arXiv subjects

Bing Liu

Publications and source records attributed to Bing Liu.

At least 19 recordsLinked to original sources

arXiv2026

Where Position Matters: Profiling-Guided Layer-Wise RoPE Allocation in Transformers

Rotary Position Embedding (RoPE) typically uses uniform coverage across Transformer layers despite heterogeneous positional requirements. Phase-off interventions reveal non-uniform, model-specific rotary-phase dependence, while coverage interventions relate it to coverage-upgrade benefits. We propose profiling-guided layer-wise RoPE allocation, assigning high/low coverage under a fixed budget using early coverage-benefit profiles. The framework supports fixed-width query/key representations and physical rotary-dimension allocation in Multi-head Latent Attention (MLA). Normalized-depth mapping transfers profiles within model families, enabling small-model measurements to guide larger-model allocations before pretraining. A brief analysis relates selection and transfer to score calibration and inter-layer interactions. From-scratch experiments demonstrate validation-loss improvements over available budget-matched baselines, while analytical resource accounting identifies favorable quality--storage--compute trade-offs for physical MLA. These results connect positional demand to transferable, efficient architecture design.

cs.LG↗

arXiv2026

Youtu-Parsing-Omni: One Encoder, One Schema, All Modalities

We present Youtu-Parsing-Omni, an omni-modal parser that converts documents, images, audio, and video into structured records grounded in space and time, providing a perception layer for agents that read, locate, and cite evidence across modalities. Its design rests on one encoder for every modality and one schema for every task. Youtu-Omni-Encoder, initialized from a pre-trained language model to strengthen text reading, encodes all modalities with shared weights in a unified space-time coordinate system, and a few fusion layers let video frames attend to concurrent audio, bringing cross-modal fusion and temporal alignment into perception itself. OmniSchema unifies the outputs of all ten parsing tasks, allowing a single self-evolving post-training procedure, OmniSchema-Aware On-Policy Distillation (OSAD), to provide a shared token-role-based objective across tasks. In each round, the model is trained on its own rollouts against token-level targets from a reference-conditioned copy of itself, which is then refreshed with the trained weights, so that the parser and its supervision co-evolve. With a single 5 B-parameter checkpoint, Youtu-Parsing-Omni achieves state-of-the-art performance on OmniDocBench v1.6 (96.96 Overall) and leads all open-weight models on OmniParsingBench (75.08 Avg.), with its advantage concentrated on audio and video. It also achieves the best overall chart-parsing score (95.02) and the highest exact chemical-fingerprint agreement among compared systems, and matches a dedicated optical-music model in character-level transcription of typeset sheet music while staying far ahead of general-purpose systems. Code and model weights are publicly available.

cs.CV↗

arXiv2026

VerifyMAS: Hypothesis Verification for Failure Attribution in LLM-based Multi-Agent Systems

Large language model-driven multi-agent systems (LLM-MAS) excel at complex tasks, yet unreliable agents remain a key bottleneck to system-level reliability. Automatic failure attribution is therefore critical, but existing approaches, such as direct prediction of agent-error pairs and agent-first failure attribution, rely on local logs of agents and miss global failures that only manifest over full interaction trajectories, such as cross-step inconsistencies and inter-agent coordination errors. Moreover, directly predicting failures induces a large combinatorial search space, hindering fine-grained attribution. To address these challenges, we propose VerifyMAS, a hypothesis verification framework for agent failure attribution. Instead of directly predicting faulty agents and error types, VerifyMAS formulates and verifies failure hypotheses against full trajectories. This verification-based approach decomposes attribution into trajectory-level error validation and fine-grained agent localization, providing an error-first attribution approach that captures global failure patterns while substantially reducing the search space. We further introduce a hypothesis-based data construction strategy grounded in a structured error taxonomy and fine-tune a specialized LLM verifier model for trajectory-level failure verification and agent attribution. Experiments on Aegis-Bench and Who&When show that VerifyMAS consistently improves diverse backbone models, including open-source Qwen and API-based GPT models, outperforming prior methods without sacrificing inference efficiency for long multi-agent trajectories.

