arXiv subjects

Yi Zhang

Publications and source records attributed to Yi Zhang.

At least 127 records · Page 7Linked to original sources

arXiv2026

SEER: Long-Context Reasoning via Selective Visual-Text Compression

Long-context reasoning remains computationally expensive for large language models due to the quadratic complexity of attention over text tokens. Visual-text compression offers a promising alternative by rendering text into images and processing them with vision-language models, often reducing token usage. However, existing approaches apply uniform compression regardless of query relevance, potentially sacrificing precision where detailed extraction is required. We present SEER, a framework that learns to select query-relevant images through visual scanning and retrieve textual content only where needed, combining the efficiency of visual compression with the precision of text-based reasoning. Through supervised fine-tuning on tool-interaction trajectories, SEER learns adaptive tool invocation for selection and retrieval. Experiments on long-context benchmarks show that SEER improves extraction precision through selective text retrieval while retaining average prompt-token savings relative to full-text baselines. On LongBench, SEER achieves 51.11% average accuracy, outperforming the visual-text baseline Glyph-9B by 2.33 points and Qwen3-8B by 3.49 points. Code can be accessed at https://github.com/jiaweixu98/SEER

cs.CL↗

arXiv2026

Scaling Representation Diversity: Modulated Attention and Reconstructive Regularization for Visual Grounding

Referring Expression Comprehension (REC) is commonly studied under dataset-specific fine-tuning, resulting in specialist models with limited cross-dataset generalization. In this work, we revisit REC from the perspective of unified open-vocabulary grounding and identify representation degeneration as a key obstacle to scaling a single generalist model. To preserve representation diversity, we propose a holistic data-model co-design framework. Architecturally, we introduce the Modulated Attention-Contrastive Head (mACH) for efficient token-level vision-language alignment and a text-conditioned JEPA auxiliary stream that provides complementary gradient support to preserve alignment-active representations without inference overhead. On the data side, we introduce Objects365-Caption, enriching Objects365 with context-aware referring expressions for large-scale language supervision. We further provide a theoretical analysis showing that complementary gradient subspaces preserve alignment capacity and thereby scale representation diversity. Extensive experiments demonstrate that our single-checkpoint framework achieves highly competitive performance on standard REC benchmarks while exhibiting strong generalization across heterogeneous grounding datasets without benchmark-specific adaptation.

cs.CV↗

arXiv2026

ASCon: A Direction-Aware Reciprocal Agent--Step Contextualization Model for Failure Attribution in Multi-Agent Systems

Failure attribution in LLM-based multi-agent systems (MAS) aims to answer who caused failures, when they occurred, and why by identifying responsible targets including faulty agents, erroneous steps, and failure modes. Existing methods have primarily focused on developing dedicated models for specific attribution targets, with limited attention to the evidential dependencies among them. Despite these attribution targets are different, they rely on common diagnostic evidence from MAS trajectories, including task constraints, agent roles, behavioral histories and inter-agent interactions. This commonality motivates us to develop a unified representation model that aggregates the trajectory evidence into individual agent and step representations, which can subsequently be adapted to different attribution targets. Accordingly, we propose ASCon, a direction-aware reciprocal \textbf{A}gent--\textbf{S}tep \textbf{Con}textualization model for multiple failure attribution targets. ASCon introduces direction-aware graph attention to model execution context, masked step-to-agent attention to construct behavior-aware agent representations, and agent-conditioned step contextualization to incorporate agent context back into step representations. The resulting contextualized representations enable different attribution targets through lightweight target-specific heads. Experiments show that ASCon can improve faulty-agent detection by 5.83\%+ in micro-accuracy, faulty-step detection by 10.63\%+ in micro-accuracy, and failure-mode detection by 14.73\%+ in Macro-F1. Meanwhile, it can also substantially enhance the LLM-based methods' attribution capabilities in out-of-domain scenarios.

