arXiv subjects

Jun Huang

Publications and source records attributed to Jun Huang.

At least 19 recordsLinked to original sources

arXiv2026

DART: Distillation-Aware Reparameterization for Training-Free LoRA Reuse in Few-Step Video Diffusion Models

Step distillation reduces the cost of video generation, but reusing a LoRA trained for a longer trajectory can alter its functional effect or degrade target quality. Static parameter compatibility offers one perspective on this problem; our observations show that similar measured geometry can coexist with different adapter behavior under a shortened denoising schedule. We propose DART, a training-free method that combines low-rank coordinate transport with target-schedule response calibration using forward evaluations and no source training videos. On a four-step Wan2.2 target, DART-F improves the joint quality score from 0.9029 to 0.9227 and changes macro functional retention from -0.4644 to +0.1349. Component analysis shows that calibration accounts for most of the quality improvement, while coordinate transport provides complementary gains when combined with calibration. Adapter-level results reveal positive functional effects for some adapters and strong attenuation with reduced negative functional effects for others. Evaluations on two additional targets show the same aggregate trend. These results motivate evaluating distilled-model LoRA reuse jointly through functional preservation and negative-transfer avoidance, without assuming recovery for every adapter.

cs.AI↗

arXiv2026

Not All Prompts Are Equal: Exploration-Guided Prompt Scaffolding for Multimodal Reinforcement Post-Training

Training prompts in online reinforcement learning (RL) differ substantially in how informative they are for the current policy: some are already saturated while others are too difficult to yield reliable learning signals, yet both receive equal rollout budget under standard training. We propose an exploration-guided prompt scaffolding framework that adapts the training prompt distribution dynamically throughout RL post-training of multimodal large language models (MLLMs). Central to our approach is the $\textit{Exploration Potential Score} (EPS)$, a lightweight rollout-based proxy for prompt utility derived from KL-regularized policy improvement theory, computable directly from on-policy rollout statistics without additional overhead. Rather than discarding low-utility prompts, we use a teacher model to generate scaffolded rewrites that preserve the original task intent while making subsequent training more informative, reframing teacher supervision as training-data refinement rather than output imitation. Integrated with GRPO on Geo3K and MMK12, our method consistently outperforms the baseline on both in-domain and out-of-distribution benchmarks, achieving up to 9.7\% relative improvement in-domain and gains of 11.5\% on MathVision and 11.1\% on MMMU-Pro.

cs.LG↗

arXiv2026

BR-FiLM: Bounded Residual Channel-Quality Conditioning for Automatic Modulation Recognition

Automatic Modulation Recognition (AMR) plays a crucial role in enabling robust, adaptive, and secure communication for military and civilian applications. Deep learning has enabled effective AMR methods that overcome the computational inefficiency of traditional approaches. However, these deep learning based methods often degrade significantly in low SNR conditions, where noise obscures modulation-discriminative waveform features. In this paper, we propose Bounded Residual Feature-wise Linear Modulation (BR-FiLM), a channel-quality conditioning block for AMR, which can be inserted into AMR classifiers with intermediate feature representations. We construct BR-FiLMNet by inserting the proposed BR-FiLM block into a Multi-Channel Convolutional Long Short-Term Deep Neural Network (MCLDNN) backbone. BR-FiLMNet conditions convolutional, recurrent, and dense features through gated residual corrections while preserving the original I/Q-driven feature path. Experimental results on RadioML 2016.10a show that BR-FiLMNet improves mean accuracy from 61.79% to 67.74% and low-SNR accuracy (SNR <= 0 dB) from 37.12% to 46.44% compared to MCLDNN. We further evaluate BR-FiLMNet against recent transformer-style baselines. The results indicate that BR-FiLMNet delivers significant and reliable performance gains, as validated through paired statistical testing.

eess.SP↗

arXiv2026

Bringing Environmental Enhancement Back to Its Physical Essence via Specular Reflecting Surfaces

