Peer Revenue Now accepting positions

Papers

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

All of arXiv

Friday 6 November

Software Engineering

Tracybot: Preserving Developer Intent by Tracing Agent-Generated Code to Prompts

Large language models (LLMs) are now widely used in software development, and LLM-based coding agents are given increasing autonomy. With coding agents, many design decisions are made in the conversation between the developer and the agent. Typically, the developer starts a conversation with an agent and states their intent, the agent proposes options if needed, and the developer plans the change accordingly. Then, the resulting code is committed. However, in such scenarios, the decisions and the intent behind the changes remain in the conversation, with no link to the code they produced. In this paper, we present Tracybot, an open-source tool that integrates with VS Code, Git, and coding agents to keep track of relevant conversations in an LLM-assisted development process. Tracybot captures each change as a Tasklet which consists of optional planning prompts used to plan a change, together with a build prompt that produces it. Tracybot has two main features: (i) it maps line-level code changes made by coding agents to the Tasklets that produced them, and (ii) it allows the developer to export a Tasklet as a decision record when needed, for example when the decision concerns the architecture of the project. We developed the tool over two iterations of a design science study, while focusing the evaluation on the usability and user-friendliness of the tool. In a focus group (7 participants) and a survey (24 participants), we found that Tracybot was perceived as useful for activities like code review, and most participants found our Tasklet design and presentation to be a clear way to represent the developer's intent. Tool: https://github.com/TracyTeam/tracybot. Video: https://youtu.be/Vg_a1F5eO04.
Read the full abstract

All authors Răzvan Albu, Aliaksei Khval, Maxine Orlova, Danis Music, Filipe Araújo Veras Rosa, Lirong Yi, Ranim Khojah

Human-Computer Interaction

Sensed but Not Told: Hidden Physiological Adaptation Shapes LLM Persuasion

Wearable sensors, including those embedded in VR headsets, can provide LLM-based chatbots with physiological signals that users do not explicitly communicate. We tested whether such hidden input changes persuasion. In a $2\times2$ within-subjects study ($N=24$), a chatbot argued against participants' self-selected opinions with or without turn-level heart-rate and pupil-diameter features; sensing was disclosed halfway through the study. Before disclosure, sensing increased perceived persuasiveness by 1.42 points. Across the study, sensing was associated with a 0.57-point greater shift toward the chatbot's position and 3.20 times the odds of any movement toward it. Persuasiveness ratings were lower after disclosure, although disclosure was confounded with session order. Dialogue logs suggested a more declarative interaction style when sensing was on: chatbot replies contained question marks less often, and conversations contained fewer turns. These findings motivate making biosignal-driven adaptation inspectable and providing a genuinely non-sensed mode.
Read the full abstract

All authors Xiaoyan Wei, Yutong Qu, Yutong Li, Haiyi Li, Wei Emma Zhang, Hsiang-Ting Chen

Number Theory

Curves with absolutely irreducible Tate modules over $p$-adic fields

Over every number field, curves whose Jacobians have absolutely irreducible $\ell$-adic Tate modules exist in every positive genus and can be chosen hyperelliptic. Over $p$-adic fields, however, inertia and Frobenius can exclude entire genera. We classify the genera that occur for $\ell\ne p$ in terms of the residue characteristic and the cardinality of the residue field. For odd $p$, we also classify the genera occurring for hyperelliptic curves. These genera have relative density zero among those occurring for arbitrary curves, yielding Jacobians whose isogeny classes contain no hyperelliptic Jacobian. For odd $p$ and even residue degree, we obtain further restrictions on the genus when the curve itself is required to have potentially good reduction. Over number fields, we obtain uniform bounds from absolute irreducibility of the Tate module at two finite places of distinct residue characteristics different from $\ell$. The dimension of the abelian variety alone bounds the number of isomorphism classes in its isogeny class and the least degrees of isogenies between them.
Read the full abstract

Data Structures and Algorithms

Precedence Constraint Scheduling on a Fixed Number of Machines

We give a deterministic exact algorithm for $P_m\mid\mathrm{prec},p_j=1\mid C_{\max}$ for every fixed $m\ge3$. The algorithm constructs an optimum schedule in $$ O(L)+m^{O(m^2)}\, n^{6m^2+13m+1} \, {\log n} $$ elementary RAM operations, where $n \ge 2$ is the number of jobs and $L$ is the binary input length. Its word size is $O(m^2 \log n)$ bits. An implementation with a fixed number of tapes uses $O(L)+m^{O(m^2)} \, n^{12m^2+26m+2} \, \log^3 n$ steps. For three machines the respective polynomial degrees are $94$ and $188$, apart from the displayed logarithmic factors.
Read the full abstract

