, Sandrine Juillard, Yoshito Ono, Garima Singh, Sébastien Vievard

Computation and Language

Which one is banana man? Evaluating vision-language models in multi-turn pragmatic interpretation

Flexible adaptation to context and shared pragmatic intuitions contribute to smooth human conversation. Iterated reference games—in which players repeatedly pick out novel referents using language—present a test case for agents' ability to perform context-sensitive pragmatic reasoning in multi-turn linguistic environments. We tested humans and vision–language models on their ability to identify the intended meaning of descriptions produced in iterated reference games, varying the provided context in terms of amount, order, and relevance. While humans performed well consistently, the models we evaluated could make use of prior context to interpret humans' referring expressions, but they struggled to build up the relevant context to interpret those expressions effectively. Our results suggest that the models we evaluated lack core skills needed for efficient linguistic collaboration.
Read the full abstract

Systems and Control

A Small-Gain-Like Framework for Large-Signal Stability Evaluation of Multi-Converter Systems

The increasing penetration of grid-connected converters has greatly altered the large-signal behavior of power systems. Their angle dynamics, shaped by diverse control algorithms and coupled through complex circuit interactions, pose substantial challenges to large-signal stability evaluation of multi-converter systems. To resolve this issue, the small-gain theorem, which characterizes the dissipation capability of interconnected systems (the small-gain-like property) via the individual dissipation capabilities of subsystems, is introduced to investigate transient angle motions in multi-converter systems. A large-signal model involving a set of interconnected relative angle motions is first developed, and the small-gain-like property is then established for multi-converter systems within certain angle limits, which further enables the construction of Lyapunov functions. Based on this, an ellipsoidal forward-invariant region is identified inside the angle-limit region, which serves as an effective estimate of the large-signal stability region for multi-converter systems. The method is further applied to a paralleled system and a four-converter system, where the large-signal stability boundaries are explicitly computed and subsequently validated through experiments. The proposed small-gain-like based large-signal stability evaluation method enables quantitative stability assessment in multi-converter systems with diverse control algorithms, which may provide a scalable framework for large-signal stability evaluation and parameter design in modern power systems.
Read the full abstract

All authors Qiannan Qu, Kaiwen Chen, Xin Xiang, Wuhua Li, Yunjie Gu

Dynamical Systems

Ergodic $\times p$-invariant measures on $\mathbb{T}^2$ with no dimension dropping projections

Fix an integer $p\geq 2$ and $0<s<1$. We construct an ergodic $\times p$-invariant measure $\mu$ on $\mathbb T^2$ of dimension $s$, such that every line projection preserves dimension, including when the corresponding projected IFS has exact overlaps. In fact, our measure assigns mass $O(w^s)$ to every planar tube of width $w$. The result is sharp at the endpoint $s=1$ as shown by Pyörälä, Shmerkin, Suomala and Wu (2025). In contrast, we show that every quasi-Bernoulli $T_p$-invariant measure with positive dimension admits a dimension-dropping projection.
Read the full abstract

Cryptography and Security

OASIS: Optimizing Attacker Sequences for Hard-Label Black-Box Text Attacks

Different attack methods follow different search trajectories, they succeed on different subsets of samples, whereas existing hard-label black-box text attacks mainly focus on improving individual attackers or manually combining them. We present OASIS, a method for optimizing attacker sequences in hard-label black-box text attacks. OASIS first performs a one-time bi-objective attack chain search over candidate sequences to balance attack success rate and perturbation, and then reuses the selected fixed global chain during attack chain execution. Experiments across multiple datasets, victim models, and large language models show that OASIS consistently outperforms strong standalone baselines and simple manually constructed chains. These results suggest that attacker composition is not merely an implementation choice, but a practical optimization target for improving hard-label black-box text attacks.
Read the full abstract

Computer Vision

MotionSync: Non-Causal Refinement of Causal Tracker for Label-Efficient 3D Perception