cs.MA↗

arXiv2026

Threshold Structure of Optimal Policies in Restart POMDPs

We study a Restart POMDP (Partially Observable Markov Decision Process) on a general Borel state space, where the controller either lets the hidden state evolve unobserved or restarts the system and observes the new state. Exploiting a sufficient-statistic representation consisting of the last observed state and the elapsed time since restart, we reduce the problem to a fully observed MDP. Under a natural one-step cost deterioration condition, we prove that optimal policies have a threshold structure in the elapsed time for both the discounted and total undiscounted cost criteria. When the state space is partially ordered and the kernel is stochastically monotone, we further show that the optimal threshold is nonincreasing in the state. For the average cost criterion, under additional assumptions of geometric ergodicity and domination of the transient gain, we establish analogous threshold results via the vanishing discount approach, after showing the uniform boundedness of the optimal thresholds and relative value functions.

math.OC↗

arXiv2026

In-Loop Model Adaptation with Coupled Latent-Noise Guidance for High-Fidelity Subject-Driven Text-to-Image Generation

Text-to-image diffusion models have achieved remarkable success in generating high-quality images from a given text prompt. Subject-driven generation aims to synthesize customized images to mimic the appearance of subjects in given reference images within different visual contexts specified by the text prompts. The central challenge here is that, when the reference image changes, the diffusion model cannot efficiently adapt to different visual contexts while consistently maintaining the subject identity. Existing methods either train the model with a large domain-specific dataset or fine-tune the model using the reference image for hundreds of iterations before actual image generation. In this work, we explore a new approach, called \textit{In-Loop Model Adaptation} (IMA), which adapts the core diffusion model at each generation step during the actual process of image generation, without being trained on the reference image before the generation process. To this end, we establish a DDIM inversion chain that maps the reference image to a sequence of latent, as well as a text-to-image generation chain which generates the image from the text prompt only. We then introduce a masked latent consistency loss and a noise regularization loss to characterize the latent-noise difference between the diffusion model and these two chains at each generation step. This coupled latent-noise loss is used to guide the in-loop model adaptation to preserve the subject identity specified by the reference image while maintaining accurate alignment with the text prompt, resulting in high-fidelity text-to-image generation. Our extensive experiments demonstrate that our proposed IMA method significantly improves the performance of subject-driven text-to-image generation.

cs.CV↗

arXiv2026

Equilibrium and Selfish Behavior in Network Contagion

In this paper we consider non-atomic games in populations that are provided with a choice of preventive policies to act against a contagion spreading amongst interacting populations, be it biological organisms or connected computing devices. The spreading model of the contagion is the standard SIR model. Each participant of the population has a choice from amongst a set of precautionary policies with each policy presenting a payoff or utility, which we assume is the same within each group, the risk being the possibility of infection. The policy groups interact with each other. We also define a network model to model interactions between different population sets. The population sets reside at nodes of the network and follow policies available at that node. We define game-theoretic models and study the inefficiency of allowing for individual decision making, as opposed to centralized control. We study the computational aspects as well.

cs.GT↗

arXiv2026

The elementary theory of full $n$-branching ordinal trees with successor functions

We investigate the first-order theories of full $n$-branching ordinal trees $\mathfrak{T}_α^n$. We obtain a complete classification of the standard trees up to elementary equivalence: every $\mathfrak{T}_α^n$ is elementarily equivalent to one of four canonical types determined by the ordinal $α$. For each canonical type we establish effective quantifier elimination and prove decidability of the theory. Along the way we develop the realizability theory of colored ordinal characters and obtain a sharp bound on the length of minimal witnesses. We also clarify the relationship between these trees and monadic second-order logic over ordinals, and show that the equal-height relation is not first-order definable in any standard tree.

math.LO↗

arXiv2026

TrustRoboReward: Preference-Ordered Isotonic Score Editing for Multi-Paradigm Robot Reward Models