Statistical Mechanics

Mathematical analysis of the Markovian Mpemba effect

The Mpemba effect is a counterintuitive phenomenon in which two equilibrium systems at different temperatures, say hot and warm systems, when quenched at the same bath temperature, for some pair of hot and warm temperatures, the hot system comes to bath temperature faster than the warm system. In recent years, studies on the Mpemba effect for Markovian systems under external potential, which are continuous-time, continuous-state-space Markov processes, have been carried out theoretically and experimentally by resorting to pseudo-distance measures which are not true distance metrics. Moreover, maximum crossover time is given for the Mpemba effect for all monotone distance measures or functions (which include true distance metrics and pseudo-distance measures). However, no mathematical analysis is carried out for the mere possibility of the presence of the Mpemba effect for Markovian systems under an external potential with a true distance metric. In this paper, by using a true distance metric which arises naturally from the weighted Hilbert space in which the adjoint of the generator of the Markov process is a Hermitian operator, we have shown the mathematical analysis of the possibility of the presence of the Mpemba effect by using Poincaré inequality and the geometry of Hilbert space.
Read the full abstract

Quantum Gases

Motion-Induced Polaron Resonance beyond the Orthogonality Catastrophe

We investigate the spectroscopic response of a repulsively interacting, infinitely heavy impurity moving at constant velocity $v$ through a one-dimensional ideal Fermi gas. Using an exact functional-determinant approach, we calculate the Ramsey response and spectral function, fully accounting for many-body particle-hole excitations. We uncover a strongly nonmonotonic dependence on impurity velocity: decoherence is enhanced around the Fermi velocity, whereas sufficiently rapid motion restores a narrow, long-lived polaron resonance. A long-time wave-packet analysis yields analytical expressions for the polaron energy and decay rate over the full velocity range and quantitatively captures the numerical results. At high velocity, the polaron energy approaches the mean-field value, the decay rate vanishes as $1/v$, and the spectral weight approaches unity. Thus, rapid prescribed motion provides a dynamical route to suppressing the orthogonality catastrophe and restoring polaron coherence even for an infinitely heavy impurity, where recoil is entirely absent.
Read the full abstract

Operator Algebras

Isomorphisms and representations of noncommutative function algebras

Let $V$ be a noncommutative (NC) variety in a NC operator ball. We study the dual operator algebra $H^\infty(V)$ of bounded NC functions on $V$, and the operator algebra $A(V)$ obtained as the closure of NC polynomials in $H^\infty(V)$. We identify the space of weak-$*$ continuous finite-dimensional representations of $H^\infty(V)$ as the similarity envelope $\widetilde{V}$ of $V$, and equip $\widetilde{V}$ with a metric which turns $\widetilde{V}$ into a complete invariant for $H^\infty(V)$, in the sense that $H^\infty(V)$ is weak-$*$ continuously isomorphic to $H^\infty(W)$ if and only if $\widetilde{V}$ and $\widetilde{W}$ are NC biholomorphic via a bi-Lipschitz map. Similar results are obtained in the completely bounded category as well as for the algebras $A(V)$ when $V$ is assumed homogeneous. In the case of homogeneous varieties $V$ and $W$, we obtain a sharper classification result, by which $H^\infty(V)$ is (completely) boundedly isomorphic to $H^\infty(W)$ if and only if there is a linear map which gives a bi-Lipschitz bijection between $\widetilde{W}$ and $\widetilde{V}$. The reduction to the linear case in the generality we work in depends on showing that the existence of maps implies the existence of origin-fixing maps, which we show using a novel “two-circle trick" of independent interest. In the important case where the varieties are the full NC operator balls we find a striking rigidity phenomenon: if $A(B_E)$ and $A(B_F)$ are completely boundedly isomorphic then $E$ and $F$ are completely isometric. We also present a variety of examples illustrating the results and their sharpness, showing how several results from the literature fail to generalize to the setting studied here.
Read the full abstract

