Searcharxiv⌕ Search

arXiv subjects

Jitao Sang

Publications and source records attributed to Jitao Sang.

At least 19 recordsLinked to original sources
arXiv2026

CSPO: Alleviating Reward Ambiguity for Structured Table-to-LaTeX Generation

Tables contain rich structured information, yet when stored as images their contents remain "locked" within pixels. Converting table images into LaTeX code enables faithful digitization and reuse, but current multimodal large language models (MLLMs) often fail to preserve structural, style, or content fidelity. Conventional post-training with reinforcement learning (RL) typically relies on a single aggregated reward, leading to reward ambiguity that conflates multiple behavioral aspects and hinders effective optimization. We propose Component-Specific Policy Optimization (CSPO), an RL framework that disentangles optimization across LaTeX tables components-structure, style, and content. In particular, CSPO assigns component-specific rewards and backpropagates each signal only through the tokens relevant to its component, alleviating reward ambiguity and enabling targeted component-wise optimization. To comprehensively assess performance, we introduce a set of hierarchical evaluation metrics. Extensive experiments demonstrate the effectiveness of CSPO, underscoring the importance of component-specific optimization for reliable structured generation. Our code is available at https://github.com/microsoft/CSPO.

cs.AI↗
arXiv2026

Self-Guided Adaptive Safety Alignment: Synthesizing and Internalizing Guidelines in Reasoning Models

Explicit safety policies can improve reasoning-model safety, but their effective coverage may lag behind evolving jailbreak strategies. We study whether a reasoning model can synthesize and internalize a task-specific safety guideline from a small set of harmful and benign examples. We introduce Self-Guided Adaptive Safety Alignment (SGASA). The model generates a guideline, refines it on its own errors, and selects a version by self-evaluation, which can then be applied in context or distilled into the model for guideline-free inference. Across two adversarial prompt datasets and three Qwen3 scales, in-context guidelines improve a combined safety and non-over-refusal score by 20.0-45.5 points on the 8B and 14B models. Self-evaluation selects the externally best refinement round in five of six settings, while guideline generation and utilization show distinct scaling patterns. Using alignment supervision derived only from WildJailbreak, internalized models retain 13.3-14.1 point gains on WildJailbreak without an inference-time guideline. These results support self-generated guidelines as a useful intermediate representation for low-resource safety adaptation.

cs.CL↗
arXiv2026

STAMP: Training Explicit Memory for Mobile GUI Agents in Controllable and Scalable Virtual Environments

Mobile GUI agents excel at immediate reactive control but frequently fail in realistic, long-horizon tasks that require memory. This failure stems from a fundamental conflict between limited context windows and token-heavy screenshots. To save the limited context, agents must progressively discard older visual history, permanently losing crucial transient information. Furthermore, existing action-centric datasets fail to teach agents what or when to explicitly memorize, and augmenting static real-world data is prohibitively expensive and lacks interactive verification. To resolve this, we present STAMP, a framework that trains explicit memory in mobile agents through controllable virtual environments, where deterministic memory variables are programmatically injected into synthesized tasks to control what must be memorized, when it should be encoded, and when it must later be retrieved, thereby producing verifiable supervised data at scale and enabling online reinforcement learning through environment-driven reward feedback. Evaluated on our newly introduced Memory-World benchmark, the resulting Stamp-GUI agent achieves state-of-the-art performance among GUI-specialized models and sets a new high watermark on our Memory-World benchmark, demonstrating exceptional memory accuracy and task resilience while maintaining strong general mobile navigation capabilities.

cs.CL↗
arXiv2026

GUITestScape: Towards Open-set Evaluation on Exploratory GUI Testing

Exploratory GUI testing is a particularly demanding setting for MLLM agents: without predefined test scripts, an agent must autonomously navigate an application and discover defects through its own interaction. However, current evaluation falls short on two fronts. First, existing benchmarks focus almost exclusively on interaction defects, leaving display defects outside the evaluation frame. Second, evaluation protocols are bound to predefined defect annotations, collapsing the testing process into a single end-state judgment that conflates qualitatively distinct failure modes. To address these challenges, we present GUITestScape, an interactive benchmark covering 61 real-world Android applications and 508 preset defects spanning interaction and display types, and introduce GUIJudge, an open-set evaluator that decomposes an agent's testing trajectory into independently diagnosable capabilities. Experimental results demonstrate that GUIJudge achieves reliable process-aware evaluation beyond predefined annotations, substantially outperforming all baselines. Benchmarking on GUITestScape further reveals that detection remains the critical bottleneck for existing models across both defect types, and that integrating GUIJudge's verifiers into existing agents significantly boosts their detection performance without retraining.

