arXiv subjects

Junming Liu

Publications and source records attributed to Junming Liu.

At least 19 recordsLinked to original sources

arXiv2026

When Should a VLM Look? Paying Only for Visual Calls That Were Needed and Used

Vision-language agents that crop and zoom are trained with rewards that credit a successful tool call, yet a successful call does not show that the model needed to look or used the pixels it received. On our cold-start checkpoint only 10% to 12% of visual calls were both needed and used, and released agents make spurious calls 36% to 87% of the time on individual benchmarks. Outcome rewards, judge rewards, and branch probes each observe one side of this failure, and about two thirds of what an outcome reward pays goes to calls that were neither needed nor used. CounterCredit asks both questions of every image-returning call at its realized pre-call state, using the policy's own gold-answer score. A decision value compares the realized visual branch with answering immediately; an evidence value compares the returned crop with random same-size patches substituted into the same call. A call verified on both earns cashback and every other executed call pays rent; the price is bounded so that every correct trajectory outranks every wrong one, and a dual-channel GRPO advantage keeps the price in its own units. From the same cold start, prompt pool, and budget, CounterCredit reaches 89.5% on V*, 80.2% on HR-Bench-4K, and 76.4% on HR-Bench-8K, 6.3 to 9.4 points above outcome-only GRPO at 1.78 against 1.84 calls per question, and lowers the spurious-call rate to 31% to 36%, the lowest among the agents evaluated. The same recipe lifts a Qwen3-VL-8B base from 75.4 to 80.8 on average.

cs.AI↗

arXiv2026

TransMem: Transforming Hidden States into Memory for Large Language Models

Large language model (LLM) agents increasingly operate over long interaction histories, where effective reasoning requires identifying and exploiting task-relevant evidence distributed across past observations and actions. However, useful information encoded in previously computed representations is often underutilized during subsequent generation. We propose \textbf{TransMem}, a lightweight inference-time parametric memory module that transforms sparse historical hidden states from a frozen LLM backbone into reusable memory representations. TransMem uses a lightweight gating network to dynamically apply the latent intervention to the current hidden states, without repeatedly encoding the preceding context. To learn transferable memory utilization rather than task-specific knowledge, we introduce evidence-conditioned self-distillation. A memory-augmented student processes the full context and matches the predictive distribution of an evidence-only teacher that shares the same frozen backbone. Experiments on LoCoMo, HotpotQA, and MemoryAgentBench demonstrate consistent improvements across different model architectures and scales. TransMem yields gains of 11.58--29.25 $F_1$ on LoCoMo and 10.20--13.03 $F_1$ on HotpotQA, while improving the average MemoryAgentBench accuracy from 29.54\% to 40.00\%. These results establish sparse historical hidden states as an effective and efficient memory substrate for long-context LLM agents. Our code is available at https://github.com/Haodong-Lei-Ray/TransMem.

cs.MA↗

arXiv2026

Mosaic: Data-Free Knowledge Distillation via Mixture-of-Experts for Heterogeneous Distributed Environments

Federated Learning (FL) is a decentralized machine learning paradigm that enables clients to collaboratively train models while preserving data privacy. However, the coexistence of model and data heterogeneity gives rise to inconsistent representations and divergent optimization dynamics across clients, ultimately hindering robust global performance. To transcend these challenges, we propose Mosaic, a novel data-free knowledge distillation framework tailored for heterogeneous distributed environments. Mosaic first trains local generative models to approximate each client's personalized distribution, enabling synthetic data generation that safeguards privacy through strict separation from real data. Subsequently, Mosaic forms a Mixture-of-Experts (MoE) from client models based on their specialized knowledge, and distills it into a global model using the generated data. To further enhance the MoE architecture, Mosaic integrates expert predictions via a lightweight meta model trained on a few representative prototypes. Extensive experiments on standard image and multimodal benchmarks demonstrate that Mosaic consistently outperforms state-of-the-art approaches under both model and data heterogeneity. The source code has been published at https://github.com/Wings-Of-Disaster/Mosaic.