All authors Marcel Scherer, Orr Shalit, Eli Shamovich, Mansi Suryawanshi

Machine Learning

k-Volume: a Unified Active Learning Strategy

In the classification context where revealing the label of a sample is costly, we propose a new Active Learning (AL) strategy named k-Volume (kVol). It randomly selects a sample to be labeled according to the volume of the smallest ball containing its k nearest neighbors. kVol is highly versatile: it can be applied either in the input space or in the output space, and it enjoys the desirable properties of the three main families of AL strategies: density, diversity, and (when performed in the output space) uncertainty. From a theoretical perspective, we establish that kVol is asymptotically equivalent to uniform sampling over the support of the (input or output) data distribution. Based on experiments on various synthetic and real-life tabular datasets and images, we observe that kVol is surprisingly stable: whatever the classifier or the experimental setup, it still lies among the best AL strategies unlike other competitors that fail dramatically. In particular, kVol outperforms state-of-the-art competitors such as CoreSet, TypiClust or Uncertainty on several real-life datasets, at a low computational cost.
Read the full abstract

All authors Eliot Tron, Alain Celisse, Christophe Denis, Alice Le Brigant

Computer Vision

A Bridge from Paste to Place: Physics-Guided One-Step Image Composition via Object-Scene Interaction

Realistic image composition makes a pasted object appear naturally present in a scene while preserving the pasted layout and fine foreground and background details. Reference-based insertion methods preserve overall object identity but can alter pose and fine details, often at substantial computational cost. Foreground harmonization and shadow generation address individual aspects of this task, but neither alone ensures natural integration. An efficient, unified solution remains an open challenge, and large-scale, high-quality datasets and benchmarks for this task are lacking. To address these challenges, we introduce Paste2Place, an end-to-end, physics-guided flow-based model that learns a direct latent bridge from pasted inputs to realistic composites, guided by image-space corrections analytically derived from predicted physical components of object and scene appearance changes. To support training and evaluation, we introduce P2P-Data with 86K manually reviewed synthetic and real-scene training pairs, including dense physical annotations from our Blender rendering and validation pipeline for the synthetic subset. P2P-Bench combines paired compositions and manually pasted inputs from collected images into 1.5K examples for evaluating composition under user-chosen placements. Extensive qualitative and quantitative experiments demonstrate that Paste2Place achieves substantial advantages in image composition quality and inference speed across diverse scenarios.
Read the full abstract

All authors Hongzhen Li, Youwei Pan, Yingfang Zhu, Sen Zhu, Fengjie Zhu, Leilei Cao

Number Theory

The exceptional set of Goldbach numbers among polynomial values

Let $\phi$ be a fixed polynomial with integer coefficients, positive leading coefficient and degree $k\geq2$. We prove that the values $2\phi(n)$, $1\leq n\leq N$, can be written as the sum of two primes, apart from at most $N^{\alpha_k+o(1)}$ exceptional indices, where $\alpha_k<1$. For quadratic polynomials the exponent is $1/2$, for cubic polynomials $5/8$, for quartic polynomials $3/4$, and for $k>4$ it is $1-1/(k+1)$. The proof combines a fourth-moment estimate over primes with mean-value estimates over arbitrary subsets of polynomial sequences.
Read the full abstract

All authors I.Althöfer, D.Chen, A.Perelli, J.Pintz

Artificial Intelligence

Against All Likelihood

Maximum likelihood is the dominant training principle for large language models, yet likelihood is neither necessary nor sufficient for truth. Unlikely events happen every day, and frequently repeated statements can be false. Drawing on a philosophical tradition that ties objectivity to invariance rather than probability (from Nozick's account of objective facts and Wigner's principles of invariance to the structural realism of Poincaré and Cassirer), we argue that the truth of a statement should be judged not by its probability under a single distribution, but by its invariance across relevant transformations of the environment. We connect this view to machine learning literature on out-of-distribution generalization and invariant learning, sketch how invariance could be tested at inference time with language models, and discuss implications for continual learning.
Read the full abstract

Signal Processing

DNN-Driven Proactive UAV Trajectory and Beam Steering for Blockage-Resilient Communications

This paper proposes Autonomous UAV Communication (AutUAVCom), a proactive communication system that jointly adapts the flight trajectory and RF beam steering to minimize UAV energy consumption while satisfying secrecy and reliability requirements for rescue teams operating in dynamic, cluttered, and GPS-denied environments. Unlike existing modelbased proactive approaches, AutUAVCom employs a deep neural network-based Tracking and Prediction (T&P) module that combines onboard sensing with spatiotemporal and body-orientation information to track users, localize blocking objects, and predict future trajectories and potential beam blockages. Building on the T&P module, we formulate a constrained optimization problem and develop a two-step iterative numerical approach. Monte Carlo simulations show that, relative to a reactive system, AutUAVCom achieves 10% and 19% gains in energy efficiency and channel secrecy, respectively, while reducing the outage probability by 85%, under a mean blockage arrival interval of 120 seconds and a mean blockage duration of 20 seconds.
Read the full abstract

