Searcharxiv⌕ Search

arXiv subjects

Bin Liu

Publications and source records attributed to Bin Liu.

At least 19 recordsLinked to original sources
arXiv2026

RobustSGPO: Search-Space Control for Agent Harness Evolution

Semantic-gradient-based prompt optimization (SGPO) improves agent harnesses using execution feedback, but its local update rule leaves the choice of edit scope and operation unresolved. We introduce RobustSGPO, which specifies the requested edit, constructs and checks the patch, and continues search from either the incumbent or retained snapshots. We evaluate permission scheduling, cumulative controls, and task-family transfer in the AgentX brainstorming workflow using 120 tasks, 95 runs, and 7,350 candidate attempts. Periodic $1\to2\to3$ scheduling exceeds fixed maximum permission by 0.28 test-score points. RobustSGPO increases completion on 30 held-out tasks from 60.0% to 80.0% and improves test quality from 3.77 to 4.14 under a 20-million-token budget. Category retention reduces source-task degradation after a shift, whereas random retention reaches a higher destination endpoint. Search-space control benefits quality through executable edits and alternative starting points, with measurable retention overhead.

cs.AI↗
arXiv2026

Can isolated binaries form unequal-mass binary black-hole mergers with a high-spin primary black hole?

GWTC-5 has revealed a subpopulation of merging binary black holes (BBHs) with a high-spin primary black hole (BH) and possibly unequal BH masses. GW241110 additionally exhibits a large spin-orbit misalignment, which suggests a hierarchical-merger origin. However, other formation scenarios are possible or even likely, especially for events without constraints on spin-orbit misalignment. As an alternative to hierarchical mergers, we investigate whether binary evolution can produce unequal-mass BBH mergers with a high-spin primary BH. Rather than performing comprehensive population-synthesis calculations, we examine the evolutionary pathways of forming merging BBHs and assess their uncertainties. We identify two possible pathways for producing unequal-mass BBHs with a high-spin primary. In initially wide binaries, mass-ratio reversal can make the tidally spun-up second-born BH both more massive and more rapidly rotating than the first-born BH; alternatively, in an initially close, unequal-mass binary, the primary star may evolve chemically homogeneously, while the secondary star evolves normally, producing a high-spin first-born BH that is more massive than its companion. Generally, large spin-orbit misalignment can be produced by large natal kicks or tertiary-induced nodal precession and/or Zeipel-Lidov-Kozai oscillations. We conclude that hierarchical mergers are not uniquely required to produce unequal-mass BBHs with a high-spin primary BH, although each isolated-binary pathway faces important theoretical constraints. Future detections of more merger events with primary BH spins around 0.7 would discriminate between binary evolution and hierarchical merger origin.

astro-ph.SR↗
arXiv2026

TransRetrieval: Scaling Up Transformer-Based Retrieval for Industrial Recommendation

Applying scaling laws to recommendation retrieval is hindered by feature heterogeneity: naively stacking Transformer layers yields diminishing returns because heterogeneous fields produce severe token-norm divergence. We present TransRetrieval, a Transformer-based retrieval framework that scales with both computational budget and cross-domain data. The key enabler is (1) weighted average aggregation, which restores the homogeneous-token assumption Transformers rely on. Building on this, we introduce (2) target token compression that cuts per-candidate FLOPs by 85% while preserving cross-attention expressiveness, and (3) position-style domain embeddings that unify multiple domains at negligible additional cost, turning cross-domain data into a scaling asset. On a 40-billion-interaction industrial dataset and the public KuaiRand benchmark, scaling compute from 0.1 to 2 MFLOPs per target yields +19.3/+22.2 pt Recall@2000, confirming robust log-linear scaling. In online A/B tests, TransRetrieval lifts platform revenue by 2.53% under the same end-to-end latency constraint as the production baseline.

cs.IR↗
arXiv2026

Is Next-Chunk Reasoning RL Really Better than SFT? Revisiting Training Strategies under no-CoT Data