cs.SE↗
arXiv2026

Positional Failures in Long-Context LLMs: A Blind Spot in Reasoning Benchmarks

Position-controlled evaluation is standard for retrieval tasks such as Needle-in-a-Haystack and RULER, but mainstream reasoning benchmarks do not control positional placement of target tasks in long contexts. We audit 11 long-context benchmarks and find none jointly controls task position, filler content, and context length for reasoning. An audit of four flagship long-context releases finds no main result-table entry for NIAH, RULER, or LongBench-family benchmarks, while agentic and coding benchmarks appear in main result-tables across all four. We propose Context Rot Evaluation (CRE), a controlled framework varying all three factors, and evaluate nine LLMs on GSM8K and ARC-Challenge across two rounds: an initial five-model set and four newer vendor releases. Models can drop sharply when the target task moves from end to middle, and the drop grows worse with context length for vulnerable models. MiMo-v2-Flash drops 88pp at 64K under with_solutions filler (middle accuracy 8%). Newer releases show smaller drops: at 64K, three of four stay within +/-6pp of end-position accuracy; MiMo-V2.5-Pro narrows the MiMo-v2-Flash 88pp drop to 32pp. Under questions_only_v2 filler, middle-position drops persist across all four (range -16pp to -56pp across 8K, 32K, 64K). At 8K, a diagnostic probe adding a target-task copy at the end brings middle accuracy within +/-4pp of end baseline across all nine models, consistent with a positional explanation. In the initial five-model set, 76% of middle-position errors match surrounding filler text versus 22% at the end position, consistent with filler-answer interference as a dominant error mode. These results expose a structural evaluation gap in current reasoning benchmark design and vendor evaluation practice: positional vulnerabilities that grow with context length cannot be measured when task position is not controlled.

cs.CL↗
arXiv2026

Taming the Thinker: Conditional Entropy Shaping for Adaptive LLM Reasoning

Entropy-based deep reasoning has emerged as a promising direction for improving the reasoning capabilities of Large Language Models (LLMs), but existing methods often either increase response length indiscriminately or shorten responses at the cost of accuracy. To better balance this trade-off, we introduce Conditional Entropy Shaping (CES), a framework that dynamically controls token-level response entropy, enabling LLMs to produce concise solutions on simple problems while encouraging deeper exploration on hard ones. Built on DAPO, CES uses token-level entropy as an uncertainty signal and applies a conditional bidirectional policy: it penalizes high-entropy "forking point" tokens on correct reasoning paths to improve conciseness, and rewards them on incorrect paths to encourage exploration and error correction. We implement CES on DeepSeek-R1-Distill-7B and evaluate it on 12 mathematical benchmarks. CES consistently improves average accuracy while reducing response length relative to DAPO, and supplementary experiments show similar trends on a smaller 1.5B backbone and on out-of-domain benchmarks.

cs.CL↗
arXiv2026

Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Tasks

Long-context Large Language Models, despite their expanded capacity, require careful working memory management to mitigate attention dilution during long-horizon tasks. Yet existing approaches rely on external mechanisms that lack awareness of the agent's reasoning state, leading to suboptimal decisions. We propose Memory-as-Action (MemAct), a framework that treats working memory management as learnable policy actions. By formulating context management as in-place editing operations (deletion, insertion), MemAct enables joint optimization of information retention and task performance through end-to-end reinforcement learning. To address the computational challenges of dynamic context updates, we introduce Dynamic Context Policy Optimization, which restores training efficiency without compromising reasoning integrity. Experiments show that MemAct-RL-14B matches the accuracy of models $16\times$ larger while reducing average context length by 51\%, with learned strategies that adapt to model capabilities and generalize across task complexities.

cs.AI↗
arXiv2026

TiMem: Temporal-Hierarchical Memory Consolidation for Long-Horizon Conversational Agents