All authors Abdallah S. Ghazy, Hussein A. Ammar, James Bayes, Francois Chan

Information Theory

Universal Decoding for Gaussian MIMO-ISI: Spherical Ensembles and Symmetric Neighborhoods

We construct a universal decoder for cyclic Gaussian multiple-input multiple-output channels with intersymbol interference and unknown channel taps and spatial noise covariance. Codewords are drawn independently and uniformly from a sphere of fixed total energy. Under stated bounds on frequency variation, the decoder attains the ensemble-average error exponent of informed maximum-likelihood decoding for fixed channels and channels with growing memory. We approximate the frequency response by a constant matrix within each frequency block. For this auxiliary channel, every matched log-likelihood is affine in a common sufficient statistic. At fixed received output, we compute the statistic density of an independent competitor. The decoder selects the candidate with the smallest statistic density, equivalently minimizing a weighted sum of residual log-determinants. The density and its support volume give an upper bound on the universal competitor probability. Affinity ensures that at least half of a suitable symmetric neighborhood of the transmitted statistic has matched likelihood no smaller than that of the transmitted word. Integrating a density lower bound over this half gives the matched competitor bound. These bounds establish equality of error exponents, which a Gaussian channel comparison transfers to the original channel. The same decoder attains the matched exponent for independent frequency blocks with prescribed energies satisfying the stated conditions. For energies proportional to block length, an additional gain condition ensures that this exponent is at least that of the total-energy spherical ensemble.
Read the full abstract

Artificial Intelligence

Exploring Weight-Space Programming: Capabilities, Generalization, and Limitations

Weight-Space Programming (WSP) enables reasoning agents to directly inspect and modify neural network parameters, offering an alternative to gradient-based model adaptation. However, its capabilities, cross-task generalization, and practical limitations remain insufficiently understood. We systematically investigate WSP through 16 editing trajectories across seven benchmarks and two model scales (Qwen3.5-0.8B and Qwen3.5-9B). Our findings reveal three key insights. First, WSP can achieve substantial performance gains through highly localized interventions. On full held-out evaluations, accuracy improves by up to 29.45, 4.42, 36.72, and 17.11 percentage points on MATH-500, GPQA Diamond, IFEval, and SciBench, respectively. Notably, modifying only 1,023 parameters yields a 29.45-point improvement. Second, WSP exhibits promising but uneven cross-task generalization. A Math-edited checkpoint improves GPQA Diamond, IFEval, and SciBench by 3.03, 12.18, and 20.10 points, respectively, without further editing, while leaving SciCode performance unchanged. Third, effective editing remains challenging and inconsistent. Outcomes vary substantially across trajectories, and six of the 16 trajectories retain the original weights when candidate improvements cannot be reliably validated. Together, these results highlight WSP's potential for highly localized model adaptation while revealing important challenges in intervention discovery, reliability, and cross-task transfer. Homepage: https://wsp-scope.github.io/ GitHub: https://github.com/WSP-Scope/WSP-Scope-Project
Read the full abstract

All authors Junru Hu, Cheng Tang, Wei Li, Hongze Zhu, Chun-Mei Feng, Qi Zhang, Yirong Chen, Ming Hu

Theoretical Economics

Decomposing the Cost of Complexity Under Risk

We study preferences under risk when the probability structure of a lottery can affect value separately from its final prizes. Our first result characterizes perturbed expected utility representation in which lottery value separates into expected utility over prizes and an additive perturbation term depending only on the probability tree. The key behavioral restrictions are branch cancellation, which separates prize evaluation from probability structure, and within-tree dominance, which captures that evaluations within the same probability tree are simple and follow dominance. Strengthening branch cancellation to lottery cancellation, which extends the invariance requirement from common prizes to common sublotteries, characterizes additive separability of the perturbation across branches. We also characterize when the tree dependent perturbation term depends only on the number of distinct possible prizes: under weaker continuity requirements, adding same-support independence yields a support size representation for nondegenerate lotteries, and a certainty equivalent condition extends it to degenerate lotteries. Finally, concavity gives the perturbation term its interpretation as a complexity cost.
Read the full abstract