Recent work proposes next-chunk reasoning RL for leveraging no-CoT data---corpora such as worked solutions and textbook derivations that contain reasoning-rich content but lack explicit chain-of-thought annotations. The method trains a model to generate implicit reasoning traces and rewards them by their ability to predict the next chunk of text. While promising, existing evaluations primarily compare against conventional SFT baselines, leaving open whether the gains come from the RL formulation itself or from more effectively exposing the model to no-CoT data. We address this question with a controlled study of next-chunk reasoning RL and a simple but previously overlooked alternative: Mixed SFT, a single supervised fine-tuning stage that jointly trains on no-CoT and long-CoT data. Despite its simplicity, Mixed SFT achieves a clearly higher post-RLVR performance ceiling than next-chunk reasoning RL while requiring over 60 times less training compute. The advantage is consistent across in-domain mathematical reasoning and out-of-domain reasoning tasks. Moreover, we show that higher pre-RLVR accuracy does not necessarily translate into higher post-RLVR accuracy, highlighting the need to evaluate no-CoT training strategies in the context of the full post-training pipeline.

cs.AI↗
arXiv2026

Optimal Grading: A Unified Approach

We develop a unified approach to optimal grading in an all-pay contest in which a designer assigns a fixed vector of heterogeneous prizes to maximize expected total effort. The approach covers two information regimes and identifies a common principle: iron locally misordered incentive returns and assign prizes assortatively across the resulting grades. Under rank-only grading, assignments depend only on ordinal ranks. Ironing cumulative rank coefficients---via the least concave majorant or the pool-adjacent-violators algorithm---determines which adjacent ranks are pooled and which prizes are randomized within each grade. Under performance-contingent grading, assignments may depend on numerical effort. The optimum irons virtual ability, forms endogenous type grades, and assigns prize blocks assortatively across grades. A failing grade below a minimum passing effort and a collection of effort brackets implement the direct optimum while preserving full prize assignment.

econ.TH↗
arXiv2026

Gaussian Sculpting: End-to-End Controllable Surface Reconstruction via Field Optimization

3D Gaussian Splatting (3DGS) has recently enabled real-time novel view synthesis with impressive quality. However, it struggles to recover accurate surfaces under limited viewpoints and due to the inherent irregularity of Gaussian primitives. The resulting geometric errors are notoriously difficult to correct manually. To address these issues, we propose Gaussian Sculpting, a fully differentiable end-to-end framework for high-quality surface reconstruction. Our key insight is to anchor Gaussians onto an evolving differentiable surface, allowing them to guide signed distance field (SDF) optimization instead of extracting the surface only during post-processing. To enable stable gradient isolation during joint optimization, we design a bi-level training strategy in which the outer loop optimizes the geometry represented by the SDF, while the inner loop updates the Gaussians with the geometry fixed. We further impose constraints on Gaussian parameters to ensure consistency with the underlying surface, thereby improving both geometric and appearance fidelity during optimization. In addition, we introduce a multi-resolution subdivision scheme based on octree-like partitioning to preserve fine details while reducing memory consumption. Experiments on object-level scenes demonstrate that our method effectively removes redundant surfaces, recovers missing structures caused by limited viewpoints, and achieves strong reconstruction quality even at relatively low resolutions.

cs.CV↗
arXiv2026

Yesterday's Shield, Today's Spear: A Self-Evolving Safety Guardrail in Production

Deployed LLM safety guardrails are predominantly static: trained once and frozen at release, while new jailbreak techniques and previously un-addressed harmful categories emerge within days, leaving the defense perpetually a step behind. We present SESG (Self-Evolving Safety Guardrails), a multi-agent system running in production. SESG monitors the live traffic behind a deployed guardrail and surfaces two classes of failure: jailbreaks novel in form and harmful categories novel in content. Once a failure is confirmed, a generation agent synthesizes paired training data targeted at it; a validation agent rebalances the batch toward the direction in which the deployed model errs, so that the model's own mistakes steer its training set; and a routing agent matches the training action to the diagnosed gap and returns the next version to production. Over six rounds of live evolution (V0 to V6), a 1.7B guardrail adapts to a new threat in 16-24 hours, with about 2 hours of human effort, versus the 40-90 hours of the manual process it replaces. On six emerging threats, it outperforms static guardrails from 0.6B to 9B and an adaptive baseline while preserving its general screening competence. Since April 2026, SESG has been the primary update pipeline of Sangfor's guardrail, autonomously closing 14 of 15 new threat scenarios in two months. We release 9 test sets for the 6 new threats at https://github.com/Trams1017/SESG. Warning: This paper contains examples that may be harmful or offensive.

