Searcharxiv⌕ Search

arXiv subjects

Huan Zhou

Publications and source records attributed to Huan Zhou.

At least 37 records · Page 2Linked to original sources

NL2Repo-Bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents

Recent advances in coding agents suggest rapid progress toward autonomous software development, yet existing benchmarks fail to rigorously evaluate the long-horizon capabilities required to build complete software systems. Most prior evaluations focus on localized code generation, scaffolded completion, or short-term repair tasks, leaving open the question of whether agents can sustain coherent reasoning, planning, and execution over the extended horizons demanded by real-world repository construction. To address this gap, we present NL2Repo Bench, a benchmark explicitly designed to evaluate the long-horizon repository generation ability of coding agents. Given only a single natural-language requirements document and an empty workspace, agents must autonomously design the architecture, manage dependencies, implement multi-module logic, and produce a fully installable Python library. Our experiments across state-of-the-art open- and closed-source models reveal that long-horizon repository generation remains largely unsolved: even the strongest agents achieve below 40% average test pass rates and rarely complete an entire repository correctly. Detailed analysis uncovers fundamental long-horizon failure modes, including premature termination, loss of global coherence, fragile cross-file dependencies, and inadequate planning over hundreds of interaction steps. NL2Repo Bench establishes a rigorous, verifiable testbed for measuring sustained agentic competence and highlights long-horizon reasoning as a central bottleneck for the next generation of autonomous coding agents.

cs.CL↗

LLM Swiss Round: Aggregating Multi-Benchmark Performance via Competitive Swiss-System Dynamics

The rapid proliferation of Large Language Models (LLMs) and diverse specialized benchmarks necessitates a shift from fragmented, task-specific metrics to a holistic, competitive ranking system that effectively aggregates performance across multiple ability dimensions. Primarily using static scoring, current evaluation methods are fundamentally limited. They struggle to determine the proper mix ratio across diverse benchmarks, and critically, they fail to capture a model's dynamic competitive fitness or its vulnerability when confronted with sequential, high-stakes tasks. To address this, we introduce the novel Competitive Swiss-System Dynamics (CSD) framework. CSD simulates a multi-round, sequential contest where models are dynamically paired across a curated sequence of benchmarks based on their accumulated win-loss record. And Monte Carlo Simulation ($N=100,000$ iterations) is used to approximate the statistically robust Expected Win Score ($E[S_m]$), which eliminates the noise of random pairing and early-round luck. Furthermore, we implement a Failure Sensitivity Analysis by parameterizing the per-round elimination quantity ($T_k$), which allows us to profile models based on their risk appetite--distinguishing between robust generalists and aggressive specialists. We demonstrate that CSD provides a more nuanced and context-aware ranking than traditional aggregate scoring and static pairwise models, representing a vital step towards risk-informed, next-generation LLM evaluation.

cs.LG↗

AInsteinBench: Benchmarking Coding Agents on Scientific Repositories

We introduce AInsteinBench, a large-scale benchmark for evaluating whether large language model (LLM) agents can operate as scientific computing development agents within real research software ecosystems. Unlike existing scientific reasoning benchmarks which focus on conceptual knowledge, or software engineering benchmarks that emphasize generic feature implementation and issue resolving, AInsteinBench evaluates models in end-to-end scientific development settings grounded in production-grade scientific repositories. The benchmark consists of tasks derived from maintainer-authored pull requests across six widely used scientific codebases, spanning quantum chemistry, quantum computing, molecular dynamics, numerical relativity, fluid dynamics, and cheminformatics. All benchmark tasks are carefully curated through multi-stage filtering and expert review to ensure scientific challenge, adequate test coverage, and well-calibrated difficulty. By leveraging evaluation in executable environments, scientifically meaningful failure modes, and test-driven verification, AInsteinBench measures a model's ability to move beyond surface-level code generation toward the core competencies required for computational scientific research.

cs.SE↗

Virtual Width Networks

We introduce Virtual Width Networks (VWN), a framework that delivers the benefits of wider representations without incurring the quadratic cost of increasing the hidden size. VWN decouples representational width from backbone width, expanding the embedding space while keeping backbone compute nearly constant. In our large-scale experiment, an 8-times expansion accelerates optimization by over 2 times for next-token and 3 times for next-2-token prediction. The advantage amplifies over training as both the loss gap grows and the convergence-speedup ratio increases, showing that VWN is not only token-efficient but also increasingly effective with scale. Moreover, we identify an approximately log-linear scaling relation between virtual width and loss reduction, offering an initial empirical basis and motivation for exploring virtual-width scaling as a new dimension of large-model efficiency.