Computer Vision

SPSC: Zero-Shot Cross-Subject EEG-to-Image Reconstruction via Semantic Preservation and Subspace Calibration

Pretrained vision-language models and diffusion generators have driven rapid progress in brain-to-image generation, while EEG-based reconstruction is still constrained by low signal-to-noise ratio and cross-subject semantic shifts. Zero-shot cross-subject reconstruction poses even greater challenges: EEG semantic embeddings learned from source subjects deviate from the manifold of image embeddings of the unseen target subject, particularly when neither target labels nor target image embeddings are available. To address this issue, we introduce the Semantic Preservation and Subspace Calibration (SPSC) framework, which first uses Object Mask Contrastive Regularization (OMCR) with object identity for backend training, and then constructs Subspace Candidate Gate (SCG) to achieve label-free semantic correction for target subjects during inference. Specifically, OMCR regularizes the conditioning space with object-level positive samples to preserve object-level feature neighborhoods within the reconstruction prior and low-level branch, while SCG exploits training image prototypes and batch statistics of unlabeled target EEG to select candidates for each task within subspaces of image prototypes. Under the leave-one-subject-out (LOSO) setting across ten subjects on the THINGS-EEG dataset, SPSC improves the cross-subject reconstruction from 0.552 to 0.713 on CLIP and from 0.612 to 0.745 on Alex.5. These significant gains indicate that the semantically calibrated conditions can better support generation for the target subjects under label-free inference.
Read the full abstract

All authors Yutang Lu, Shuai Jiang, Fei Su, Zhicheng Zhao

Quantum Physics

Enhanced-sensitivity optical sensing of magnetic fields using postselected amplification and continuous-wave power recycling

Given the importance of precision weak magnetic field measurement, and noting that postselected amplification has not yet been well applied in this area, in this work we propose to combine the technique of postselected amplification with continuous-wave power recycling for enhanced-sensitivity optical sensing of weak magnetic fields. Based on rigorous derivations, we obtain compact expressions for the total postselection probability and the signal-to-noise ratio. We also analyze the effects originating from various non-ideal factors. In contrast to pulse recycling, continuous-wave recycling has lower control requirements and is more experimentally feasible.
Read the full abstract

All authors Yazhi Niu, Jialin Li, Lupei Qin, Xin-Qi Li

Optimization and Control

Approximation Algorithms for Right-Hand-Side Parametric Linear Programming Problems

In parametric linear optimization problems, the objective function or the feasible set depends linearly on one or several real-valued parameters. The goal is to provide an optimal solution set, i.e., a set containing an optimal solution for each non-parametric problem induced by a particular choice of parameter values. In this paper, we consider right-hand-side single-parametric linear programming problems (RHS-PLPPs) that are usually very difficult to solve exactly. An optimal solution set of minimum cardinality can be infinite, while a representation by optimal bases may still require super-polynomially many different bases. Hence, we develop approximation methods for RHS-PLPPs that compute approximation sets of polynomially bounded cardinality, each containing, for every parameter value, a feasible solution whose objective value is at most a prescribed multiplicative factor times the optimum. Our first approach constructs an approximation set using a greedy staircase procedure and yields an FPTAS under general sufficient conditions. We show that these conditions hold for some important classes of RHS-PLPPs. For monotone instances, in which the feasible sets are nested as the parameter varies, our algorithm yields an approximation set of minimum cardinality. Beyond monotonicity, we establish near-minimum cardinality guarantees under more general sufficient conditions. Finally, we introduce a dual approximation algorithm that returns solutions with objective value no greater than the optimum while allowing a controlled multiplicative violation of the parameter-dependent right-hand sides.
Read the full abstract

All authors Levin Nemesch, Stefan Ruzika, Clemens Thielen, Alina Wittmann

Artificial Intelligence

KineCouple: Substrate-conditioned residual readout for enzyme kinetic prediction

Accurate prediction of enzyme kinetic parameters can support enzyme engineering and metabolic modeling by complementing experimental characterization. Existing prediction methods often pool residue-level protein features before combining them with substrate features. Such substrate-independent pooling cannot adapt residue weighting to the substrate during aggregation, potentially diluting information relevant to a specific enzyme-substrate pair. Since the contributions of individual residues to catalysis can depend on the substrate, substrate-conditioned aggregation offers a way to emphasize residue features relevant to the prediction. We introduce KineCouple, which uses substrate queries and residue-level keys and values in cross-attention, allowing the substrate to guide residue selection before aggregation. The model preserves the global protein representation and adds a gated correction defined as the difference between the conditioned readout and a uniform readout. ProtT5 and MolT5 remain frozen, and separate models predict kcat and Km on a logarithmic scale. In pooled out-of-fold predictions on the CataPro benchmarks, KineCouple reduces kcat RMSE from 1.333 to 1.330 and increases PCC from 0.492 to 0.496 relative to reproduced CataPro. For Km, RMSE decreases from 0.998 to 0.997 and PCC increases from 0.633 to 0.634. Spearman correlation also improves for both tasks.
Read the full abstract