cs.AI↗
arXiv2026

No evidence for a supermassive black hole binary in GSN 069

Quasi-periodic eruptions (QPEs) are recurrent soft X-ray flares from galactic nuclei and provide a new time-domain probe of stellar-mass objects (SMOs) orbiting supermassive black holes (SMBHs). In an extreme-mass-ratio inspiral (EMRI) system interacting with an accretion disk, QPEs are produced when the SMO repeatedly crosses an accretion disk, so that the eruption times trace the orbital motion of the EMRI. We investigate whether such timing information can be used to probe a more distant SMBH companion. We develop two complementary diagnostics: (1) the motion of the EMRI host SMBH around the SMBH-binary (SMBHB) center of mass induces a light-travel-time modulation in the observed QPE arrival times, specifically an \emph{in-phase} modulation in arrival times of even and odd eruptions; (2) if the QPE source contains a surviving stellar orbiter, the external SMBH must not drive the SMO into tidal disruption through eccentricity excitation by the von Zeipel--Lidov--Kozai (ZLK) mechanism. Using GSN 069 as an example, we find \emph{no} in-phase modulation in the QPE timing (i.e., no evidence for a SMBHB) and constrain the excluded parameter space of the companion SMBH. These results demonstrate that QPE timing and stellar survival offer complementary routes for constraining otherwise hidden SMBH companions in nearby galactic nuclei.

astro-ph.HE↗
arXiv2026

FAST Ultra-Deep Survey: the baryonic Tully-Fisher relation in FUDS0 field

The Baryonic Tully-Fisher relation (BTFR) is one of the tightest scaling relations for disk galaxies in the local Universe, and therefore is an important tool for studying the fomation and evolution of galaxies. However, the evolution of the BTFR over cosmic time is poorly understood due to the limited sample of HI galaxies beyond the local Universe, limitations of optically-derived rotation curves, and selection effects. In this work, we explore the BTFR at redshifts up to $z=0.42$ from galaxies detected in the pilot FAST Ultra-Deep Survey (FUDS) field, FUDS0. As found in previous work, we identify two components in the plane of baryonic mass versus rotational velocity, $C_{\rm BTFR}$ (tight) and $C_{\rm Outlier}$ (dispersed). A Gaussian mixture model is employed to recover the BTFR, yielding the best fit parameters for the slope $k=3.32_{-0.11}^{+0.12}$, zero point $b=10.07_{-0.03}^{+0.03}$, and intrinsic scatter $\sigma_{\rm BTFR}=0.036_{-0.009}^{+0.010}$. A random forest classifier is used to investigate the origin of the outlier component. We find that low signal significance and inaccurate inclinations are the key factors that contribute to the outlier population, indicating that observational effects are the dominant origin. Evolutionary trends are examined in three different redshift bins. Both the slope and zero point show consistency within 1-$\sigma$ uncertainty in the two low redshift bins, indicating no significant evolution. The indirectly inferred BTFR parameters from the $C_{\rm Outlier}$ component in the highest redshift bin aligns with the conclusion. The ongoing full FUDS survey will provide a larger sample to enable more accurate constraints on BTFR evolution.

astro-ph.GA↗
arXiv2026

AirKey: Multimodal Acoustic-Assisted WiFi Sensing for Zero-Training Robust PIN Inference