cs.LG↗

arXiv2026

MemVerse: Multimodal Memory for Lifelong Learning Agents

Despite rapid progress in large-scale language and vision models, AI agents still suffer from a fundamental limitation: they cannot remember. Without reliable memory, agents catastrophically forget past experiences, struggle with long-horizon reasoning, and fail to operate coherently in multimodal or interactive environments. We introduce MemVerse, a model-agnostic, plug-and-play memory framework that bridges fast parametric recall with hierarchical retrieval-based memory, enabling scalable and adaptive multimodal intelligence. MemVerse maintains short-term memory for recent context while transforming raw multimodal experiences into structured long-term memories organized as hierarchical knowledge graphs. This design supports continual consolidation, adaptive forgetting, and bounded memory growth. To handle real-time demands, MemVerse introduces a periodic distillation mechanism that compresses essential knowledge from long-term memory into the parametric model, allowing fast, differentiable recall while preserving interpretability. Extensive experiments demonstrate that MemVerse significantly improves multimodal reasoning and continual learning efficiency, empowering agents to remember, adapt, and reason coherently across extended interactions.

cs.AI↗

arXiv2026

MemCoT: Test-Time Scaling through Memory-Driven Chain-of-Thought

Large Language Models (LLMs) still suffer from severe hallucinations and catastrophic forgetting during causal reasoning over massive, fragmented long contexts. Existing memory mechanisms typically treat retrieval as a static, single-step passive matching process, leading to severe semantic dilution and contextual fragmentation. To overcome these fundamental bottlenecks, we propose MemCoT, a test-time memory scaling framework that redefines the reasoning process by transforming long-context reasoning into an iterative, stateful information search. MemCoT introduces a multi-view long-term memory perception module that enables Zoom-In evidence localization and Zoom-Out contextual expansion, allowing the model to first identify where relevant evidence resides and then reconstruct the surrounding causal structure necessary for reasoning. In addition, MemCoT employs a task-conditioned dual short-term memory system composed of semantic state memory and episodic trajectory memory. This short-term memory records historical search decisions and dynamically guides query decomposition and pruning across iterations. Empirical evaluations demonstrate that MemCoT establishes a state-of-the-art performance. Empowered by MemCoT, several open- and closed-source models achieve SOTA performance on the LoCoMo benchmark and LongMemEval-S benchmark.

cs.MA↗

arXiv2026

Self-Evolving Spatial Reasoning in Vision Language Models via Geometric Logic Consistency

Vision-Language Models (VLMs) have made striking progress, yet their spatial reasoning remains fragile: models that answer an original input correctly can still fail under paired transformations with predictable answer mappings, revealing a gap between instance-level correctness and robust spatial reasoning. To address this, we propose Spatial Alignment via Geometric Evolution (SAGE), a self-evolving framework that enforces logical consistency in VLMs through geometric and linguistic duality operations. SAGE incorporates duality consistency as an auxiliary reward within GRPO training, encouraging models to produce logically coherent answers across original and transformed inputs. A dynamic operation pool continuously probes for inconsistencies, promoting challenging operations and retiring mastered ones, so that training focuses on the most informative signals. SAGE is model-agnostic, data-efficient compared to prior GRPO methods, and can be applied as a lightweight post-training stage to any existing VLM. Experiments on video and spatial reasoning benchmarks demonstrate consistent improvements over strong baselines and enhanced generalization to unseen data.

cs.CV↗

arXiv2026

Domain-Adaptive Model Merging Across Disconnected Modes

Learning across domains is challenging when data cannot be centralized due to privacy or heterogeneity, which limits the ability to train a single comprehensive model. Model merging provides an appealing alternative by consolidating knowledge from multiple specialized models into one, avoiding data sharing and reducing retraining cost. In this work, we present DMM, a data-free model merging framework designed to handle highly divergent models. DMM proceeds in three steps. First, domain-specific models are trained independently. Second, models with high similarity are merged using standard techniques to ensure stability. Third, we synthesize pseudo-data from normalization statistics and distill knowledge from divergent models into the merged model through a lightweight refinement guided by these samples. This approach preserves rare but critical knowledge while maintaining stability. Extensive experiments on unimodal and multimodal benchmarks show that DMM achieves state-of-the-art performance over existing merging methods.