Three-dimensional box-and-track annotation is the cost bottleneck in autonomous-driving data engines, and the offline systems built to relieve it replace the online perception stack outright, so a team needing both regimes maintains and reconciles two. MotionSync makes the causal/non-causal boundary an explicit architectural seam instead. A strictly causal tracker, built on a strong published baseline and extended with innovation-driven uncertainty calibration, frame-rate-invariant kinematic association gates, and multi-hypothesis motion with learned mode selection, emits a valid online result. A non-causal pass then revises the buffered trajectories with Rauch–Tung–Striebel smoothing applied separately to pose, extent and yaw, physics-validated gap completion, and semantic pruning of ghost tracks against LiDAR point labels. The refiner never writes back, so one system serves both regimes and refinement's effect is a delta over an unaltered causal estimate. Used as an auto-labeller, a fixed 3D detector trained on 25% human labels plus MotionSync pseudo-labels reaches 96.9% of its full-supervision mean average precision (mAP) on Waymo, and at a 10% budget the non-causal pass accounts for +3.3 mAP/L2 over pseudo-labels from the same tracker's causal stage. Re-fitting the online tracker on its own refined output recovers 73% of the benefit of human supervision, while its causal output is worse supervision than no re-fitting at all. As a tracker MotionSync is at parity with the leading published offline entries on the headline metric and ahead of them on error composition, which is where a refinement pass can act at all: it reduces misses and fragmentations together, the signature of gap completion rather than of a tuned detector.
Read the full abstract

All authors Rahul Ahuja, Bala Murali Manoghar Sai Sudhakar, Shashwata Gupta, Venkatraman Narayanan, Varun Ravi Kumar, Senthil Yogamani

Mesoscale and Nanoscale Physics

Extreme Polarization of the Optical Gap and High-Energy Exciton Landscape in CrSBr

We reveal a strongly anisotropic excitonic landscape in monolayer and bulk-like CrSBr using optical absorption spectroscopy and $GW$-Bethe-Salpeter equation $\textit{ab initio}$ calculations. The direct absorptive determination of the lowest bright optical onsets i.e. $X_0^a$ and $X_0^b$ excitons for the two in-plane polarization eigenaxes yield an in-plane optical gap anisotropy of $470 \pm 15$ meV. This is the highest observed value for any material in the near-infrared-to-visible spectral region to the best of our knowledge. Energetically above, we identify multiple strongly polarized excitons spanning $1.25$ eV to $3.1$ eV selectively aligned along the two orthogonal axes. A resonance $X^-$, located $24$ meV below the $X_0^b$ progressively transfers oscillator strength to $X_b^0$, a behavior consistent with a coupled trion (Fermi-polarion)/exciton pair. Our experiments also provide polarization-resolved broadband dielectric functions of CrSBr. These results establish CrSBr as a strongly polarization-selective excitonic system and highlight its potential for polarization-selective optoelectronics enabled with its large optical-gap anisotropy.
Read the full abstract

All authors Sayantan Patra, Sourabh Jain, Bhumika Chauhan, Marie-Christin Heißenbüttel, Abhisek Saidarsan, Ranjuna M. K., Kseniia Mosina, Zdeněk Sofer, Michael Rohlfing, Thorsten Deilmann, Ashish Arora

Probability

Central limit theorem for Wasserstein projection - the case of convex order

The main focuses of the article are limit theorems of Wasserstein projection in the convex order which are useful for inference tasks. The main results rely on the establishment of dual attainment, stability and several useful observations. The first is a clean criterion to invoke the Wasserstein projection dualities on the classical cost $c=h(x-y)$. A new strategy was introduced to establish the dual attainment. Backward and forward dual transport are proved to enjoy a conjugate relationship. Sufficient conditions are given that ensure uniqueness of the optimal dual potential. These ingredients laid the groundwork for the proof of optimal dual potential stability, thereby permitting us to prove the central limit theorems under mild moment assumptions. Additional discussions show that the moment assumptions are sharp. These results also answered several open questions raised by Professor Benjamin Jourdain.
Read the full abstract

Computation and Language

TACS: Trajectory-Aware Candidate Selection for LLM Jailbreak Suffix Optimization

Gradient-based jailbreak suffix optimization methods typically update the suffix by retaining the candidate with the lowest current loss. We show that this seemingly natural design is fundamentally myopic: candidates that look better under the current-step proxy often fail to produce better jailbreak outcomes later in the search, revealing a form of selection-stage reward hacking. This suggests that candidate selection, rather than candidate generation alone, is a hidden bottleneck in suffix optimization. To address this issue, we propose TACS, a trajectory-aware candidate selection framework for jailbreak suffix optimization. Instead of selecting candidates solely by their immediate loss, TACS augments per-step evaluation with a trajectory-aware proxy and stabilizes selection with reference-policy regularization and a discriminator-estimated chi-squared correction, encouraging choices that remain effective beyond the current step. Experiments on HarmBench show that TACS consistently outperforms strong baselines under the same search budget, substantially improving attack success rates while exhibiting more stable optimization behavior throughout the search. Our findings highlight that mitigating selection-stage reward hacking caused by myopic candidate selection is critical for improving jailbreak suffix optimization.
Read the full abstract