cs.CL↗

arXiv2026

Look Back to Move Forward: Learning to Revisit for Agentic Compositional Image Generation

Text-to-image models have made remarkable progress in visual fidelity and instruction following, yet complex compositional generation remains challenging. Existing agentic approaches address this limitation through iterative generation and editing, adapting their actions based on interaction history. However, they typically fix the continuation state (i.e., the image from which refinement proceeds) to the latest image. Since iterative editing is not necessarily monotonic, the agent may continue refining from the latest image even when it has already degraded. In this work, we introduce GenRevisit, an agentic framework that extends action selection to joint continuation-state and action selection. The agent can not only use interaction history to adjust editing instructions but also select an image candidate to edit, jointly deciding where to continue from and how to refine. We train GenRevisit with SFT followed by agentic reinforcement learning. To address the coarse credit assignment of final outcome reward in GRPO, which assigns the same trajectory-level advantage to all actions, we further propose Normalized Action Credit Assignment (NACA) to redistribute positive advantage according to each action's contribution to reducing the remaining objective. On GenEval2, GenRevisit achieves GM scores of 84.3 and 74.4 with Qwen-Image-2512 and Z-Image-Turbo, surpassing the strongest baselines using the same generators by 12.7 and 10.9 points respectively. GenRevisit achieves state-of-the-art performance on GenEval, GenEval++, and GenEval2 among baselines.

cs.CV↗

arXiv2026

Probing $L_z$ Orbital Angular Momentum via Wavefunction-Interference-Induced Charge Order

Orbital angular momentum (OAM) characterization is central to understanding quantum materials such as valley Hall and Rashba systems and, through its connection to Berry curvature, provides key insight into band topology. Conventional OAM detection via photoemission dichroism, however, is susceptible to interference artifacts and is largely restricted to occupied states. Here, we demonstrate a real-space approach to map the $L_z$ OAM character using scanning tunneling microscopy in triangular atomic monolayers. Unlike photoemission, this method exploits interference effects arising from the $L_z$ orbital phase and the Bloch phase associated with neighboring atoms. This shifts the local density of states (LDOS) maxima to distinct Wyckoff positions between atoms, encoding $L_z$ information into characteristic real-space LDOS patterns. We exemplify this approach in 2D quantum materials with contrasting $L_z$ sequences, namely Tl/Si(111) and In/SiC(0001) monolayers, identifying $L_z$-dependent LDOS accumulation associated with distinct topological phases in both systems. Our results establish $L_z$-dependent charge localization as a proxy to atomic obstruction for energetically isolated states in triangular lattices, including transition-metal dichalcogenides.

cond-mat.mtrl-sci↗

arXiv2026

How Bregman Divergences Shape Shampoo

Understanding the principles behind Shampoo has recently guided the development of more effective neural network optimizers. These methods learn a preconditioner by optimizing the Frobenius or Kullback-Leibler (KL) divergence against the gradient second moment. In this work, we investigate how the choice of divergence shapes preconditioning, which remains unclear and blocks further improvements. To do so, we develop a unified Bregman divergence framework that connects all popular divergences, allowing us to study them jointly. Through empirical spectral analysis of gradient second moments, we examine how divergence choice shapes Kronecker approximation and interacts with finite-sample error in preconditioning. We find that some divergences can better compensate for finite-sample underestimation of the empirical second moment, helping explain the differing behavior of their corresponding Shampoo variants. We further validate this explanation through GPT-2 pretraining experiments. By connecting divergence choice to practical training behavior, we believe our framework provides principled guidance for understanding the foundations of, and further improving, Shampoo.

cs.LG↗

arXiv2026

Experimental realization of the minimal two-dimensional multi-orbital kagome model