Reward models are a bottleneck for reinforcement learning in embodied AI. Long-horizon robotic manipulation requires scalable vision feedback beyond handcrafted rewards or task-specific annotations. Existing open-source VLM reward judges like RoboReward adopt simple 1--5 trajectory progress scoring, lacking pairwise preferences for RLHF, DPO and Bradley-Terry frameworks, while failing to optimize video scene understanding. Augmenting RoboReward with pairwise comparison and video-QA supervision causes inconsistency between pairwise preferences and pointwise scores, introducing training noise and hurting downstream performance---an issue aggregation methods such as TrustJudge cannot resolve. To address this, we propose TrustRoboReward, a multi-paradigm reward modeling framework equipped with Preference-Ordered Isotonic Score Editing (POISE). We construct a unified four-paradigm dataset with trajectory progress scoring (Score-A), video-QA answer quality scoring (Score-B), and their pairwise counterparts (Pair-A, Pair-B). Pairwise labels align better with human judgment than pointwise scores, inspiring us to calibrate pointwise scores to avoid score-pair reversals against pairwise preferences. POISE rectifies pointwise scores and eliminates cross-paradigm reversal conflicts unresolved by TrustJudge. Theoretically, POISE reduces score-pair reversal conflicts from 20.15% to 0%, whereas TrustJudge retains 20.46% conflicts on the same corpus. Evaluated on our benchmark, Qwen3-VL-4B trained with POISE achieves an overall reward score of 77.96%, nearly matching GPT-5-mini (78.09%, gap 0.13%) and outperforming the strongest RoboReward-4B baseline by 10.13%. It also lifts test-time score-pair consistency to 71.90%, exceeding RoboReward-4B (57.26%) and GPT-5-mini (68.09%). Integrating TrustJudge aggregation during inference boosts the overall score to 78.57%, surpassing the GPT-5-mini teacher model.

cs.AI↗

arXiv2026

From Errors to Rules: Iterative Prompt Optimization for Text Classification

Prompt optimization for text classification spans diverse approaches, from demonstration selection to exploration-based search to error-driven diagnosis, each with known but incompletely characterized strengths and limitations. We conduct a comprehensive empirical study across diverse classification benchmarks (2 to 150 classes) comparing these paradigms through both quantitative evaluation and qualitative analysis of optimization traces, revealing that each paradigm excels on structurally different task types and that no single method dominates. Guided by these insights, we propose Error-Guided Optimization (ERGO), an error-driven method that iterates over the full training set in non-overlapping batches, diagnoses classification failures, and generates targeted decision rules through a diagnose-prescribe-rewrite feedback loop. ERGO achieves the best accuracy on tasks where errors concentrate in specific confused label pairs (which we term boundary-learnable tasks): TREC: 90.0%, CLINC150: 94.4%, converges in 3-5 iterations, and produces interpretable decision rules. While ERGO does not achieve the highest overall average, it fills a complementary role: demonstration-based ICL wins on coverage-dependent tasks, exploration-based search wins on many-class intent, and ERGO wins where decision boundaries are learnable from error patterns. We provide a complementarity framework linking task characteristics to optimal paradigm selection, offering practical guidance for practitioners.