2608.29564 · 30 Aug · We identified an error in the theoretical analysis, which affects the validity of the main conclusions of the manuscript. Since the current version does not adequately support this conclusion, we have decided to withdraw the paper

Machine Learning

HoopMind: A Real-Time Neural Game-Tree System for Opponent-Aware Possession Planning

School coaches prepare for opponents with game film and intuition. The analytics tools of professional teams stay out of reach. We ask how far public data can close this gap. Professional basketball is our case study, chosen for its data rather than the league. We fuse five public sources into one per-shot dataset of 4.23M shots over 21 seasons. The sources are shot locations, two play-by-play feeds, official matchup tracking, and player biometrics. Alignment across them is 99.5% to 100%. We also report two data pitfalls that are easy to miss. We then model a half-court possession as a sequential game. Shot values come from ShotNet, an embedding multilayer perceptron (MLP). On a held-out season it beats a zone-rate baseline and a logistic baseline, and its probabilities are well calibrated. A depth-limited expectimax search then solves the offensive decision tree, with branch-and-bound pruning to keep it real time. All training runs offline, so the online system stays light. A scouting planner and a playable simulator both run in a single browser page.
Read the full abstract

Machine Learning

Asynchronous Cooperative Online Learning for Multi-Robot Control under Computational Delays

Ensuring the safe operation of multi-agent systems (MASs) under uncertain environments is crucial for cooperative robotic, where external disturbances and inaccurate dynamic models can significantly compromise performance and reliability. To address this challenge, calibrated machine learning models, particularly Gaussian process (GP) regression, are extensively employed due to their interpretable performance quantification. As the interconnected communication of MASs facilitates cooperative learning, agents are able to enhance learning performance by exchanging local GP inferences with their neighbors and aggregating the received information via distributed GP strategies. However, variations in computational power and prediction tasks among agents inevitably lead to heterogeneous computational delays and differences in query points, which are often overlooked in existing aggregation methods. To overcome these limitations, this work proposes an asynchronous cooperative learning strategy that explicitly accounts for prediction accuracy, query point variations and delay effects. Additionally, a distributed control law based on an adjoint MAS is developed to ensure the desired control performance. Simulations on unmanned surface vehicles validate the effectiveness of the proposed approach, demonstrating substantial improvements in both learning and control performance compared to the state-of-the-art approaches.
Read the full abstract

All authors Xiaobing Dai, Zewen Yang, Wei Ren, Sandra Hirche

Logic

Ball codes: A coding characterization of Hausdorff and packing dimensions

We prove a purely classical, coding-theoretic characterization of Hausdorff and packing dimensions in \(\mathbb R^n\). A ball code assigns names to closed balls of \(\mathbb R^n\): it is a partial map from a prefix-free set of finite binary strings to balls. For every nonempty \(E\subseteq\mathbb R^n\), the Hausdorff dimension of \(E\) is the minimum, over all ball codes, of the supremum over \(x\in E\) of the lower asymptotic rate at which \(x\) can be described by names of balls containing it; the packing dimension is obtained in the same way from the upper rates. The Hausdorff characterization is a Euclidean counterpart of Ryabko's coding theorem for combinatorial sources. We obtain the packing characterization from the standard characterization of packing dimension by upper modified box-counting dimension, without encoding \(\mathbb R^n\) into a sequence space. We then derive the point-to-set principle of J.~Lutz and N.~Lutz by replacing a minimizing ball code with a rational one and storing the resulting countable codebook in an oracle.
Read the full abstract

Machine Learning

Which LLM for Which Work? Budgeted Model Allocation under Uncertain Evaluation

A company with a fixed artificial intelligence (AI) budget must decide which large language model (LLM) handles each recurring workload. What it lacks is the quality table, how well each model performs on each workload. Given that table, the decision is a multiple-choice knapsack problem and is routine to solve, so estimating it is the difficulty, and that estimation fails in two ways. Models are rarely compared on the same work, and the recorded score is usually a proxy rather than the outcome the company values. Causal and off-policy methods repair the first but condition on the second, while evaluator-validation methods estimate the second but stop short of the decision. Worse, buying more re-evaluation cannot settle the second: randomization governs which requests are scored, not how a score is produced, so the table stays uncertain however much evaluation is purchased. Yet the deployment decision may still be determined even when the table is not. We therefore ask whether one assignment stays optimal across every quality table consistent with the evidence. For the fixed-budget problem, this admits an exact two-solve certificate: solve once at the estimated table and once at a least-favourable table. Agreement certifies the assignment; disagreement identifies the model-workload pairs where further evidence can matter. We propose CASE (causal active sequential experimentation), which targets evaluation to those pairs and repeats the test as evidence accumulates. On a production log, the measurement failure is the larger of the two: correcting assignment exactly still leaves most of the loss, and randomized re-evaluation does not remove it. In our experiments, the available evidence often does not determine the assignment. On paid software tasks, better information about model quality yields more savings than further optimization of the assignment on the same estimates.
Read the full abstract