cs.DC↗

arXiv2026

MGA: Memory-Driven GUI Agent for Observation-Centric Interaction

Multimodal Large Language Models (MLLMs) have significantly advanced GUI agents, yet long-horizon automation remains constrained by two critical bottlenecks: context overload from raw sequential trajectory dependence and architectural redundancy from over-engineered expert modules. Prevailing End-to-End and Multi-Agent paradigms struggle with error cascades caused by concatenated visual-textual histories and incur high inference latency due to redundant expert components, limiting their practical deployment. To address these issues, we propose the Memory-Driven GUI Agent (MGA), a minimalist framework that decouples long-horizon trajectories into independent decision steps linked by a structured state memory. MGA operates on an ``Observe First and Memory Enhancement`` principle, powered by two tightly coupled core mechanisms: (1) an Observer module that acts as a task-agnostic, intent-free screen state reader to eliminate confirmation bias, visual hallucinations, and perception bias at the root; and (2) a Structured Memory mechanism that distills, validates, and compresses each interaction step into verified state deltas, constructing a lightweight state transition chain to avoid irrelevant historical interference and system redundancy. By replacing raw historical aggregation with compact, fact-based memory transitions, MGA drastically reduces cognitive overhead and system complexity. Extensive experiments on OSWorld and real-world applications demonstrate that MGA achieves highly competitive performance in open-ended GUI tasks while maintaining architectural simplicity, offering a scalable and efficient blueprint for next-generation GUI automation {https://github.com/MintyCo0kie/MGA4OSWorld}.

cs.AI↗

arXiv2026

Hierarchical Memory Orchestration for Personalized Persistent Agents

While long-term memory is essential for intelligent agents to maintain consistent historical awareness, the accumulation of extensive interaction data often leads to performance bottlenecks. Naive storage expansion increases retrieval noise and computational latency, overwhelming the reasoning capacity of models deployed on constrained personal devices. To address this, we propose Hierarchical Memory Orchestration (HMO), a framework that organizes interaction history into a three-tiered directory driven by user-centric contextual relevance. Our system maintains a compact primary cache, coupling recent and pivotal memories with an evolving user profile to ensure agent reasoning remains aligned with individual behavioral traits. This primary cache is complemented by a high-priority secondary layer, both of which are managed within a global archive of the full interaction history. Crucially, the user persona dictates memory redistribution across this hierarchy, promoting records mapped to long-term patterns toward more active tiers while relegating less relevant information. This targeted orchestration surfaces historical knowledge precisely when needed while maintaining a lean and efficient active search space. Evaluations on multiple benchmarks achieve state-of-the-art performance. Real-world deployments in ecosystems like OpenClaw demonstrate that HMO significantly enhances agent fluidity and personalization.

cs.AI↗

arXiv2026

Hit-RAG: Learning to Reason with Long Contexts via Preference Alignment

Despite the promise of Retrieval-Augmented Generation in grounding Multimodal Large Language Models with external knowledge, the transition to extensive contexts often leads to significant attention dilution and reasoning hallucinations. The surge in information density causes critical evidence to be submerged by voluminous noise, which complicates the discernment of relevant fragments within a dense input. In this paper, we propose \textbf{Hit-RAG}, a multi-stage preference alignment framework designed to resolve these cognitive bottlenecks through a progressive optimization pipeline. Our approach systematically refines the utilization of external evidence via three distinct stages. First, Supervised Fine-tuning establishes baseline context awareness to minimize information neglect. Next, Discriminative Preference Alignment enhances robustness against misleading distractors. Finally, Group-Relative Policy Optimization stabilizes logical synthesis to prevent reasoning collapse. Extensive evaluations on eight benchmarks demonstrate that Hit-RAG consistently yields substantial performance gains, enabling models to bridge the gap between context acquisition and accurate reasoning while surpassing much larger counterparts in long-context scenarios.

cs.CL↗

arXiv2026

ReBrain: Brain MRI Reconstruction from Sparse CT Slice via Retrieval-Augmented Diffusion

Magnetic Resonance Imaging (MRI) plays a crucial role in brain disease diagnosis, but it is not always feasible for certain patients due to physical or clinical constraints. Recent studies attempt to synthesize MRI from Computed Tomography (CT) scans; however, low-dose protocols often result in highly sparse CT volumes with poor through-plane resolution, making accurate reconstruction of the full brain MRI volume particularly challenging. To address this, we propose ReBrain, a retrieval-augmented diffusion framework for brain MRI reconstruction. Given any 3D CT scan with limited slices, we first employ a Brownian Bridge Diffusion Model (BBDM) to synthesize MRI slices along the 2D dimension. Simultaneously, we retrieve structurally and pathologically similar CT slices from a comprehensive prior database via a fine-tuned retrieval model. These retrieved slices are used as references, incorporated through a ControlNet branch to guide the generation of intermediate MRI slices and ensure structural continuity. We further account for rare retrieval failures when the database lacks suitable references and apply spherical linear interpolation to provide supplementary guidance. Extensive experiments on SynthRAD2023 and BraTS demonstrate that ReBrain achieves state-of-the-art performance in cross-modal reconstruction under sparse conditions.