All authors Chun Kang, Yunbo Zhang, Fanhu Zeng, Jian Liu, Siwei Deng, Lei Sha

Robotics

Can LLMs Debug Robots by Editing Their Weights?

A language model can repair a robot policy's lexical instruction grounding by editing just 65 token-embedding rows. We bring weight-space programming to robot foundation policies: GPT-6 Astra reads weights and uses a CPU workspace and budgeted simulator feedback, without gradient updates or new demonstrations. Three independent sessions on $\pi_{0.5}$ all improve success on 765 held-out paraphrases; the best raises success from 86.3% to 92.0% (48 repairs, four regressions; exact paired $p=1.31\times10^{-10}$). Among the 335 instructions containing edited tokens, success rises from 74.3% to 87.5%; the other 430 episodes have zero outcome changes. Matched random-direction edits do not help. Full replacement using the editor's own lexicon explains much of the gain but produces more regressions; its direct difference from the editor is not significant at 0.05. Across 13 further robot sessions, the editor declines 12 candidates; the sole retained edit does not show a significant held-out gain. Separately, benchmark metadata lowers FastWAM success from 70.0% to 48.6%, while the evaluated VLA's aggregate rate is unchanged. The evidence supports precise lexical repair, not general acquisition of missing robot capabilities.
Read the full abstract

Cryptography and Security

Cutpurse: Decomposing Explanation-Guided Model Extraction on Tabular Data

Explanation APIs are often treated as increasing model-extraction risk, but an end-to-end attack comparison cannot show whether an explanation reveals model-specific information or merely supplies a convenient next-query coordinate. We study this distinction for explanation-guided extraction of tabular classifiers. We separate an explanation into an attribution channel, which ranks influential features, and a coordinate channel, which supplies candidate feature values for perturbing those features. We then test whether the useful bin edges are privileged to the service or reproducible from the adversary's own data. Across six datasets and five target families, using attribution rankings to choose which features to perturb provides no systematic extraction advantage. Lime's discretization bin edges can improve fidelity by up to 41 percentage points on continuous features, but those edges are quantiles of the explainer's background data rather than properties of the target model. In controlled comparisons, an attacker-side grid reproduces the useful bin-edge gains without querying an explanation API. We use these findings to derive Cutpurse, which combines a self-computed grid with diverse query generation and boundary search. Against a service that withholds bin edges, Cutpurse improves fidelity by 16.3 percentage points on continuous-feature targets. It also exceeds the undefended published attack by 7.0 percentage points. Thus, withholding default Lime bin edges alone does not prevent effective extraction under the in-domain auxiliary-input, probability-output, and same-family surrogate assumptions studied here. Our results concern default Lime and Shap configurations for tabular classifiers. They do not establish the same conclusion for other explainers, target classes, or data-free attackers.
Read the full abstract

All authors Seyma Selcan Magara, Hanru Zhang, Fatme El Zahraa Hamad, Nico Pfeifer, Mete Akgün

Data Structures and Algorithms

Pairwise Reconnection for Cost-Distance Steiner Trees

The Uniform Cost-Distance Steiner Tree Problem asks for a rooted tree that minimizes the connection cost plus a weighted sum of root-to-terminal distances. We develop an algorithm that constructs two candidate trees: one based on direct connections to the root and another that pairs selected vertices and connects each pair to the root. The algorithm selects the tree with the smaller objective value. The resulting randomized algorithm achieves expected approximation factors of $1.698$ in the spanning case and $2.041$ in the general case, and it can be derandomized in polynomial time. The pairwise matching analysis charges part of the augmentation cost directly to the connection and delay costs of an optimal solution, rather than only to the separated lower bound used in previous analyses. This allows the $1.698$ factor to fall below the maximum multiplicative gap of $1+1/\sqrt{2}$ between the optimum value and that lower bound. We give instance families with approximation-ratio lower bounds approaching $1+1/\sqrt{2}$ for cut-and-reconnect algorithms that start from a minimum spanning tree and $3/2$ for the present algorithm. Finally, we show that the Uniform Cost-Distance Steiner Tree Problem and the Steiner-tree version of the Cable-Trench Problem are approximation equivalent up to an arbitrarily small additive loss.
Read the full abstract

Machine Learning

Calibrated Uncertainty Quantification on TDC ADMET: What Can Be Resolved Across 22 Datasets?