cs.AI↗

arXiv2026

An Early Warning of Emerging Biosecurity Risks in Frontier LLMs

Frontier large language models (LLMs) are increasingly integrated into scientific workflows, yet their growing biological capabilities may outpace current safeguards. To assess the biological risks of frontier models, we develop Intern-BioBreaker, a specialized bio-red-teaming model, together with an integrated computational-to-physical framework that couples model-level stress testing with wet-lab validation. Within this framework, Intern-BioBreaker generates targeted jailbreak prompts to test whether aligned models can be induced to provide operational guidance for safety-sensitive biological tasks or produce sequence-level outputs with potentially harmful properties. Selected sequence outputs are then carried forward for DNA synthesis, host expression, and orthogonal protein verification to assess whether model-generated designs can yield the intended biological products. Our evaluation reveals a concerning gap between text-level safeguards and the risks posed by capable scientific models: (i) Intern-BioBreaker outperforms baseline attack models and reveals widespread bio-risk jailbreak vulnerabilities across both open-weight and proprietary frontier LLMs, with several targets reaching near-saturated or 100% task-level attack success rate (ASR); (ii) in sequence-level case studies, GPT-5.5 can be induced to generate modified viral candidate sequences with pathogenic potential; the corresponding translated proteins may exhibit even stronger receptor-binding affinity and thus enhanced infection potential; and (iii) end-to-end verification shows that selected model-generated biological designs are not merely textual artifacts, but can be physically realized under controlled experimental settings. These findings underscore the need for stronger biological red-teaming, nucleic acid synthesis screening, and safety mechanisms that keep pace with model capabilities.

cs.CL↗

arXiv2026

Differential 6-DOF Pose Estimation with Provable First-Order Immunity to Camera Calibration Errors

Accurate six-degree-of-freedom (6-DOF) motion estimation is essential for robotic manipulation, autonomous systems, and structural displacement monitoring. Conventional 3D-2D methods estimate absolute camera poses independently at each time and recover platform motion through camera-to-platform extrinsics, making them sensitive to extrinsic calibration errors, especially for micromotion. We present a differential pose estimation method that directly recovers platform motion from inter-frame image displacements and known 3D control points. By differencing perspective projection equations, using a depth-invariance approximation, and modeling motion on SE(3), the method avoids independent absolute-pose estimation and supports both monocular and multi-camera systems. We prove that translational extrinsic errors cancel exactly, while rotational errors induce a bounded perturbation determined by calibration error, motion magnitude, and observation geometry. We also derive generic observability conditions, a Cramer-Rao lower bound, and a bias-eliminated consistent estimator, and characterize the validity limits of the approximations. Extensive synthetic and real-world experiments establish a new state of the art for 6-DOF platform micromotion estimation, outperforming representative PnP and generalized-PnP methods in accuracy, calibration robustness, and computational efficiency. With five control points and 0.5-pixel image noise, the monocular solver obtains a combined pitch-yaw rotation RMSE of 10.09 arcsec, a translation RMSE of 3.70 mm, and a runtime of 0.34 ms. The binocular solver achieves a rotation RMSE of 10.58 arcsec, a translation RMSE of 3.91 mm, and a runtime of 0.27 ms. Code will be released upon publication at https://github.com/zyoungszu/pami2026.

cs.CV↗

arXiv2026

Topological Rainbow Trapping for Spatial-frequency Demultiplexing of Underwater Acoustic Signals

Efficient separation and localization of multifrequency acoustic waves are essential for underwater target recognition and acoustic energy harvesting. The underwater implementation of topological rainbow trapping remains challenging because of complex fluid-solid interactions and the difficulty of integrating long-range transport with frequency-selective localization in an open system. Here, we theoretically develop and experimentally demonstrate two underwater spatial-frequency demultiplexing mechanisms based on the acoustic analogues of the QVHE and QSHE. Both mechanisms employ SSAWs, whose fields are confined near a structured surface and decay evanescently into the surrounding water, enabling experiments without an enclosed waveguide. In the QVHE mechanism, a spatial gradient along a valley-Hall edge channel shifts the local edge-state dispersion, causing different frequency components to become localized at distinct positions and thereby realizing spectral and spatial demultiplexing. In the QSHE mechanism, one-dimensional topological edge states are coupled to frequency-selective zero-dimensional higher-order corner states. Multifrequency signals first propagate robustly along a common boundary and are then transferred to prescribed remote corners according to frequency, producing a transport-then-confinement process. This mechanism combines defect-tolerant edge transport, frequency-selective corner localization, and remote rainbow trapping. Numerical simulations and experiments verify the frequency-dependent localization and the persistence of the designed transport pathways in the presence of structural defects. The proposed open SSAW platform performs robust frequency demultiplexing at the physical layer, reducing reliance on digital signal processing and offering potential for underwater target recognition and frequency-selective acoustic energy harvesting.