Systems and Control

Conformal Prediction Regions for Continuous-Time Trajectories under Random Sampling

Uncertainty quantification for continuous-time trajectories is a prerequisite in many safety-critical engineering domains. However, a major challenge in data-driven uncertainty quantification is that calibration trajectories are sampled only at discrete, often sparse, and random intervals. Standard conformal prediction methods typically fail to provide guarantees in between sampling times. In this work, we introduce a new technique to obtain valid conformal prediction regions for continuous-time trajectories that are sampled at discrete and possibly random times. To accomplish this goal, we make three contributions: (1) we provide an algorithm that leverages regularity properties of the underlying trajectories to obtain valid prediction regions in between samples, (2) we provide methods that estimate valid bounds on the aforementioned regularity properties from an additional high-frequency calibration dataset, and (3) we introduce and compare several algorithms to deal with random sampling times. Finally, we present experiments where we validate that our methods achieve valid coverage across the entire continuous trajectory.
Read the full abstract

All authors Joaquin Alvarez, Matteo Sesia, Jyotirmoy V. Deshmukh, Lars Lindemann

Number Theory

Luo's Spectral Large Sieve Inequality on Short Intervals

Let $u_j $ traverse an orthonormal basis of Hecke–Maass forms for $\mathrm{SL}_2 (\mathbb {Z}) $ with Hecke eigenvalues $\lambda_j (n)$ and Laplace eigenvalue $1/4+t_j^2$. In this paper, we consider the short-interval variant of the twisted spectral large sieve inequality of Luo for $ \lambda_j (n) n^{it_j} $ on the range $t_j \leqslant T$ and prove that the `Eisenstein–Kloosterman' cancellation discovered by Luo is effective on the interval $ T < t_j \leqslant T + M $ as long as $\sqrt{T} < M \leqslant T$. Moreover, our approach yields an improvement of the large sieve inequality of Luo.
Read the full abstract

Quantum Physics

Thresholdless dynamical instability in transverse $\mathcal{PT}$-symmetric scattering systems: The hidden role of bound states in the continuum

The stationary scattering properties of transverse parity-time ($\mathcal{PT}$) symmetric systems have been extensively studied, yet their dynamical stability, a prerequisite for any stationary description, remains largely unexplored. Here we uncover a thresholdless dynamical instability in such systems, driven by symmetry-protected bound states in the continuum (BICs). Without gain and loss, the up-down mirror symmetry of the transverse geometry generically protects a BIC, which manifests as the coalescence of an $S$-matrix pole and zero on the real axis of the complex wave-number plane. A Hermitian symmetry-breaking perturbation shifts the poles into the lower half-plane, converting the BIC into a resonance with a Fermi-golden-rule decay width. An anti-Hermitian $\mathcal{PT}$-symmetric perturbation instead reverses the sign of the second-order energy shift, driving the poles into the upper half-plane and producing a time-growing bound state with $\operatorname{Im}E=\gamma^{2}\Gamma_{V}/2+O(\gamma^{3})$, where $\gamma$ is the gain-loss strength and $\Gamma_{V}$ is the golden-rule coupling of the BIC to the continuum. The instability therefore sets in at arbitrarily small $\gamma$ whenever $\Gamma_{V}>0$. We confirm this mechanism in a three-site side-coupled model that is unstable despite possessing a unitary scattering matrix, and in a four-site rhombic model where transverse and longitudinal gain-loss placements yield vanishing and finite thresholds, respectively. These results establish a microscopic stability criterion and a design principle for stable $\mathcal{PT}$-symmetric scattering devices.
Read the full abstract

General Relativity and Quantum Cosmology

How much averaged null energy must an anisotropic bounce violate?