Contactless keystroke inference via WiFi sensing highlights severe privacy threats, yet its real-world feasibility is hindered by two fundamental physical and deployment bottlenecks: the strict requirement for network privileges to acquire stable sensing streams, and the inherent "waveform fusion" ambiguity of pure WiFi signals during rapid, muscle-memory typing. To overcome these limitations, we propose AirKey, a novel cross-modal sensing framework that achieves highly stealthy, zero-training PIN eavesdropping. First, to bypass network deployment barriers, AirKey exploits fundamental IEEE 802.11 mechanisms to predictably elicit Acknowledgment (ACK) responses from unmodified target devices. By passively harvesting Channel State Information (CSI) from these ACKs using a low-cost microcontroller, AirKey secures a continuous spatial sensing stream entirely without network association. Crucially, to resolve the WiFi waveform fusion bottleneck, AirKey introduces a cross-modal complementarity mechanism. By utilizing lightweight acoustic signals as precise temporal anchors, the system robustly guides the segmentation of overlapping CSI trajectories. This joint spatiotemporal fusion strictly intersects CSI-derived spatial similarities with acoustic-guided inter-keystroke timing. Extensive real-world evaluations demonstrate that AirKey achieves over 4x higher accuracy than state-of-the-art unimodal zero-training schemes, successfully recovering device-unlock PINs within 6 attempts. Ultimately, this work exposes a critical vulnerability in contemporary smart interfaces, underscoring the severe privacy implications of ubiquitous multimodal sensing.

cs.CR↗
arXiv2026

The PSR J0435+3233 Triple System

The detailed evolution of triple star systems is complicated and poorly known. Based on the optical/infrared and gamma-ray archived data, we identified that the pulsar, PSR~J0435+3233, is a gamma-ray pulsar in a hierarchical triple system, with a helium white dwarf (WD) as a close inner binary companion and a Sun-like star as the distant tertiary. PSR~J0435+3233 and the WD companion are in a circular orbit with a period of $P_{\rm orb1} = 8$~days and an eccentricity of $e=0.00016$.The tertiary is a G-type subgiant with a mass of $0.98(12) M_\odot$ at a distance of $2.1(4)$ kpc from the Earth. By simultaneously fitting the observed spin-period variations of the gamma-ray emission (over 16.7 years) and radio emission (over 4.6 years) from PSR~J0435+3233, the changes of the inner orbital parameters, the Shapiro delay, Gaia astrometry, and the outer companion mass, we determined the outer elliptical orbit for the tertiary, with a period $P_{\rm orb2} \sim 26900$~days and an eccentricity $e_2 = 0.5983$. The outer orbit is either nearly perpendicular to the inner orbit (mutual inclination $\sim 84^\circ$), or exhibits a moderate mutual inclination of $\sim 55^\circ$. For the former geometry, the pulsar, the WD, and the tertiary star have masses of $1.15^{+0.06}_{-0.04} M_\odot$, $0.271^{+0.010}_{-0.006} M_\odot$, and $0.96(4) M_\odot$, respectively; for the latter geometry, the corresponding masses are $1.29^{+0.14}_{-0.11} M_\odot$, $0.296^{+0.022}_{-0.018} M_\odot$, and $1.12^{+0.06}_{-0.05} M_\odot$. This is a unique triple system for detailed multi-band observations and for studying the evolutionary path and dynamic processes of a primordial triple star system. It will ultimately evolve into a system consisting of a neutron star and two white dwarfs.

astro-ph.HE↗
arXiv2026

DistillAlign: Coordinating Mode Covering and Mode Seeking in Autoregressive Video Distillation

Existing autoregressive video distillation methods commonly adopt a Distribution Matching Distillation (DMD)-based multi-stage pipeline. However, they typically decouple the initialization and DMD stages -- which then pursue different target distributions -- and judge the intermediate student mainly by visual scores such as VBench. In this paper, we revisit this design from a distributional perspective. Given the mode-seeking nature of the distribution matching loss, a good initialization should match the mode coverage of the target DMD teacher, rather than merely pursuing high quality. To analyze this, we introduce a distributional evaluation protocol that measures precision and coverage between student and teacher distributions in a shared latent space. It exposes differences hidden by visual scores: some initializations reach high precision but low coverage, leading to suboptimal refinement, while mode-covering ones preserve broader support. Furthermore, even when the target distributions are aligned, DMD's reverse-KL objective can still drive the student toward high-probability teacher regions in late training, reducing coverage and diversity. To address this, we propose joint distillation, which combines DMD's mode-seeking objective with a Consistency Distillation-based mode-covering constraint. Experiments show that our method improves generation quality, coverage, and diversity; notably, even with a Wan-1.3B DMD teacher, it outperforms baselines refined with Wan-14B, underscoring the importance of distributional alignment in autoregressive video distillation.