Intelligent control of wireless propagation environments is crucial for future network capacity and reliability. Unlike circuit-controlled reconfigurable intelligent surfaces (RIS), mechanically actuated specular reflecting surfaces (SRS) offer a simpler and potentially more cost-effective alternative. In this paper, based on the tractable ray-based cascaded channel model with power-projection correction, we investigate the fundamental operational behaviors of an ideal SRS in free space. Specifically, in the angle-aligned near field, edge reflections cause non-constructive combining, resulting in a damped oscillatory convergence of the gain to an aperture-independent constant. We further obtain the far-field behavior, unbounded-aperture asymptotics, an optimal aperture size and reflection angle, and a gain-based near/far-field boundary. For misalignment, we provide accurate approximations for small and large apertures via center-point and stationary-point analyses. We also define the SRS beam pattern, derive analytical 3-dB beamwidths, and quantify the effective region where a main lobe exists. Finally, we derive a closed-form achievable-rate for an SRS-aided communication system. Numerical results validate the proposed expressions, reveal distinct near-/far-field behaviors of specular reflection, and show that SRS can outperform RIS in the far field due to continuous aperture and angular-resolution control and stronger power projection.

cs.IT↗

arXiv2026

Bootstrapping Vision-Language Model for Hysteroscopic Surgical Scene Segmentation

Hysteroscopic surgical scene segmentation plays a pivotal role in understanding the hysteroscopic intraoperative environment as well as computer-assisted intervention. However, this task presents unique challenges due to the high morphological similarity among different lesions and the presence of artifacts such as specular reflections, motion blur, and fluid occlusions in surgical videos. In this work, we propose the first vision-language model (VLM)-based hysteroscopic surgical scene segmentation method, which performs pixel-wise localization for fifteen representative categories in hysteroscopic surgical scenes. Our VLM-hyster has a segmentation backbone that utilizes the pretrained image encoder for robust visual feature extraction, coupled with a transformer-based decoder for dense prediction. Moreover, we design category-specific text prompts and incorporate a masked distillation branch to filter out visual features with low correlation to the text prompts, enabling the model to focus more effectively on category-specific image regions and thereby enhancing segmentation performance. We collect a large multicentric hysteroscopic surgical scene dataset, containing 4,020 high-resolution images with detailed mask annotations, for model training and evaluation. Experimental results demonstrate that VLM-hyster substantially outperforms state-of-the-art AI models. Furthermore, extensive assessments by gynecologists, as well as multicentre and prospective validations, demonstrate VLM-hyster's robustness and generalizability. The results suggest that VLM-hyster earns considerable potential in enabling AI-assisted localization of surgical instruments and lesions in hysteroscopic surgeries. Code is available at https://github.com/viscom-tongji/VLM-hyster.

cs.CV↗

arXiv2026

FedSceneX: Time-to-Target Orchestration for Same-Scene Multimodal Federated Edge Learning

Federated learning at the sensing edge is typically evaluated by communication rounds, yet a round does not represent a fixed amount of work. Even on identical hardware, the methods we compare require 3.3 to 9.8 hours per round, which makes round-based comparisons misleading. The problem is more obvious for same-scene multimodal clients, since camera, video, LiDAR, and radar workloads differ substantially in training and communication cost, while existing methods treat the modality composition of each round as fixed. To address it, we introduce FedSceneX, an orchestrator that jointly determines round composition to maximize learning value per active hour. The optimization method, Value-per-Hour Pricing (VHP), converts the fractional objective through a parametric transformation and dualizes the uplink constraint, yielding a closed-form client price whose weights capture resource shadow costs. Based on these prices, FedSceneX selects clients subject to a modality coverage constraint, allocates precision through reverse water filling, and assigns updates to edge servers. On the full nuScenes benchmark with fifteen clients and twelve baselines, FedSceneX reduces the active time per round to 3.31 hours, compared with 4.85 to 9.78 hours for the baselines. Across all random seeds, it achieves the highest accuracy within a twenty-hour budget while preserving all four modalities. Its advantage persists from ten to forty-five hours, after which conventional methods overtake it.

cs.NI↗

arXiv2026

CrystalMem: Elastic Memory for Self-Evolving LLM Agents via Knowledge Crystallization