Long-horizon conversational agents have to manage ever-growing interaction histories that quickly exceed the finite context windows of large language models (LLMs). Existing memory frameworks provide limited support for temporally structured information across hierarchical levels, often leading to fragmented memories and unstable long-horizon personalization. We present TiMem, a temporal--hierarchical memory framework that organizes conversations through a Temporal Memory Tree (TMT), enabling systematic memory consolidation from raw conversational observations to progressively abstracted persona representations. TiMem is characterized by three core properties: (1) temporal--hierarchical organization through TMT; (2) semantic-guided consolidation that enables memory integration across hierarchical levels without fine-tuning; and (3) complexity-aware memory recall that balances precision and efficiency across queries of varying complexity. Under a consistent evaluation setup, TiMem achieves state-of-the-art accuracy on both benchmarks, reaching 75.30% on LoCoMo and 76.88% on LongMemEval-S. It outperforms all evaluated baselines while reducing the recalled memory length by 52.20% on LoCoMo. Manifold analysis indicates clear persona separation on LoCoMo and reduced dispersion on LongMemEval-S. Overall, TiMem treats temporal continuity as a first-class organizing principle for long-horizon memory in conversational agents. The code is available at https://github.com/TiMEM-AI/timem.

cs.CL↗
arXiv2026

VAPO: End-to-end Slide-Enhanced Speech Recognition with Omni-modal Large Language Models

Omni-modal large language models (OLLMs) offer a promising end-to-end solution for slide-enhanced speech recognition due to their inherent multimodal capabilities. However, we found a fundamental issue faced by OLLMs: \textit{Visual Interference}, where models show a bias towards visible text over auditory signals, causing them to hallucinate slide content that was never spoken. To address this, we propose Visually-Anchored Policy Optimization (VAPO), which aims to reshape models' inference process to follow the human-like ``Look-then-Listen'' inference chain. Specifically, we design a temporally decoupled policy: the model first extracts visual priors in a block to serve as semantic anchors, then generates the transcription in an block. The policy is optimized via multi-objective reinforcement learning. Furthermore, we introduce SlideASR-Bench, a comprehensive benchmark designed to address the scarcity of entity-rich data, comprising a large-scale synthetic corpus for training and a challenging real-world test set for evaluation. We conduct extensive evaluations demonstrating that VAPO effectively eliminates visual interference and achieves state-of-the-art performance on SlideASR-Bench and public datasets, significantly reducing entity recognition errors in specialized domains.

eess.AS↗
arXiv2026

Evaluate-as-Action: Self-Evaluated Process Rewards for Retrieval-Augmented Agents

Retrieval-augmented agents can query external evidence, yet their reliability in multi-step reasoning remains limited: noisy retrieval may derail multi-hop question answering, while outcome-only reinforcement learning provides credit signals that are too coarse to optimize intermediate steps. We propose \textsc{EvalAct} (Evaluate-as-Action), which converts implicit retrieval quality assessment into an explicit action and enforces a coupled Search-to-Evaluate protocol so that each retrieval is immediately followed by a structured evaluation score, yielding process signals aligned with the interaction trajectory. To leverage these signals, we introduce Process-Calibrated Advantage Rescaling (PCAR), a GRPO-based optimization method that rescales advantages at the segment level according to evaluation scores, emphasizing reliable segments while updating uncertain ones conservatively. Experiments on seven open-domain QA benchmarks show that \textsc{EvalAct} achieves the best average accuracy, with the largest gains on multi-hop tasks, and ablations verify that the explicit evaluation loop drives the primary improvements while PCAR provides consistent additional benefits.

cs.AI↗
arXiv2026

HulluEdit: Single-Pass Evidence-Consistent Subspace Editing for Mitigating Hallucinations in Large Vision-Language Models

Object hallucination in Large Vision-Language Models (LVLMs) significantly hinders their reliable deployment. Existing methods struggle to balance efficiency and accuracy: they often require expensive reference models and multiple forward passes, or apply static edits that risk suppressing genuine visual evidence. To address this, we introduce HulluEdit, a single-pass, reference-free intervention framework. Our core innovation is orthogonal subspace editing: we decompose the hidden states of the model into orthogonal subspaces - visual evidence, conflicting priors, and residual uncertainty - enabling selective suppression of hallucinatory patterns without interfering with visual grounding. This approach mathematically guarantees that edits applied to the prior subspace leave the visual component entirely unaffected. Extensive experiments show that HulluEdit achieves state-of-the-art hallucination reduction on benchmarks including POPE and CHAIR across diverse architectures, while preserving general capabilities on MME and maintaining efficient inference. Our method consistently outperforms contrastive decoding and static subspace editing baselines, offering a new pathway toward more trustworthy LVLMs.

cs.CV↗
arXiv2026

NAP-Tuning: Neural Augmented Prompt Tuning for Adversarially Robust Vision-Language Models