cs.LG↗

Extension of the Gyárfás-Sumner conjecture to signed graphs

The balanced chromatic number of a signed graph G is the minimum number of balanced sets that cover all vertices of G. Studying structural conditions which imply bounds on the balanced chromatic number of signed graphs is among the most fundamental problems in graph theory. In this work, we initiate the study of coloring hereditary classes of signed graphs. More precisely, we say that a set F = {F_1, F_2, ..., F_l} is a GS (for Gyárfás-Sumner) set if there exists a constant c such that signed graphs with no induced subgraph switching equivalent to a member of F admit a balanced c-coloring. The focus of this work is to study GS sets of order 2. We show that if F is a GS set of order 2, then F_1 is either (K_3, -) or (K_4, -), and F_2 is a linear forest. In the case of F_1 = (K_3, -), we show that any choice of a linear forest for F_2 works. In the case of F_1 = (K_4, -), we show that if each connected component of F_2 is a path of length at most 4, then {F_1, F_2} is a GS set.

math.CO↗

Brooks' theorem for signed graphs with $Δ=3$

Circular $r$-coloring of a signed graph $(G,σ)$ is a mapping of its vertices to a circle of circumference $r$ such that: I. each pair of vertices with a negative connection is at distance at least $1$, and II. for each pair with a positive connection, the distance of one from the antipodal of the other is at least $1$. A signed graph $(G,σ)$ admits a circular $r$-coloring for some values of $r$ if and only if it has no negative loop. The smallest value of such $r$ is the circular chromatic number, denoted $χ_{c}(G,σ)$. The circular chromatic number is a refinement of the balanced chromatic number, which is mostly studied under the equivalent term $0$-free coloring in the literature. Extending Brooks' theorem, Má\v cajová, Raspaud, and Škoviera showed that if $Δ(G)$ is an even number, $G$ is connected, and $(G,σ)$ is not (switching) isomorphic to $(K_{Δ+1},-)$ or $C_{-\ell}$ (when $Δ(G)=2$), then $χ_c(G,σ)\leq Δ(G)$ and that the upper bound is tight. For the odd values of $Δ(G)$, assuming a connected signed graph $(G,σ)$ is not isomorphic to $(K_{Δ+1},-)$, determining the best upper bound for $χ_c(G, σ)$ proves to be more of a challenge. In this work, addressing the first step of this question, we show that if $(G, σ)$ is a signed graph of maximum degree 3 with no component isomorphic to $(K_4, -)$, then $χ_{c}(G, σ)\leq \frac{10}{3}$. The upper bound is tight even among signed cubic graphs of girth 5. In particular, there is a signature on the Petersen graph for which the upper of $\frac{10}{3}$ is achieved.

math.CO↗

Identifying Microlensing by Compact Dark Matter through Diffraction Patterns in Gravitational Waves with Machine Learning

Gravitational wave lensing, particularly microlensing by compact dark matter (DM), offers a unique avenue to probe the nature of dark matter. However, conventional detection methods are often computationally expensive, inefficient, and sensitive to waveform systematics. In this work, we introduce the Wavelet Convolution Detector (WCD), a deep learning framework specifically designed to identify wave-optics diffraction patterns imprinted in gravitationally lensed signals. The WCD integrates multi-scale wavelet analysis within residual convolutional blocks to efficiently extract time-frequency interference structures, and is trained on a realistically generated dataset incorporating compact DM mass functions and astrophysical lensing probabilities. This work is the first machine learning-based approach capable of identifying such wave-optics signatures in lensed gravitational waves. Tested on simulated binary black hole events, the model achieves 92.2\% accuracy (AUC=0.965), with performance rising to AUC$\sim$0.99 at high SNR. Crucially, it maintains high discriminative power across a wide range of lens masses without retraining, demonstrating particular strength in the low-impact-parameter and high-lens-mass regimes where wave-optics effects are most pronounced. Compared to Bayesian inference, the WCD provides orders-of-magnitude faster inference, making it a scalable and efficient tool for discovering compact DM through lensed gravitational waves in the era of third-generation detectors.