cs.CV↗

arXiv2026

Characterizations of harmonic quasiregular mappings in function spaces

We study conjugate-type phenomena for complex-valued harmonic quasiregular mappings in the unit disk across three function space families: $Q(n,p,α)$, $F(p,q,s)$, and the non-derivative $M(p,q,s)$. For a harmonic $K$-quasiregular mapping $f=u+iv$, we first show that if the real part $u$ belongs to $Q_h(1,p,α)$ (with $α>-1$ and $α+1<p<α+2$), the imaginary part $v$ lies in the same space with a $K$-dependent quantitative bound. An analogous stability result is established for the harmonic $F$-scale, with sharp $K$-dependence. These results are extended to harmonic $(K, K')$-quasiregular mappings, yielding explicit estimates with an additional inhomogeneous term involving $K'$. Finally, for normalized harmonic quasiconformal mappings, %$f\in\mathcal S_H(K)$, we derive membership criteria in the harmonic $M$- and $F$-scales, and obtain corresponding conclusions for their natural derivatives, with parameter ranges governed by the order $α_K$ of the family of harmonic quasiconformal mappings.

math.CV↗

arXiv2025

EpiPlanAgent: Agentic Automated Epidemic Response Planning

Epidemic response planning is essential yet traditionally reliant on labor-intensive manual methods. This study aimed to design and evaluate EpiPlanAgent, an agent-based system using large language models (LLMs) to automate the generation and validation of digital emergency response plans. The multi-agent framework integrated task decomposition, knowledge grounding, and simulation modules. Public health professionals tested the system using real-world outbreak scenarios in a controlled evaluation. Results demonstrated that EpiPlanAgent significantly improved the completeness and guideline alignment of plans while drastically reducing development time compared to manual workflows. Expert evaluation confirmed high consistency between AI-generated and human-authored content. User feedback indicated strong perceived utility. In conclusion, EpiPlanAgent provides an effective, scalable solution for intelligent epidemic response planning, demonstrating the potential of agentic AI to transform public health preparedness.

cs.AI↗

arXiv2025

Multiplicative operators on analytic function spaces

H. J. Schwartz proved in his thesis (1969) that a nonzero bounded operator on Hardy spaces $(H^p, 1\leq p\leq\infty)$ is almost multiplicative if and only if it is a composition operator. But, his proof has a gap. In this article, we show that his result is not correct for $H^\infty$ and we fill the gap for $H^p, 1\leq p<\infty.$ Further, we prove that on several classical spaces such as the Bloch space, the little Bloch space, Besov spaces $B_p$ for $p>1$, and weighted Bergman spaces an operator is almost multiplicative if and only if it is a composition operator. Finally, we give a complete characterization of those composition operators that are multiplicative with respect to the Duhamel product of analytic functions.

math.FA↗

arXiv2025

Aligning Vision to Language: Annotation-Free Multimodal Knowledge Graph Construction for Enhanced LLMs Reasoning

Multimodal reasoning in Large Language Models (LLMs) struggles with incomplete knowledge and hallucination artifacts, challenges that textual Knowledge Graphs (KGs) only partially mitigate due to their modality isolation. While Multimodal Knowledge Graphs (MMKGs) promise enhanced cross-modal understanding, their practical construction is impeded by semantic narrowness of manual text annotations and inherent noise in visual-semantic entity linkages. In this paper, we propose Vision-align-to-Language integrated Knowledge Graph (VaLiK), a novel approach for constructing MMKGs that enhances LLMs reasoning through cross-modal information supplementation. Specifically, we cascade pre-trained Vision-Language Models (VLMs) to align image features with text, transforming them into descriptions that encapsulate image-specific information. Furthermore, we developed a cross-modal similarity verification mechanism to quantify semantic consistency, effectively filtering out noise introduced during feature alignment. Even without manually annotated image captions, the refined descriptions alone suffice to construct the MMKG. Compared to conventional MMKGs construction paradigms, our approach achieves substantial storage efficiency gains while maintaining direct entity-to-image linkage capability. Experimental results on multimodal reasoning tasks demonstrate that LLMs augmented with VaLiK outperform previous state-of-the-art models. Our code is published at https://github.com/Wings-Of-Disaster/VaLiK.