cs.CV↗
arXiv2026

DLAM: Distributional Latent Actions with Temporal Constraints

Vision-language-action (VLA) models remain constrained by scarce action-labeled robot data, whereas action-free videos offer abundant observations of physical change. Latent action models can extract such priors, but reconstruction-trained codes may predict future observations without the structure required for joint generation with robot actions. Existing structured methods add temporal constraints but retain deterministic transition points, so residual errors in locally inferred transitions may propagate and compound under recursive composition. We introduce DLAM, a distributional latent-action model that represents each transition as a diagonal Gaussian. Reconstruction conditioned on the reference frame grounds the mean in observed visual change, while normalized composition and reversal over equal-gap triplets constrain both the mean and dimension-wise variance. Variance composition uses a lightweight shared-correlation coefficient to account for dependence between adjacent transitions that share an intermediate frame, whereas reversal negates the mean and preserves the variance. For downstream policy learning, we freeze the encoder and train a flow-matching policy to jointly generate mean transition sequences and robot actions. On held-out transitions, DLAM learns more temporally consistent latent dynamics than existing latent-action baselines and achieves stronger direct and cumulative reconstruction on held-out videos. Under the same controlled $\pi_0$ transfer protocol, it also improves policy performance on MetaWorld MT50, LIBERO, and real-world manipulation tasks. Controlled ablations show that normalized mean constraints account for most of the reconstruction gain, while learned variance and correlation-aware composition provide complementary improvements in downstream control.

cs.RO↗
arXiv2026

Regularizing modality contribution drift in multimodal continual learning

Multimodal continual learning (MMCL) aims to learn emerging knowledge from multimodal data while preserving knowledge. To mitigate forgetting, current MMCL methods usually focus on cross-modal representation alignment or semantic similarity, but they overlook whether the relative contributions of individual modalities and their interactions remain stable across incremental tasks. We term this decision-level shift Modality Contribution Drift (MCD) and quantify it with the MCD score, which combines contribution-strength and relative-reliance changes under controlled interventions on modality subsets. Theoretical and empirical analyses further explain why current MMCL methods cannot reliably mitigate this drift. To this end, we propose Continual Modality Contribution Drift Regularization (CMCDR), which preserves the modality contribution structure of previously learned tasks. Since MMCL settings differ in whether old exemplars are available, CMCDR includes both replay-based and replay-free versions. The replay-based version uses modality-subset interventions as diagnostic probes on stored old samples, compares their contribution profiles between the current model and a frozen previous model, and constrains changes in old-sample modality-specific and interaction contributions. The replay-free version uses current-task samples as probes and distills the frozen model's old-task contribution responses, thereby regularizing the observed contribution profile without exemplars. Experiments on multimodal class-incremental learning and continual visual question answering validate the generality and effectiveness of CMCDR.

cs.LG↗
arXiv2026

ANFI: Rethinking Neighbor Feature Interaction in Person Re-ID

In person re-identification, neighbor-based methods have achieved significant success by interacting with neighbor samples to obtain more robust representations. However, existing methods rely only on affinity relations, causing their success to depend heavily on the reliability of selected neighbors. We find that affinity-only interaction often fails in challenging scenarios due to the inevitable presence of noisy neighbors. To enable effective interactions under noisy neighborhoods, we revisit neighbor-based methods under distinct reliability conditions and propose a novel Adaptive Neighbor Feature Interaction (ANFI) method. The core idea of ANFI is to account for negative effects from noisy neighbors, allowing samples to remain distinguishable from false positive neighbors. Unlike existing methods, ANFI models not only affinity relations but also discrepancy relations, and employs sample-wise adaptive weighting for these two types of relations. Given that capturing negative effects from noisy neighbors differs significantly from traditional relation learning, we derive discrepancy relations from a new neighborhood similarity, which provides more information than pairwise similarity. In addition, we propose Noisy Relation Supervision (NRS) to train ANFI, gradually injecting robustness to noisy relations into the model. Extensive experiments conducted under standard, cross-modal, and cross-domain settings, including comparisons with neighbor-based methods and re-ranking methods, demonstrate the superiority of our method across various neighbor distributions.