We show that in Einstein gravity a null-complete, non-static Bianchi I universe (homogeneous, spatially flat and anisotropic) violates the averaged null energy condition (ANEC) along every complete light ray outside at most four null directions, and along all of them if its total energy density, cosmological constant included, is ever positive. Along each complete ray the averaged null energy is at most $-J_p/(8\pi G)$, with equality whenever the integral converges, where $J_p$ is the squared expansion and shear accumulated by a thin bundle of light rays. With spatial sections $\mathbb{R}^3$ the achronal ANEC therefore leaves only Minkowski space, as for flat and open Friedmann models. Under the pointwise null energy condition, every null geodesic of a non-static Bianchi I universe is incomplete. The isotropic bounce of effective loop quantum cosmology has $J_p=1.478\,\omega_B H_{\max}$ (light frequency at the bounce times largest Hubble rate), and at the same critical density the cheapest ray of a strongly anisotropic bounce pays four to five times as much. A Casimir-supported turning point on a three-torus has a deficit of the same order, but at Hubble rates well below the Planck scale its light rays circumnavigate the torus and are not achronal, so the achronal ANEC does not apply to them.
Read the full abstract

Commutative Algebra

Truncations of the ring of number-theoretic functions, revisited

Let $K$ be a field containing $\mathbb{Q}$, let $\Gamma$ be the ring of all functions from the positive integers to $K$ under Dirichlet convolution, and let $\Gamma_n$ be its truncation to functions supported on $[1,n]$. In [Snellman, Homology Homotopy Appl. 2 (2000), 17-27; arXiv:math/9904143] it was shown that $\Gamma_n$ is a polynomial ring modulo a monomial ideal $I_n$ which is stable after reversing the order of the variables, and the Poincare-Betti series of $\Gamma_n$ was computed in terms of the numbers $C_{n,v}$ of minimal generators of $I_n$ of least support $v$. We prove that $C_{n,v} = \Phi(n,p_v)$, Legendre's sifting function: the number of integers in $[1,n]$ free of prime factors $\le p_v$. This identifies an invariant of a minimal free resolution with a classical object of sieve theory, and it refines to the grading by total degree, with the number of prime factors carried in a variable. As consequences we obtain: a proof of Conjecture 4.6 of the 2000 paper, which was left open there; the asymptotics $C_n \sim \pi(n)^2/2$ of the total number of minimal generators, showing that the lower bound $C_n \ge \binom{\pi(n)+1}{2}$ of that paper is asymptotically sharp, together with the exact order $C_n - \binom{\pi(n)+1}{2} \sim \frac{16}{3} n^{3/2} / \log^3 n$ of the error; and the identification of $C_n$ with the OEIS sequence A182843. We also record errata for the 2000 paper: one stated result is false, one worked example is miscomputed, and two proofs are incomplete. Corrected statements and complete proofs are given.
Read the full abstract

2608.29555 · 30 Aug · 25 pages, 2 figures. v2: adds Theorem 4', the graded refinement of the main theorem, and Section 4.2, an erratum to Example 4.5 of the 2000 paper. Written with substantial assistance from a large language model (Claude, Anthropic); see the "Disclosure of AI assistance" section. The author takes full responsibility for the results

Differential Geometry

Limiting Behavior of a Class of Hermitian Yang–Mills Metrics, II: Exponential Approximation

This paper is a sequel to [9], where the first-named author constructed a family of approximate Hermitian Yang–Mills metrics $H_{0,\epsilon}$ on stable rank-two holomorphic vector bundles arising from double spectral covers over the product of two one-dimensional complex tori. We prove that these approximate metrics give an all-order, exponentially accurate asymptotic description of the exact Hermitian Yang–Mills metrics in the large Kähler limit. More precisely, the mean curvature of $H_{0,\epsilon}$ decays exponentially in every $C^k$-norm. Moreover, if $H_{1,\epsilon}$ denotes the exact Hermitian Yang–Mills metric and \[ H_\epsilon=H_{0,\epsilon}^{-1}H_{1,\epsilon}, \] then, after normalization, for every nonnegative integer $k$, there exist positive constants $C_k$ and $c_k$ such that \[ \|H_\epsilon- Id\|_{C^k}\leq C_k e^{-\frac{c_{k}}{\epsilon}}. \] The main analytic difficulty lies in the global $C^0$-comparison. Obtaining $C^0$-estimates for the coupled nonlinear Hermitian Yang–Mills system is intrinsically difficult; moreover, the equation controls only the contraction of the curvature, and hence only certain combinations of second derivatives, whereas one needs global control of the full matrix-valued metric.
Read the full abstract