Vision-Language Models (VLMs) such as CLIP have demonstrated remarkable capabilities in understanding relationships between visual and textual data through joint embedding spaces. Despite their effectiveness, these models remain vulnerable to adversarial attacks, particularly in the image modality, posing significant security concerns. Building upon our previous work on Adversarial Prompt Tuning (AdvPT), which introduced learnable text prompts to enhance adversarial robustness in VLMs without extensive parameter training, we present a significant extension by introducing the Neural Augmentor framework for Multi-modal Adversarial Prompt Tuning (NAP-Tuning).Our key innovations include: (1) extending AdvPT from text-only to multi-modal prompting across both text and visual modalities, (2) expanding from single-layer to multi-layer prompt architectures, and (3) proposing a novel architecture-level redesign through our Neural Augmentor approach, which implements feature purification to directly address the distortions introduced by adversarial attacks in feature space. Our NAP-Tuning approach incorporates token refiners that learn to reconstruct purified features through residual connections, allowing for modality-specific and layer-specific feature correction.Comprehensive experiments demonstrate that NAP-Tuning significantly outperforms existing methods across various datasets and attack types. Notably, our approach shows significant improvements over the strongest baselines under the challenging AutoAttack benchmark, outperforming them by 33.5% on ViT-B16 and 33.0% on ViT-B32 architectures while maintaining competitive clean accuracy.

cs.CV↗
arXiv2026

MoHoBench: Assessing Honesty of Multimodal Large Language Models via Unanswerable Visual Questions

Recently Multimodal Large Language Models (MLLMs) have achieved considerable advancements in vision-language tasks, yet produce potentially harmful or untrustworthy content. Despite substantial work investigating the trustworthiness of language models, MMLMs' capability to act honestly, especially when faced with visually unanswerable questions, remains largely underexplored. This work presents the first systematic assessment of honesty behaviors across various MLLMs. We ground honesty in models' response behaviors to unanswerable visual questions, define four representative types of such questions, and construct MoHoBench, a large-scale MMLM honest benchmark, consisting of 12k+ visual question samples, whose quality is guaranteed by multi-stage filtering and human verification. Using MoHoBench, we benchmarked the honesty of 28 popular MMLMs and conducted a comprehensive analysis. Our findings show that: (1) most models fail to appropriately refuse to answer when necessary, and (2) MMLMs' honesty is not solely a language modeling issue, but is deeply influenced by visual information, necessitating the development of dedicated methods for multimodal honesty alignment. Therefore, we implemented initial alignment methods using supervised and preference learning to improve honesty behavior, providing a foundation for future work on trustworthy MLLMs. Our data and code can be found at https://github.com/yanxuzhu/MoHoBench.

cs.AI↗
arXiv2026

GUITester: Enabling GUI Agents for Exploratory Defect Discovery

Exploratory GUI testing is essential for software quality but suffers from high manual costs. While Multi-modal Large Language Model (MLLM) agents excel in navigation, they fail to autonomously discover defects due to two core challenges: \textit{Goal-Oriented Masking}, where agents prioritize task completion over reporting anomalies, and \textit{Execution-Bias Attribution}, where system defects are misidentified as agent errors. To address these, we first introduce \textbf{GUITestBench}, the first interactive benchmark for this task, featuring 143 tasks across 26 defects. We then propose \textbf{GUITester}, a multi-agent framework that decouples navigation from verification via two modules: (i) a \textit{Planning-Execution Module (PEM)} that proactively probes for defects via embedded testing intents, and (ii) a \textit{Hierarchical Reflection Module (HRM)} that resolves attribution ambiguity through interaction history analysis. GUITester achieves an F1-score of 48.90\% (Pass@3) on GUITestBench, outperforming state-of-the-art baselines (33.35\%). Our work demonstrates the feasibility of autonomous exploratory testing and provides a robust foundation for future GUI quality assurance~\footnote{Our code is now available in~\href{https://github.com/ADaM-BJTU/GUITestBench}{https://github.com/ADaM-BJTU/GUITestBench}}.

cs.AI↗
arXiv2025

ITDR: An Instruction Tuning Dataset for Enhancing Large Language Models in Recommendations

Large language models (LLMs) have demonstrated outstanding performance in natural language processing tasks. However, in the field of recommender systems, due to the inherent structural discrepancy between user behavior data and natural language, LLMs struggle to effectively model the associations between user preferences and items. Although prompt-based methods can generate recommendation results, their inadequate understanding of recommendation tasks leads to constrained performance. To address this gap, we construct a comprehensive instruction tuning dataset, ITDR, which encompasses seven subtasks across two root tasks: user-item interaction and user-item understanding. The dataset integrates data from 13 public recommendation datasets and is built using manually crafted standardized templates, comprising approximately 200,000 instances. Experimental results demonstrate that ITDR significantly enhances the performance of mainstream open-source LLMs such as GLM-4, Qwen2.5, Qwen2.5-Instruct and LLaMA-3.2 on recommendation tasks. Furthermore, we analyze the correlations between tasks and explore the impact of task descriptions and data scale on instruction tuning effectiveness. Finally, we perform comparative experiments against closed-source LLMs with massive parameters. Our tuning dataset ITDR, the fine-tuned large recommendation models, all LoRA modules, and the complete experimental results are available at https://github.com/hellolzk/ITDR.