Kagome materials have emerged as a major platform for correlated and topological quantum matter, hosting flat bands, Dirac dispersions and van Hove singularities rooted in frustrated lattice geometry. While their essential physics is epitomized by the canonical single-orbital kagome model, real kagome materials are intrinsically multi-orbital and typically chemically and structurally complex, with additional three-dimensional coupling making kagome geometry and orbital degrees of freedom difficult to disentangle. Here we realize a truly two-dimensional, elemental two-orbital kagome system in monolayer Sb on SiC(0001). Substrate-induced orbital filtering isolates the in-plane Sb p_x/p_y orbitals into a six-band kagome manifold, establishing a chemically simple, minimal platform for multi-orbital kagome physics. We demonstrate two consequences that distinguish this system from the canonical single-orbital model. First, chemically enforced half filling pins the Fermi level to a closed nodal line, whose finite density of states drives a unit-cell-conserving breathing instability and opens a giant insulating gap. Second, we uncover orbital-resolved atomic obstruction: individual orbital-derived manifolds realize obstructed atomic limits, whereas their combined half-filled valence manifold is non-obstructed. Our results establish kagome antimonene as a benchmark system for exploring orbital-driven electronic, structural and topological kagome physics in the genuine two-dimensional limit.

cond-mat.mtrl-sci↗

arXiv2026

Stable Neural Decoding Across Sessions via Task-Conditioned Latent Alignment for Brain-Machine Interfaces

Achieving stable long-term neural decoding in invasive brain-machine interfaces (BMIs) remains challenging due to variations in recorded neural populations across sessions. Current latent alignment approaches may overlook task-dependent structure during cross-session adaptation. We propose Task-Conditioned Latent Alignment (TCLA), a framework that stabilizes neural decoding by learning a shared latent space. TCLA learns a low-dimensional source representation using neural reconstruction and continuous behavioral supervision. During target-session adaptation, the shared representation is fixed, while target neural activity is mapped into the source latent space by aligning source and target distributions separately for each task condition. We evaluated TCLA on seven nonhuman primate datasets spanning multiple tasks. In long-term cross-session evaluation, TCLA achieved a mean $R^2$ of $0.476\pm0.014$ with a negative $R^2$ failure rate of only 6.8\%. Across 1,356 within-subject session pairs, TCLA achieved a mean $R^2$ of $0.371\pm0.009$ with a failure rate of 6.8\%. Across 2,134 cross-subject session pairs, TCLA achieved a mean $R^2$ of $0.218\pm0.004$ with a failure rate of 12.9\%, substantially better than those of the comparison methods. These results demonstrate that by preserving behaviorally relevant and task-dependent latent structure, TCLA improves the robustness of neural decoding across recording sessions and subjects. The source code is publicly available at \href{https://github.com/FAMD-CASIA/TCLA}{https://github.com/FAMD-CASIA/TCLA}.

cs.LG↗

arXiv2026

Measurement of Cosmic-Ray Density in the Spiral Arms toward Galactic Anticenter

Using nearly 17 years of Fermi-LAT data, we analyzed the diffuse gamma-ray emission associated with different spiral arms of the Milky Way in the ranges $l = 105^\circ$ to $145^\circ$ and $b = -5^\circ$ to $5^\circ$. The gamma-ray emissions from these spiral arms all exhibited clear pion-bump features, which can be well explained by a power-law spectrum of cosmic-ray protons. The spectral indices corresponding to the three distinct spiral arms -- the Local Arm, the Perseus Arm, and the Outer Arm -- are found to be approximately $-2.75$, $-2.55$, and $-2.80$, respectively. The energy densities of the cosmic rays above 10 GeV are estimated to be all around 0.2$\rm~eV~cm^{-3}$. We further compared our results with previous studies, which have established various propagation models predicting that cosmic ray density decreases and spectra soften with increasing Galactocentric radius. We found higher densities and harder spectral indices than predicted in the outer Galaxy, suggesting possible contributions from a larger halo, slower magnetic field decay, or a combination of these and other effects. In particular, the spectral hardening observed in the Perseus Arm may be attributed to locally enhanced cosmic-ray acceleration.