cs.CV↗
arXiv2026

SciExplore: Evaluating Autonomous Agents from Scientific Navigation to Information Integration

Scientific research involves complex information-seeking and reasoning workflows across heterogeneous sources. However, existing benchmarks primarily emphasize general-domain retrieval or static scientific question answering, and therefore fail to assess key capabilities required in realistic scientific research workflows. We introduce SciExplore, a benchmark designed to evaluate scientific information-seeking and reasoning capabilities of LLMs and agents. SciExplore comprises four task types covering 103 expert-curated tasks across more than ten scientific disciplines: scientific database navigation, ambiguous literature retrieval, missing reference completion, and cross-source structured knowledge synthesis, which probe progressively higher-level abilities from entity-level reasoning and document-level identification to evidence-level grounding and domain-level synthesis. We evaluate over ten state-of-the-art LLMs and autonomous agents on SciExplore, revealing substantial performance gaps with performance degrading sharply as task complexity increases and extremely low accuracy on the most challenging structured synthesis tasks. These results highlight significant limitations of current models and agents in realistic scientific information-seeking scenarios.

cs.AI↗
arXiv2026

What Transfers Under Source Shift? Definitions, Examples, and Fine-Tuning for Climate Disclosure Classification

Climate disclosure classification is a fundamental task for analysing corporate climate disclosures, yet such disclosures appear in many different sources -- annual reports, press releases, and earnings calls -- that differ in length, purpose, and writing style. Existing evaluations are mostly conducted within a single source, leaving open whether common LLM adaptation strategies remain effective under source shift. We reframe climate disclosure classification as a cross-source adaptation problem and study three widely used adaptation strategies -- definitions, examples, and fine-tuning -- across eleven open- and closed-source LLMs, using two corpora that share the same label space but come from different sources. We find that all strategies bring positive cross-source gains on average, but the strongest in-source strategies are not the strongest cross-source ones: similarity-based retrieval and LoRA fine-tuning gain most in-source but lose most of that advantage under source shift; randomly selected few-shot examples, a weaker in-source baseline, retain their advantage more reliably; definitions transfer most consistently, though only when their granularity matches the target text. Across these strategies, when the source changes, simpler is often safer.

cs.CL↗
arXiv2026

Beyond Visibility and Technical Reuse: Public Application Transformation in Open-Source Model Ecosystems

Open-source model platforms have made it easier to publish AI models, but model release alone does not reveal whether models become visible, technically reused, or incorporated into public applications. This study introduces public application transformation as a platform-visible dimension of model impact and examines it through structured Model-Space links on Hugging Face. We construct a platform-scale dataset of 2.56 million model repositories, 1.06 million Spaces, 810,087 dataset repositories, and 1.22 million account profiles, together with Model-Space, Dataset-Space, and model-to-model technical reuse links. The analysis shows that public application transformation is highly selective and concentrated: only a small share of models are linked to Spaces, and most Model-Space links are concentrated among a limited set of models. More importantly, application transformation is associated with platform visibility but is not equivalent to technical reuse, indicating that downloads, likes, downstream model reuse, and application-facing uptake capture different forms of model impact. Additional analyses show that application-transformed models tend to exhibit stronger metadata-based readiness and enter heterogeneous Space configurations involving datasets, SDKs, and task-specific application categories. By tracing how models move from repositories into public applications and demos, this study extends the measurement of open-source model impact from artifact availability and technical reuse to platform-mediated transformation across AI information objects.

cs.SE↗