physics.app-ph↗

arXiv2026

Towards Effective Federated Multimodal Graph Learning via Navigating Multifaceted Heterogeneity

Multimodal-attributed graphs (MAGs), where nodes carry heterogeneous semantic content across multiple modalities while edges encode relational dependencies, have been widely adopted across diverse domains. Federated multimodal graph learning (FMGL) extends federated graph learning (FGL) to MAGs, enabling collaborative optimization across decentralized MAGs without exposing raw data. However, naively applying existing FGL methods to FMGL is insufficient, as they fail to navigate the multifaceted heterogeneity inherent in decentralized MAGs, including task heterogeneity across diverse client objectives, modality heterogeneity from discrepant modality quality and semantic domains, and topology heterogeneity arising from divergent topological patterns with low cross-modality correlation. To address these challenges, we propose Federated multimodal graph learning with Topology-aware Cross-modal Routing (FedTCR), the first systematic algorithm designed for FMGL. To handle task heterogeneity, FedTCR employs a two-stage paradigm that comprises federated task-agnostic pre-training followed by isolated task-oriented fine-tuning. To jointly address modality and topology heterogeneity, FedTCR introduces a topology-aware cross-modal routing mechanism. Concretely, each client distills modality-specific knowledge into compact prototypes via topology-aware importance-weighted aggregation informed by graph structure; the server then evaluates cross-client cross-modal relationships among these structure-informed prototypes and routes informative ones as contrastive references, driving a tri-level cross-modal contrastive learning scheme that jointly aligns cross-client modalities while preserving discrimination. Experiments across 7 domains demonstrate that FedTCR outperforms state-of-the-art baselines on both graph-centric and modality-centric tasks.

cs.LG↗

arXiv2026

SaFRO: Satisfaction-Aware Fusion via Dual-Relative Policy Optimization for Short-Video Search

Multi-Task Fusion plays a pivotal role in industrial short-video search systems by aggregating heterogeneous prediction signals into a unified ranking score. However, existing approaches predominantly optimize for immediate engagement metrics, which often fail to align with long-term user satisfaction. While Reinforcement Learning (RL) offers a promising avenue for user satisfaction optimization, its direct application to search scenarios is non-trivial due to the inherent data sparsity and intent constraints compared to recommendation feeds. To this end, we propose SaFRO, a novel framework designed to optimize user satisfaction in short-video search. We first construct a satisfaction-aware reward model that utilizes query-level behavioral proxies to capture holistic user satisfaction beyond item-level interactions. Then we introduce Dual-Relative Policy Optimization (DRPO), an efficient policy learning method that updates the fusion policy through relative preference comparisons within groups and across batches. Furthermore, we design a Task-Relation-Aware Fusion module to explicitly model the interdependencies among different objectives, enabling context-sensitive weight adaptation. Extensive offline evaluations and large-scale online A/B tests on Kuaishou short-video search platform demonstrate that SaFRO significantly outperforms state-of-the-art baselines, delivering substantial gains in both short-term ranking quality and long-term user retention.

cs.IR↗

arXiv2026

Thinking Once Is Enough: Intermediate-Layer Evidence Routing for High-Resolution VQA