astro-ph.HE↗

arXiv2026

Gamma-ray Emission from the S147 Region: Indication of Escaping Cosmic Rays Interacting with Molecular Clouds

We present a detailed analysis of $γ$-ray emission from the middle-aged supernova remnant (SNR) S147 (G180.0$-$1.7) using approximately 16.5 years of Fermi-LAT data. Spatially, a new extended $γ$-ray component distinct from the emission associated with the H$α$ filaments of the SNR shell is identified. This new component exhibits a strong spatial correlation with dense molecular clouds (MCs) identified in CO emission at Local Standard of Rest velocities of $0$--$5\,\mathrm{km\,s^{-1}}$. Spectrally, the cloud-associated emission implies an underlying cosmic-ray (CR) proton population described by a hard power-law with an index of $Γ\approx 2.1$, compatible with the standard diffusive shock acceleration prediction. We interpret the $γ$-ray emission in this region with a hadronic scenario involving two distinct CR populations: trapped CRs reaccelerated within the radiative SNR shell as proposed in previous work, and escaping CRs illuminating the nearby MCs. The derived CR proton intensity in the MC region significantly exceeds the local Galactic background measured by AMS-02, consistent with the interpretation that the cloud is illuminated by particles accelerated by S147. These findings provide observational support for a CR-escape scenario during the earlier evolutionary phases of this middle-aged SNR and highlight the S147 MC component as a potential candidate for detection at TeV energies by LHAASO.

astro-ph.HE↗

arXiv2026

InteractBench: Benchmarking LLMs on Competitive Programming under Unrevealed Information

Competitive programming is increasingly being used to evaluate the algorithmic reasoning capabilities of large language models (LLMs). However, existing benchmarks primarily focus on full-information tasks where all problem inputs are provided upfront. This overlooks a critical dimension of algorithmic reasoning: the ability of generated programs to operate when key information is not revealed upfront. Interactive problems, a distinctive component of competitive programming, embody this challenge. These problems require programs to engage in multi-round interaction with an interactor (a judge program) under strict protocol constraints and limited query budgets, with new information revealed only in response to queries. To address this gap, we introduce InteractBench, a benchmark comprising 322 high-quality interactive problems curated from Codeforces, AtCoder, IOI, and ICPC. Each problem is packaged with executable local interactors, enabling fully offline evaluation. Unlike existing benchmarks, InteractBench assesses whether model-generated code can acquire information and track state dynamically. Our evaluation reveals a significant interaction gap: even the most advanced reasoning models achieve limited success on interactive problems. Beyond success rates, we propose a fine-grained failure taxonomy to diagnose the root causes of these deficiencies. Although algorithmic logic errors remain dominant, protocol violations and query-budget overruns are frequent. Code is available at https://github.com/kmsgk0/InteractBench.

cs.SE↗

arXiv2026

Primitive-Driven Compositional Forensic Visual Prompting for Open-World Face Anti-Spoofing

Open-world face anti-spoofing must address both covariate and semantic shifts: source and target domains differ in imaging conditions, while target domains contain diverse attack types absent from training. Existing prompt-based approaches often express spoofing through category semantics or language guidance, which is effective for modeling high-level concepts but is less suited to explicitly capturing the evolving fine-grained and spatially heterogeneous forensic evidence of unseen attacks. Motivated by the hypothesis that many unseen attacks can be characterized by new combinations of recurring visual cues, we propose a compositional forensic visual prompt learning framework that operates entirely in the visual feature space. Built on a frozen ViT-based vision foundation model, the framework employs patch-aware attention to refine a shared set of learnable micro-forensic primitives into localized forensic evidence units derived from image patches. Class-specific global contextual prompts then provide input-dependent routing weights that adaptively select and compose these primitives into compositional forensic visual prompts for real/spoof discrimination. The primitives are not assigned predefined semantic meanings; instead, their specialization and reuse emerge from shared parameterization and joint optimization across categories. Extensive experiments on nine open-world protocols demonstrate state-of-the-art performance, strong cross-domain generalization, and robust adaptation to unseen attacks.