Absorption, distribution, metabolism, excretion, and toxicity (ADMET) models often rely on small, heterogeneous datasets, yet their predictions may determine which compounds proceed to experimental testing. Useful evaluations must therefore assess both predictive performance and the quality of reported probabilities or intervals. We ask which common uncertainty-quantification methods perform best on the 22 Therapeutics Data Commons (TDC) ADMET datasets and what this benchmark can resolve. We compare eleven calibrated pipelines using nested five-fold cross-validation across ten chemical partitions and five model seeds. Ensemble Monte Carlo (MC) dropout has the best average rank for TDC performance, Brier score, and 90% interval score. For TDC performance, however, it is not separated from the other ensemble pipelines; for Brier and interval score, LightGBM is also not ruled out as best. LightGBM wins more individual datasets than any other method for TDC performance and Brier score, but performs poorly on several endpoints. Smaller evaluations can identify a different leader and often fail to recover the broader comparison. These results make LightGBM and an ensemble neural model complementary baselines. Comparative claims should be checked across multiple chemical partitions using statistical analyses matched to the question being asked.
Read the full abstract

Artificial Intelligence

Enabling Privacy-Preserving Natural Language Access to Industrial Smart Meter Data

Energy management officers and grid operators increasingly rely on smart meter data for anomaly detection, performance monitoring, and strategic decision-making. However, extracting actionable insights from large-scale, heterogeneous telemetry databases requires advanced SQL and data engineering expertise, creating a critical bottleneck for non-technical personnel. Furthermore, the strict data privacy regulations and security constraints inherent to critical energy infrastructure often prohibit the routing of sensitive consumer data through cloud-based Large Language Models (LLMs). To address these challenges, we present OpenMeter, a schema-agnostic Multi-Agent System (MAS) designed to translate natural language queries into executable code and human-readable insights. This architecture lowers the technical barrier for day-to-day utility operations by empowering non-expert personnel to interact with complex databases through intuitive natural language dialogue, transforming raw telemetry into a highly valuable tool for routine grid workflows and strategic decisions. Multi-scenario ablation studies over this environment demonstrate that full multi-agent orchestration shows its largest observed accuracy gains on resource-constrained local hardware (+6.1 pp EX-R over the strongest single-agent configuration); at frontier scale, individual components retain value but should be enabled selectively, indicating that the cost of full orchestration is justified specifically where privacy demands local execution.
Read the full abstract

Machine Learning

Mitigating Covariate Shift in Behavioral Cloning via Robust Stationary Distribution Correction

We consider offline imitation learning (IL), which aims to train an agent to imitate from the dataset of expert demonstrations without online interaction with the environment. Behavioral Cloning (BC) has been a simple yet effective approach to offline IL, but it is also well-known to be vulnerable to the covariate shift resulting from the mismatch between the state distributions induced by the learned policy and the expert policy. Moreover, as often occurs in practice, when expert datasets are collected from an arbitrary state distribution instead of a stationary one, these shifts become more pronounced, potentially leading to substantial failures in existing IL methods. Specifically, we focus on covariate shift resulting from arbitrary state data distributions, such as biased data collection or incomplete trajectories, rather than shifts induced by changes in dynamics or noisy expert actions. In this paper, to mitigate the effect of the covariate shifts in BC, we propose DrilDICE, which utilizes a distributionally robust BC objective by employing a stationary distribution correction ratio estimation (DICE) to derive a feasible solution. We evaluate the effectiveness of our method through an extensive set of experiments covering diverse covariate shift scenarios. The results demonstrate the efficacy of the proposed approach in improving the robustness against the shifts, outperforming existing offline IL methods in such scenarios.
Read the full abstract

All authors Seokin Seo, Byung-Jun Lee, Jongmin Lee, HyeongJoo Hwang, Hongseok Yang, Kee-Eung Kim