High-resolution visual question answering (HR-VQA) is often treated as a problem of insufficient evidence acquisition, where failing multimodal large language models must inspect images again through cropping, re-encoding, or multi-round search. We show that this view is incomplete: in many cases, fine-grained evidence has already survived visual encoding and become identifiable and influential within an intermediate-layer routing window, but is later diluted before answer generation. We propose Thinking-Once, a \textbf{training-free, single-visual-pass} evidence-routing method that reconstructs question-conditioned attention at this window, preserves core entity tokens and compact background context, and routes this evidence to later layers without extra visual encoding. Across five base models, Thinking-Once consistently improves or matches the corresponding base setting, increasing the average scores on V$^*$Bench, HRBench-4K, and HRBench-8K by \textit{+3.1}, \textit{+3.0}, and \textit{+2.7} points while reducing the average peak memory by about 4,GB. On Qwen2.5-VL-7B, it improves the three benchmarks by \textit{+9.9}, \textit{+4.6}, and \textit{+5.5} points, raising the cross-benchmark mean from 72.5 to 79.1. With the ZwZ-8B base model, Thinking-Once reaches a mean score of 82.7. Against 11 open-source HR-VQA baselines, it obtains the best or tied-best score on all three benchmark averages and the best overall mean; for example, compared with DeepScan, it reduces V$^*$Bench inference time by \textbf{97.2\%} while improving the cross-benchmark mean from 77.8 to 79.1. These results show that HR-VQA can be improved by routing already encoded evidence rather than repeatedly acquiring new visual inputs. Code is available in the appendix.

cs.CV↗

arXiv2026

Photon Gas Thermodynamics in Doubly Special Relativity at the Planck Scale

We investigate the thermodynamics of a photon gas within the Magueijo-Smolin formulation of doubly special relativity, a framework that augments the speed of light with an observer-independent energy scale of the order of the Planck energy. We derive the logarithmic grand partition function, and the complete set of thermodynamic quantities for the photon gas, including the Helmholtz free energy, internal energy, entropy, pressure, and heat capacity, and perform their numerical evaluation. Our results smoothly reduce to the conventional special relativistic expressions in the limit where the invariant energy scale tends to infinity. For temperatures approaching the Planck scale, the finite cutoff induces a systematic suppression of all thermodynamic functions relative to their standard special-relativistic counterparts.

hep-ph↗

arXiv2026

Transition to double-cell mock Walker circulations with surface warming explained by periodic convection

Idealized mock Walker simulations are widely used to study the interactions between overturning circulation and convection in the tropics. Previous studies documented a transition from a single-cell to a double-cell mock Walker circulation when the average sea surface temperature exceeds 300 K. Here, we ascribe the transition to the emergence of periodic convection with warming due to stronger convectively-coupled waves. In cold simulations, the warm pool is dominated by steady deep convection, which results in a single overturning cell. In hot simulations, the warm pool is alternately dominated by deep convection and a stratiform mode, resulting in lower and upper cells respectively. This study suggests that the Walker circulation in a warmer climate may feature complex structural changes in addition to a weakening in strength, and highlights the profound impacts of convection on overturning circulation.

physics.ao-ph↗

arXiv2026

A Human-Augmenting Agentic Workflow for Observational Causal Inference

Data analysis agents are becoming increasingly common tools for applied and scientific research. Yet, for highly specialized tasks such as Observational Causal Inference (OCI), human oversight remains necessary to ensure the validity of results. We introduce `oci-agent`, an open-source Python package that implements a human-in-the-loop agentic workflow for observational causal inference. `oci-agent` is designed to automate vital but laborious aspects of applied causal inference, such as covariate balance checking, propensity score trimming, and sensitivity analysis, so that humans can focus on more nuanced tasks, such as framing questions, scrutinizing assumptions, and evaluating diagnostics and results. We initially open-sourced `oci-agent` in June 2026 with support for doubly robust learning of the average treatment effect of a single binary treatment. Since then, we have added support for heterogeneous treatment effect estimation and for multiple continuous treatments via partially linear models. In this paper, we describe the principles behind `oci-agent` and offer internal Netflix case studies and evaluations on public data of its capabilities. Across numerous evaluations, `oci-agent` outperforms less structured baselines while remaining competitive with hand-tuned benchmarks. `oci-agent` is used extensively for causal inference at Netflix and has orchestrated more than 100 analyses per month since its release in June.

stat.CO↗