astro-ph.IM↗

Degree-truncated choosability of graphs

A graph $G$ is called degree-truncated $k$-choosable if for every list assignment $L$ with $|L(v)| \ge \min\{d_G(v), k\}$ for each vertex $v$, $G$ is $L$-colourable. Richter asked whether every 3-connected non-complete planar graph is degree-truncated 6-choosable. We answer this question in negative by constructing a 3-connected non-complete planar graph which is not degree-truncated 7-choosable. Then we prove that every 3-connected non-complete planar graph is degree-truncated 16-DP-colourable (and hence degree-truncated $16$-choosable). We further prove that for an arbitrary proper minor closed family ${\mathcal G}$ of graphs, let $s$ be the minimum integer such that $K_{s,t} \notin \mathcal{G}$ for some $t$, then there is a constant $k$ such that every $s$-connected graph $G \in {\mathcal G}$ other than a GDP tree is degree-truncated DP-$k$-colourable (and hence degree-truncated $k$-choosable), where a GDP-tree is a graph whose blocks are complete graphs or cycles. In particular, for any surface $Σ$, there is a constant $k$ such that every 3-connected non-complete graph embeddable on $Σ$ is degree-truncated DP-$k$-colourable (and hence degree-truncated $k$-choosable). The $s$-connectedness for graphs in $\mathcal{G}$ (and 3-connectedness for graphs embeddable on $Σ$) is necessary, as for any positive integer $k$, $K_{s-1,k^{s-1}} \in \mathcal{G}$ ($K_{2,k^2}$ is planar) is not degree-truncated $k$-choosable. Also, non-completeness is a necessary condition, as complete graphs are not degree-choosable.

math.CO↗

Constraints on fast radio burst population from the first CHIME/FRB catalog with the Hierarchical Bayesian Inference

Fast Radio Bursts (FRBs) have emerged as one of the most dynamic areas of research in astronomy and cosmology. Despite increasing number of FRBs have been reported, the exact origin of FRBs remains elusive. Investigating the intrinsic redshift distributions of FRBs could provide valuable insights into their possible origins and enhance the power of FRBs as a cosmological probe. In this paper, we propose a hierarchical Bayesian inference approach combining with several viable models to investigate the redshift distribution of the CHIME/FRB catalog 1. By utilizing this method, we aim to uncover the underlying patterns and characteristics of the FRB population, i.e. intrinsic redshift distribution of FRB. Taking uncertainties within the observational data and selection effects into consideration, we obtained that the redshift distribution of FRBs is significantly delayed with respect to that of the star formation history.

astro-ph.HE↗

Fractional balanced chromatic number and arboricity of planar (signed) graphs

A fractional coloring of a signed graph $(G, σ)$ is an assignment of nonnegative weights to the balanced sets (sets which do not induce a negative cycle) such that each vertex has an accumulated weight of at least 1. The minimum total wight among all such colorings is defined to be the fractional balanced chromatic number, denoted by $χ-{fb}(G, σ)$. This value is clearly upper bounded by the fractional arboricity of $G$, denoted $a_f (G)$, where weights are assigned to sets inducing no cycle rather than sets inducing no negative cycle. In this work we present an example of a planar signed simple graph of fractional balanced chromatic number larger than 2, thus in particular refuting a conjecture of Bonamy, Kardos, Kelly, and Postle suggesting that the fractional arboricity of planar graphs is bounded above by 2. By iterating the construction, we show that the supremum of the fractional balanced chromatic number of planar signed simple graphs is at least as $83/41 = 2 + 1/41$. With similar operations, we built a sequence of planar graphs whose limit of fractional arboricity is $a_f (G) = 2 + 2/25$.

math.CO↗

Truncated degree DP-colourability of $K_{2,4}$-minor free graphs

Assume $G$ is a graph and $k$ is a positive integer. Let $f$ from $V(G)$ to $ N$ be defined as $f(v)$ is the minimum of $k$ and $d(v)$. If $G$ is $f$-DP-colourable (respectively, $f$-choosable), then we say $G$ is $k$-truncated degree DP-colourable (respectively, $k$-truncated degree-choosable). Hutchinson proved that 2-connected maximal outerplanar graphs other than the triangle are $5$-truncated degree-choosable, and asked whether the result can be extended to all outerplanar graphs, and the question remained open. This paper proves that 2-connected $K24$-minor free graphs other than cycles and complete graphs are $5$-truncated degree DP-colourable. This not only answers Hutchinson's question in the affirmative, but also extends to a larger family of graphs, and strengthens choosability to DP-colourability.

math.CO↗

PromptMID: Modal Invariant Descriptors Based on Diffusion and Vision Foundation Models for Optical-SAR Image Matching