Memory for self-evolving large language model (LLM) agents is often provisioned as if its byte budget only grows. Cloud platforms, however, adjust quotas with load and cost, and we show that capability does not follow the budget back up: after a squeeze-and-recover cycle, the agent settles below its pre-squeeze level, a gap we call memory hysteresis. The cause is structural. Deletion and one-way compression discard the material needed for later rebuilding, and we prove that any policy that only keeps or drops entries carries a residual-deficit floor. We propose CrystalMem (Crystallized Memory), an elastic memory sidecar that demotes entries across four fidelity states under a crystallization-energy schedule, orders demotions by advantage-weighted influence with dependency coupling, and recovers capability through verified recrystallization under explicit compute and byte caps. Across seven environments, seventeen methods, and six backbones, with multi-tenant serving and a physical edge-cloud deployment, CrystalMem achieves the highest restored capability in every setting and closes the loop left open by every baseline. From a 50% byte budget, CrystalMem matches the strongest budgeted baseline at full provision on every environment; at equal budgets, it leads by +4.6 pp on average.

cs.AI↗

arXiv2026

When Unlearning Fails: Reliable Data Deletion under Post-Training in Agent Networks

Self-improving federated agent networks keep training after deployment by collecting new trajectories with the current policy and feeding them back into later rounds. This closed loop makes unlearning harder than a one-time model repair. When a data owner requests deletion, the target data may have already shaped later retained trajectories, so retraining or model-side unlearning can leave an influence echo that returns as the network continues to operate. We show that this echo survives retained-data retraining, grows with the amount of forget-shaped retained data, and can be traced from deployment, collection, and aggregation records. To address this problem, we propose MUTE, a Muting Unlearned Trajectories' Echoes method for reliable deletion in self-improving federated agent networks. MUTE estimates downstream influence from a lightweight server ledger, removes the current residue through a forget-retain update, contains high-influence retained trajectories through quarantine or down-weighting, and audits later behavior to schedule additional erasure under an uplink budget. Experiments on LIBERO with two vision-language-action backbones, three deletion granularities, and a physical Jetson-based edge testbed show that MUTE keeps behavioral leakage and influence regeneration low while preserving task utility and using much less communication than full retraining.

cs.NI↗

arXiv2026

Energy Market and Carbon Emission Spillovers in Critical Minerals Investment: A Dynamic Connectedness Approach

Design/methodology/approach A time-varying parameter vector autoregression (TVP-VAR) model is employed to quantify dynamic connectedness and directional volatility spillovers using daily data from May 1, 2013, to May 2, 2023. The study isolates the impact of extreme events by splitting the data into pre- and post-COVID-19 samples based on the February 2020 stock market crash. Purpose This paper examines the daily financial risk spillovers associated with investing in critical minerals. It examines the dynamic interconnectedness between seven critical mineral Exchange-Traded Fund (ETF) portfolios and key economic-wide variables, including the energy market, carbon emissions, market sentiment, and global infrastructure. Findings Portfolios with high Environmental, Social, and Governance (ESG) scores significantly contribute to shock spillovers. Net directional connectedness analysis reveals that West Texas Intermediate (WTI) crude oil and carbon emission futures consistently act as "net receivers," absorbing volatility from the system. Conversely, Cobalt and Aluminum ETFs primarily act as "net givers," transmitting volatility. The pandemic caused significant structural shifts in these transmission roles. Practical implications The identification of specific net givers and receivers provides actionable insights for investors, facilitating better hedging strategies against time-varying structural breaks and broader economic shocks. Originality This study uniquely utilizes financial ETF data rather than physical mineral prices to capture accessible investment risks. It is among the first to link ESG scores to the directional role (giver vs. receiver) of critical mineral assets within a broader macro-financial network.

econ.EM↗

arXiv2026

Student-Centered Distillation Narrows the Agentic Gap Between Small and Large LLMs