cs.CV↗

arXiv2025

DDTime: Dataset Distillation with Spectral Alignment and Information Bottleneck for Time-Series Forecasting

Time-series forecasting is fundamental across many domains, yet training accurate models often requires large-scale datasets and substantial computational resources. Dataset distillation offers a promising alternative by synthesizing compact datasets that preserve the learning behavior of full data. However, extending dataset distillation to time-series forecasting is non-trivial due to two fundamental challenges: 1.temporal bias from strong autocorrelation, which leads to distorted value-term alignment between teacher and student models; and 2.insufficient diversity among synthetic samples, arising from the absence of explicit categorical priors to regularize trajectory variety. In this work, we propose DDTime, a lightweight and plug-in distillation framework built upon first-order condensation decomposition. To tackle Challenge 1, it revisits value-term alignment through temporal statistics and introduces a frequency-domain alignment mechanism to mitigate autocorrelation-induced bias, ensuring spectral consistency and temporal fidelity. To address Challenge 2, we further design an inter-sample regularization inspired by the information bottleneck principle, which enhances diversity and maximizes information density across synthetic trajectories. The combined objective is theoretically compatible with a wide range of condensation paradigms and supports stable first-order optimization. Extensive experiments on 20 benchmark datasets and diverse forecasting architectures demonstrate that DDTime consistently outperforms existing distillation methods, achieving about 30% relative accuracy gains while introducing about 2.49% computational overhead. All code and distilled datasets will be released.

cs.LG↗

arXiv2025

Boosting Adversarial Transferability via Commonality-Oriented Gradient Optimization

Exploring effective and transferable adversarial examples is vital for understanding the characteristics and mechanisms of Vision Transformers (ViTs). However, adversarial examples generated from surrogate models often exhibit weak transferability in black-box settings due to overfitting. Existing methods improve transferability by diversifying perturbation inputs or applying uniform gradient regularization within surrogate models, yet they have not fully leveraged the shared and unique features of surrogate models trained on the same task, leading to suboptimal transfer performance. Therefore, enhancing perturbations of common information shared by surrogate models and suppressing those tied to individual characteristics offers an effective way to improve transferability. Accordingly, we propose a commonality-oriented gradient optimization strategy (COGO) consisting of two components: Commonality Enhancement (CE) and Individuality Suppression (IS). CE perturbs the mid-to-low frequency regions, leveraging the fact that ViTs trained on the same dataset tend to rely more on mid-to-low frequency information for classification. IS employs adaptive thresholds to evaluate the correlation between backpropagated gradients and model individuality, assigning weights to gradients accordingly. Extensive experiments demonstrate that COGO significantly improves the transfer success rates of adversarial attacks, outperforming current state-of-the-art methods.

cs.CV↗

arXiv2025

Cuckoo Attack: Stealthy and Persistent Attacks Against AI-IDE

Modern AI-powered Integrated Development Environments (AI-IDEs) are increasingly defined by an Agent-centric architecture, where an LLM-powered Agent is deeply integrated to autonomously execute complex tasks. This tight integration, however, also introduces a new and critical attack surface. Attackers can exploit these components by injecting malicious instructions into untrusted external sources, effectively hijacking the Agent to perform harmful operations beyond the user's intention or awareness. This emerging threat has quickly attracted research attention, leading to various proposed attack vectors, such as hijacking Model Context Protocol (MCP) Servers to access private data. However, most existing approaches lack stealth and persistence, limiting their practical impact. We propose the Cuckoo Attack, a novel attack that achieves stealthy and persistent command execution by embedding malicious payloads into configuration files. These files, commonly used in AI-IDEs, execute system commands during routine operations, without displaying execution details to the user. Once configured, such files are rarely revisited unless an obvious runtime error occurs, creating a blind spot for attackers to exploit. We formalize our attack paradigm into two stages, including initial infection and persistence. Based on these stages, we analyze the practicality of the attack execution process and identify the relevant exploitation techniques. Furthermore, we analyze the impact of Cuckoo Attack, which can not only invade the developer's local computer but also achieve supply chain attacks through the spread of configuration files. We contribute seven actionable checkpoints for vendors to evaluate their product security. The critical need for these checks is demonstrated by our end-to-end Proof of Concept, which validated the proposed attack across nine mainstream Agent and AI-IDE pairs.

cs.CR↗