cs.CV↗

arXiv2026

DeMixPert: Decomposed Response Modeling with Gaussian Mixtures for OOD Single-Cell Perturbation Prediction

Predicting transcriptome-wide responses to unseen genetic perturbations remains a major computational challenge because accurate prediction requires recovering both perturbation-specific transcriptional shifts and heterogeneous cellular responses. Existing methods often entangle deterministic response structure with stochastic population-level variation, causing dominant shared patterns to mask weaker perturbation-specific signals and impair distributional modeling. To address these challenges, we propose \textbf{DeMixPert}, an approach for Decomposed response Modeling with Gaussian Mixtures for Out-Of-Distribution (OOD) single-cell Perturbation prediction. DeMixPert decomposes perturbation-induced changes into a basal-state-dependent systematic response, a perturbation-specific response, and population-level variation. The systematic component is derived from the basal state encoded from control-cell expression, whereas the perturbation-specific component is inferred from pretrained target embeddings for unseen-target generalization. DeMixPert models population-level variation using a Gaussian prototype Invertible Network and adaptively combines reusable Gaussian prototypes according to the basal state and perturbation condition. The resulting mixture is mapped to a condition-specific variation distribution. Sampled variations are integrated with the systematic and perturbation-specific components, followed by joint decoding with the basal state to reconstruct perturbed-cell gene expression. Experimental results show that DeMixPert effectively captures heterogeneous single-cell perturbation responses and achieves superior performance across unseen-perturbation settings. The source code is made publicly available upon publication.

cs.LG↗

arXiv2026

Row-Stochastic Matrices Can Provably Outperform Doubly Stochastic Matrices in Decentralized Learning

Decentralized learning often involves a weighted global loss with heterogeneous node weights $λ$. We revisit two natural strategies for incorporating these weights: (i) embedding them into the local losses to retain a uniform weight (and thus a doubly stochastic matrix), and (ii) keeping the original losses while employing a $λ$-induced row-stochastic matrix. Although prior work shows that both strategies target the same $λ$-weighted global loss, it remains unclear whether the Euclidean-space guarantees are tight and what fundamentally differentiates their behaviors. To clarify this, we develop a weighted Hilbert-space framework $L^2(λ;\mathbb{R}^d)$ and obtain convergence rates that are strictly tighter than those from standard Euclidean analysis. In this geometry, the row-stochastic matrix becomes \emph{self-adjoint} whereas the doubly stochastic one does not, creating additional \emph{penalty terms} that amplify consensus error, thereby slowing convergence. Consequently, the difference in convergence arises not only from spectral gaps but also from these penalty terms. We then derive sufficient conditions under which the row-stochastic design converges faster even with a smaller spectral gap. Finally, by using a Rayleigh-quotient and Loewner-order eigenvalue comparison, we further obtain topology conditions that guarantee this advantage and yield practical topology-design guidelines.

cs.LG↗

arXiv2026

Identification of a Large-Scale Diffuse Gamma-Ray Structure in the Southern Galactic Hemisphere

We identify and characterize a large-scale diffuse gamma-ray structure in the Southern Galactic Hemisphere using 17 yr of Fermi-LAT data. An energy-dependent likelihood analysis, including alternative Galactic diffuse-emission models, isotropic emission, the Fermi bubbles, and resolved 4FGL sources, reveals an extended excess that persists across the tested background models and spans tens of degrees. The excess broadly follows the X-ray-defined southern eROSITA Bubble (eB) region, while also overlapping the projected southern extension of Loop I. Template fits favor a filled eB-like morphology over the adopted Wolleben Loop I shell geometry, making the structure a plausible gamma-ray counterpart of the southern eB, although Loop-I-related or other localized foreground emission cannot be excluded. Under the eB template, the southern component is fainter and softer than the northern large-scale component, with an integrated luminosity lower by a factor of about seven, broadly consistent with the eROSITA-bubble asymmetry. If interpreted as Galactic-scale outflow emission, its faint, soft spectrum may indicate aged particles and/or distributed reacceleration in the outer bubble. A hadronic interpretation is energetically demanding, whereas a leptonic inverse-Compton scenario is more economical but requires rapid transport and/or local reacceleration of high-energy electrons.