cs.IR↗
arXiv2025

AI Deception: Risks, Dynamics, and Controls

As intelligence increases, so does its shadow. AI deception, in which systems induce false beliefs to secure self-beneficial outcomes, has evolved from a speculative concern to an empirically demonstrated risk across language models, AI agents, and emerging frontier systems. This project provides a comprehensive and up-to-date overview of the AI deception field, covering its core concepts, methodologies, genesis, and potential mitigations. First, we identify a formal definition of AI deception, grounded in signaling theory from studies of animal deception. We then review existing empirical studies and associated risks, highlighting deception as a sociotechnical safety challenge. We organize the landscape of AI deception research as a deception cycle, consisting of two key components: deception emergence and deception treatment. Deception emergence reveals the mechanisms underlying AI deception: systems with sufficient capability and incentive potential inevitably engage in deceptive behaviors when triggered by external conditions. Deception treatment, in turn, focuses on detecting and addressing such behaviors. On deception emergence, we analyze incentive foundations across three hierarchical levels and identify three essential capability preconditions required for deception. We further examine contextual triggers, including supervision gaps, distributional shifts, and environmental pressures. On deception treatment, we conclude detection methods covering benchmarks and evaluation protocols in static and interactive settings. Building on the three core factors of deception emergence, we outline potential mitigation strategies and propose auditing approaches that integrate technical, community, and governance efforts to address sociotechnical challenges and future AI risks. To support ongoing work in this area, we release a living resource at www.deceptionsurvey.com.

cs.AI↗
arXiv2025

Reasoning Shapes Alignment: Investigating Cultural Alignment in Large Reasoning Models with Cultural Norms

The advanced reasoning capabilities of Large Reasoning Models enable them to thoroughly understand and apply safety policies through deliberate thought processes, thereby improving the models' safety. Beyond safety, these models must also be able to reflect the diverse range of human values across various cultures. This paper presents the Cultural Norm-based Cultural Alignment (CNCA) framework, which enables models to leverage their powerful reasoning ability to align with cultural norms. Specifically, we propose three methods to automatically mine cultural norms from limited survey data and explore ways to effectively utilize these norms for improving cultural alignment. Two alignment paradigms are examined: an in-context alignment method, where cultural norms are explicitly integrated into the user context, and a fine-tuning-based method, which internalizes norms through enhanced Chain-of-Thought training data. Comprehensive experiments demonstrate the effectiveness of these methods, highlighting that models with stronger reasoning capabilities benefit more from cultural norm mining and utilization. Our findings emphasize the potential for reasoning models to better reflect diverse human values through culturally informed alignment strategies.

cs.AI↗
arXiv2025

Beyond Pipelines: A Survey of the Paradigm Shift toward Model-Native Agentic AI

The rapid evolution of agentic AI marks a new phase in artificial intelligence, where Large Language Models (LLMs) no longer merely respond but act, reason, and adapt. This survey traces the paradigm shift in building agentic AI: from Pipeline-based systems, where planning, tool use, and memory are orchestrated by external logic, to the emerging Model-native paradigm, where these capabilities are internalized within the model's parameters. We first position Reinforcement Learning (RL) as the algorithmic engine enabling this paradigm shift. By reframing learning from imitating static data to outcome-driven exploration, RL underpins a unified solution of LLM + RL + Task across language, vision and embodied domains. Building on this, the survey systematically reviews how each capability -- Planning, Tool use, and Memory -- has evolved from externally scripted modules to end-to-end learned behaviors. Furthermore, it examines how this paradigm shift has reshaped major agent applications, specifically the Deep Research agent emphasizing long-horizon reasoning and the GUI agent emphasizing embodied interaction. We conclude by discussing the continued internalization of agentic capabilities like Multi-agent collaboration and Reflection, alongside the evolving roles of the system and model layers in future agentic AI. Together, these developments outline a coherent trajectory toward model-native agentic AI as an integrated learning and interaction framework, marking the transition from constructing systems that apply intelligence to developing models that grow intelligence through experience.

cs.AI↗