The ideal goal of image matching is to achieve stable and efficient performance in unseen domains. However, many existing learning-based optical-SAR image matching methods, despite their effectiveness in specific scenarios, exhibit limited generalization and struggle to adapt to practical applications. Repeatedly training or fine-tuning matching models to address domain differences is not only not elegant enough but also introduces additional computational overhead and data production costs. In recent years, general foundation models have shown great potential for enhancing generalization. However, the disparity in visual domains between natural and remote sensing images poses challenges for their direct application. Therefore, effectively leveraging foundation models to improve the generalization of optical-SAR image matching remains challenge. To address the above challenges, we propose PromptMID, a novel approach that constructs modality-invariant descriptors using text prompts based on land use classification as priors information for optical and SAR image matching. PromptMID extracts multi-scale modality-invariant features by leveraging pre-trained diffusion models and visual foundation models (VFMs), while specially designed feature aggregation modules effectively fuse features across different granularities. Extensive experiments on optical-SAR image datasets from four diverse regions demonstrate that PromptMID outperforms state-of-the-art matching methods, achieving superior results in both seen and unseen domains and exhibiting strong cross-domain generalization capabilities. The source code will be made publicly available https://github.com/HanNieWHU/PromptMID.

cs.CV↗

Indicated list colouring game on graphs

Given a graph $G$ and a list assignment $L$ for $G$, the indicated $L$-colouring game on $G$ is played by two players: Ann and Ben. In each round, Ann chooses an uncoloured vertex $v$, and Ben colours $v$ with a colour from $L(v)$ that is not used by its coloured neighbours. If all vertices are coloured, then Ann wins the game. Otherwise after a finite number of rounds, there remains an uncoloured vertex $v$ such that all colours in $L(v)$ have been used by its coloured neighbours, Ben wins. We say $G$ is indicated $L$-colourable if Ann has a winning strategy for the indicated $L$-colouring game on $G$. For a mapping $g: V(G) \to \mathbb{N}$, we say $G$ is indicated $g$-choosable if $G$ is indicated $L$-colourable for every list assignment $L$ with $|L(v)| \ge g(v)$ for each vertex $v$, and $G$ is indicated degree-choosable if $G$ is indicated $g$-choosable for $g(v) =d_G(v)$ (the degree of $v$). This paper proves that a graph $G$ is not indicated degree-choosable if and only if $G$ is an expanded Gallai-tree - a graph whose maximal connected induced subgraphs with no clique-cut are complete graphs or blow-ups of odd cycles, along with a technical condition (see Definition \ref{def-egt}). This leads to a linear-time algorithm that determines if a graph is indicated degree-choosable. A connected graph $G$ is called an IC-Brooks graph if its indicated chromatic number equals $Δ(G)+1$. Every IC-Brooks graph is a regular expanded Gallai-tree. We show that if $r \le 3$, then every $r$-regular expanded Gallai-tree is an IC-Brooks graph. For $r \ge 4$, there are $r$-regular expanded Gallai-trees that are not IC-Brooks graphs. We give a characterization of IC-Brooks graphs, and present a linear-time algorithm that determines if a given graph of bounded maximum degree is an IC-Brooks graph.

math.CO↗

CAMEL: Cross-Attention Enhanced Mixture-of-Experts and Language Bias for Code-Switching Speech Recognition

Code-switching automatic speech recognition (ASR) aims to transcribe speech that contains two or more languages accurately. To better capture language-specific speech representations and address language confusion in code-switching ASR, the mixture-of-experts (MoE) architecture and an additional language diarization (LD) decoder are commonly employed. However, most researches remain stagnant in simple operations like weighted summation or concatenation to fuse languagespecific speech representations, leaving significant opportunities to explore the enhancement of integrating language bias information. In this paper, we introduce CAMEL, a cross-attention-based MoE and language bias approach for code-switching ASR. Specifically, after each MoE layer, we fuse language-specific speech representations with cross-attention, leveraging its strong contextual modeling abilities. Additionally, we design a source attention-based mechanism to incorporate the language information from the LD decoder output into text embeddings. Experimental results demonstrate that our approach achieves state-of-the-art performance on the SEAME, ASRU200, and ASRU700+LibriSpeech460 Mandarin-English code-switching ASR datasets.

cs.SD↗

On the second-order zero differential properties of several classes of power functions over finite fields