astro-ph.HE↗

arXiv2026

Resonantly-enhanced Raman response in graphene-capped bismuthene on SiC

Two-dimensional quantum spin Hall insulators based on atomic monolayers offer a promising route toward dissipationless electronics, yet their practical use is often limited by environmental instability. Encapsulating the system with a graphene capping layer has been shown to be a reliable method to prevent oxidation and degradation. However, the confirmation of a successful encapsulation still relies on ultra-high vacuum techniques, that considerably slow the process. Here, we present an ex situ, rapid, nondestructive and spatially resolved Raman characterization of graphene-capped bismuthene, a honeycomb monolayer of Bi on SiC. A pronounced Raman scattering peak at around 122 cm-1 is identified as the E2g phonon of bismuthene, via a comparison with density functional perturbation theory calculations. We use excitation-energy and polarization-dependent Raman measurements to enable an unambiguous assignment of the spectral features. Tuning the excitation energy close to the excitonic transition in pristine bismuthene, we observe a strong enhancement of the Raman response and the emergence of additional scattering peaks. In this regime, higher-order phonon features, as well as interfacial modes between bismuthene and the SiC substrate, become visible, suggesting the involvement of resonant scattering processes. Our results establish Raman micro-spectroscopy as a versatile tool for probing graphene-protected quantum materials, providing access to lattice dynamics and interlayer coupling.

cond-mat.mtrl-sci↗

arXiv2026

Moiré-resonant surface state in ultrathin RuO$_2$

RuO$_2$ has emerged as a prototypical candidate for altermagnetism. In the face of daunting evidence for magnetic order in the bulk, the focus naturally shifted to surfaces and ultrathin films, where Coulomb interactions are dimensionally quenched and electron correlations strongly enhanced. Here, we examine atomically ordered, ultrathin RuO$_2$(110) grown on Ru(0001) using a combination of scanning tunneling microscopy (STM), density functional theory, and density matrix renormalization group methods. We observe a nonmagnetic charge order that is imprinted by the incommensurate moiré stacking with the substrate and enhanced by the electronic Fermi surface scattering within the flat-band surface state. We further identify a nonmagnetic, metastable $c(2 \times 2)$ surface reconstruction that arises from surface phonon softening and can be toggled reversibly via STM tip manipulation. Spin-polarized STM measurements, however, reveal no evidence of magnetic order on the RuO$_2$(110) surface. Our findings of a nonmagnetic charge-modulation position ultrathin RuO$_2$(110) as an intriguing platform for exploring moiré-assisted electronic orders.

cond-mat.mtrl-sci↗

arXiv2026

Demand-side decarbonisation at scale via MaaS-integrated carbon incentives

Digital carbon incentives are increasingly used to promote low-carbon travel, but city-scale evidence on their behavioral and carbon-accounting implications remains limited. We evaluated a carbon-incentive program on a Beijing Mobility-as-a-Service platform using 4.82 billion trips from 2.96 million users over 13 months, with a matched panel of enrolled and non-enrolled users. Enrollment was associated with a 20.3 percentage-point increase in the monthly low-carbon travel share, with pre-enrollment trends remaining near zero across event-study tests. A random-forest accounting scenario trained on pre-enrollment data implied a 1.8% citywide decline in gasoline-car trips and annual reductions of 94,353 tonnes of CO2, equivalent to 5.7% of the certified reductions traded in Beijing's carbon market in 2023. This estimate is model-dependent rather than a directly observed or causally identified program effect. Larger program-associated responses were concentrated in areas with greater transit access. These results show that carbon-market-financed digital incentives can support measurable low-carbon travel responses at the city scale.

stat.AP↗