Solar and Stellar Astrophysics

The OpenMULTI3D stellar analysis package. A 3D non-LTE spectrum synthesis code for AFGKM-type stars within the DISPATCH framework

Upcoming large spectroscopic surveys such as WEAVE, 4MOST, and SDSS-V will provide millions of high-resolution stellar spectra. This motivates the development of efficient codes to perform three-dimensional (3D) non-local thermodynamic equilibrium (non-LTE) radiative transfer (RT) calculations, in order to meet the requirements on efficiency and data analysis quality set by these surveys. We present a new open access 3D non-LTE spectrum synthesis code OpenMULTI3D, based on its predecessor MULTI3D, which was never released. This code can be used to compute large grids of spectra of AFGKM-type stars and will aid the exploitation of large-scale spectroscopic surveys such as 4MOST. We summarise the main input micro- and macro-physics of the code, describe its numerical implementation and provide selected test cases. The outputs of the code are compared with standard results in the literature, including the NLTE departure coefficients in 1D and 3D, the 3D NLTE abundance corrections, 3D synthetic spectral energy distributions for three types of stars, and 3D NLTE Balmer-line profiles of a metal-poor red giant. We also demonstrate how the code can be used to compute time-dependent spectra of stars, which may be relevant in precision stellar physics and analyses of exoplanet spectra. OpenMULTI3D offers an accessible and efficient platform for high-fidelity stellar analysis in the era of large-scale spectroscopic surveys.
Read the full abstract

All authors Richard Hoppe, Maria Bergemann, Philipp Eitner, Åke Nordlund, Jorrit Leenaarts, Andrius Popovas, Nicholas Storm

Audio and Speech Processing

AISHELL8-FISHEYE: A Fisheye Audio-Visual Dataset for Target Speaker Extraction with Distortion-Aware Baselines

Fisheye cameras offer wide visual coverage but distort facial and lip geometry, challenging audio-visual target speaker extraction. We introduce AISHELL8-FISHEYE, a 47.03-hour corpus of 94 Mandarin speakers with synchronized audio and paired fisheye and normal-camera videos. Speaker-disjoint splits and synthetic two-speaker mixtures support reproducible evaluation under matched acoustic conditions. We also establish a baseline that combines a lip-reading encoder with a frozen distortion-aware branch transferred from fisheye head pose estimation. On fisheye video, adding this branch to an adapted lip encoder improves scale-invariant signal-to-distortion ratio improvement from 10.47 to 11.84 dB. The branch does not improve performance with a frozen lip encoder, highlighting the importance of visual adaptation. A substantial gap to paired normal-camera inputs remains. The corpus and baseline resources support further research on speech extraction from distorted visual cues.
Read the full abstract

All authors Peijun Yang, Zhan Jin, Juan Liu, Hui Bu, Xiaoyi Qin, Ruyi Gan, Hao Wang, Ming Li

Geometric Topology

HOMFLY homology for oriented tangles

We introduce an extension of Khovanov and Rozansky's categorification of the HOMFLY polynomial to arbitrary oriented tangles. Unlike previous such extensions, ours is invariant under the full suite of oriented Reidemeister moves, including the "misoriented" Reidemeister IIb. For oriented links $\mathcal{L}$, our construction recovers Khovanov and Rozansky's construction while removing the preliminary step of presenting $\mathcal{L}$ as a braid closure.
Read the full abstract

Information Theory

Anchored Iterative Decoding of Reed-Solomon Product Codes for High-Rate Magnetic Tape Storage

We present a decoding algorithm for a Reed–Solomon product code targeted at the demanding reliability requirements of modern magnetic tape storage. For a code $RS(256, 224)\times RS(175, 173)$ in a BSC we achieve post-decoding bit error rate (BER) of $10^{-20}$ at $\approx 5.3\times 10^{-3}$ raw BER while maintaining the common requirement to detect and correct 4 out of 32 dead tracks. We resolve two common problems of miscorrections and stall patterns that arise under such a high raw BER via simplified versions of anchor decoding and binary image analysis. The highest efficiency is achieved by performing an unlimited iterative decoder, which performs decoding iterations until no changes are made. We give an empirical analysis of the average number of iterations: below the raw BER of $5\times 10^{-3}$, four half-iterations are performed on average while improving substantially over a decoder that always performs 4 half-iterations. We provide a CPU implementation achieving 100-200 megabytes/s single-core throughput on a regular laptop and a design of a potential hybrid FPGA-CPU implementation of our decoder.
Read the full abstract