Feistel Boomerang Connectivity Table (FBCT) is an important cryptanalytic technique on analysing the resistance of the Feistel network-based ciphers to power attacks such as differential and boomerang attacks. Moreover, the coefficients of FBCT are closely related to the second-order zero differential spectra of the function $F(x)$ over the finite fields with even characteristic and the Feistel boomerang uniformity is the second-order zero differential uniformity of $F(x)$. In this paper, by computing the number of solutions of specific equations over finite fields, we determine explicitly the second-order zero differential spectra of power functions $x^{2^m+3}$ and $x^{2^m+5}$ with $m>2$ being a positive integer over finite field with even characteristic, and $x^{p^k+1}$ with integer $k\geq1$ over finite field with odd characteristic $p$. It is worth noting that $x^{2^m+3}$ is a permutation over $\mathbb{F}_{2^n}$ and only when $m$ is odd, $x^{2^m+5}$ is a permutation over $\mathbb{F}_{2^n}$, where integer $n=2m$. As a byproduct, we find $F(x)=x^4$ is a PN and second-order zero differentially $0$-uniform function over $\mathbb{F}_{3^n}$ with odd $n$. The computation of these entries and the cardinalities in each table aimed to facilitate the analysis of differential and boomerang cryptanalysis of S-boxes when studying distinguishers and trails.

cs.CR↗

XCB: an effective contextual biasing approach to bias cross-lingual phrases in speech recognition

Contextualized ASR models have been demonstrated to effectively improve the recognition accuracy of uncommon phrases when a predefined phrase list is available. However, these models often struggle with bilingual settings, which are prevalent in code-switching speech recognition. In this study, we make the initial attempt to address this challenge by introducing a Cross-lingual Contextual Biasing(XCB) module. Specifically, we augment a pre-trained ASR model for the dominant language by integrating an auxiliary language biasing module and a supplementary language-specific loss, aimed at enhancing the recognition of phrases in the secondary language. Experimental results conducted on our in-house code-switching dataset have validated the efficacy of our approach, demonstrating significant improvements in the recognition of biasing phrases in the secondary language, even without any additional inference overhead. Additionally, our proposed system exhibits both efficiency and generalization when is applied by the unseen ASRU-2019 test set.

cs.CL↗

A new framework for calibrating COVID-19 SEIR models with spatial-/time-varying coefficients using genetic and sliding window algorithms

A susceptible-exposed-infected-removed (SEIR) model assumes spatial-/time-varying coefficients to model the effect of non-pharmaceutical interventions (NPIs) on the regional and temporal distribution of COVID-19 disease epidemics. A significant challenge in using such model is their fast and accurate calibration to observed data from geo-referenced hospitalized data, i.e., efficient estimation of the spatial-/time-varying parameters. In this work, a new calibration framework is proposed towards optimizing the spatial-/time-varying parameters of the SEIR model. We also devise a method for combing the overlapping sliding window technique (OSW) with a genetic algorithm (GA) calibration routine to automatically search the segmented parameter space. Parallelized GA is used to reduce the computational burden. Our framework abstracts the implementation complexity of the method away from the user. It provides high-level APIs for setting up a customized calibration system and consuming the optimized values of parameters. We evaluated the application of our method on the calibration of a spatial age-structured microsimulation model using a single objective function that comprises observed COVID-19-related ICU demand. The results reflect the effectiveness of the proposed method towards estimating the parameters in a changing environment.

cs.DC↗

MMGER: Multi-modal and Multi-granularity Generative Error Correction with LLM for Joint Accent and Speech Recognition

Despite notable advancements in automatic speech recognition (ASR), performance tends to degrade when faced with adverse conditions. Generative error correction (GER) leverages the exceptional text comprehension capabilities of large language models (LLM), delivering impressive performance in ASR error correction, where N-best hypotheses provide valuable information for transcription prediction. However, GER encounters challenges such as fixed N-best hypotheses, insufficient utilization of acoustic information, and limited specificity to multi-accent scenarios. In this paper, we explore the application of GER in multi-accent scenarios. Accents represent deviations from standard pronunciation norms, and the multi-task learning framework for simultaneous ASR and accent recognition (AR) has effectively addressed the multi-accent scenarios, making it a prominent solution. In this work, we propose a unified ASR-AR GER model, named MMGER, leveraging multi-modal correction, and multi-granularity correction. Multi-task ASR-AR learning is employed to provide dynamic 1-best hypotheses and accent embeddings. Multi-modal correction accomplishes fine-grained frame-level correction by force-aligning the acoustic features of speech with the corresponding character-level 1-best hypothesis sequence. Multi-granularity correction supplements the global linguistic information by incorporating regular 1-best hypotheses atop fine-grained multi-modal correction to achieve coarse-grained utterance-level correction. MMGER effectively mitigates the limitations of GER and tailors LLM-based ASR error correction for the multi-accent scenarios. Experiments conducted on the multi-accent Mandarin KeSpeech dataset demonstrate the efficacy of MMGER, achieving a 26.72% relative improvement in AR accuracy and a 27.55% relative reduction in ASR character error rate, compared to a well-established standard baseline.

eess.AS↗