Large Language Model agents achieve strong performance on multi-step reasoning and tool-use tasks, but their impressive capabilities typically rely on extremely large backbones. Existing distillation approaches train smaller students to imitate full teacher trajectories, yet reasoning and knowledge gaps between the teacher and student can cause compounding errors. We propose SCoRe, a student-centered framework in which the student generates training trajectories and the teacher corrects only the earliest error, producing training data matched to the student's abilities and exposing specific weaknesses. The student is first fine-tuned on corrected trajectories. Subsequently, short-horizon reinforcement learning starts from the verified prefix preceding the earliest error, with target rewards assigned at that step. This design enables the student to solve problems through unconstrained RL exploration rather than teacher imitation, while the short-horizon setup improves training stability. On 12 challenging benchmarks, a 7B-parameter student distilled with SCoRe closes the agentic performance gap with a 72B-parameter teacher.

cs.CL↗

arXiv2026

Infinity-Parser2 Technical Report

We present Infinity-Parser2, a large multimodal model that couples a controllable data-synthesis pipeline with multi-task reinforcement learning for end-to-end document parsing, addressing the persistent scarcity of faithfully annotated parsing corpora. Our contributions are threefold. First, we build a scalable synthesis engine, pairing a controllable rendering framework with an iterative refinement loop, and use it to construct and open-source Infinity-Doc2-5M: a 5-million-sample bilingual (Chinese/English) corpus spanning diverse document types, annotated with element bounding boxes, canonical content forms (Markdown, HTML, LaTeX, SMILES, structured charts), and full-page reading order. Second, we introduce a verifiable, multi-task reward system that enables Joint Reinforcement Learning across eight co-trained objectives (document parsing, layout analysis, table parsing, math formula parsing, chart parsing, chemical formula parsing, document VQA, and general multimodal understanding), unifying perception, structure, and reasoning in a single optimization signal. Third, we release two variants under a shared architecture: Infinity-Parser2-Flash, optimized for low-latency inference with a 3.68x throughput gain over Infinity-Parser-7B, and Infinity-Parser2-Pro, engineered for precision-critical settings. Infinity-Parser2-Pro reaches state-of-the-art 87.6% on olmOCR-Bench and 74.3% on ParseBench, surpassing DeepSeek-OCR-2, PaddleOCR-VL-1.5, and MinerU2.5, with strong generalization to charts, chemical formulas, and document VQA.

cs.AI↗

arXiv2026

FlowWAM: Optical Flow as a Unified Action Representation for World Action Models

World Action Models (WAMs) are able to leverage pretrained video generators for both world modeling and action prediction. However, directly leveraging such video generators for control raises a new challenge: how to represent actions in a suitable form that aligns with pretrained video generators while carrying enough motion cues for accurate control. Existing numerical actions fail to satisfy the former, and prior visual action representations overlook the temporal motion structure across frames. We address this issue with FlowWAM, a dual-stream diffusion framework that adopts optical flow as a unified, video-native action representation. Flow videos share the same format as RGB videos and encode rich per-pixel displacement. By jointly modeling them within a shared pretrained video generator, FlowWAM can naturally implement two modes of WAMs. In policy mode, FlowWAM generates flow for action prediction, while in world-model mode, it uses target flow sequences to guide future video generation. Moreover, since flow can be easily extracted from raw videos without action labels, FlowWAM can leverage large-scale action-unlabeled video datasets for pretraining. We empirically find that our flow-based action representation delivers gains across both modes. On RoboTwin manipulation, FlowWAM raises the success rate to 92.94% on the Clean setting and 92.14% on Random, outperforming both VLA and WAM baselines. On WorldArena world modeling, it achieves the best overall EWMScore (63.71) with an 18.4% relative improvement in trajectory accuracy. More results can be found on our project website: https://flow-wam.github.io .

cs.RO↗

arXiv2026

SGMatch: Semantic-Guided Non-Rigid Shape Matching with Flow Regularization

Establishing accurate point-to-point correspondences between non-rigid 3D shapes remains a critical challenge, particularly under non-isometric deformations and topological noise. Existing functional map pipelines suffer from ambiguities that geometric descriptors alone cannot resolve, and spatial inconsistencies inherent in the projection of truncated spectral bases to dense pointwise correspondences. In this paper, we introduce SGMatch, a learning-based framework that couples 3D-lifted semantic cues with trajectory-level feature transport regularization. Specifically, we design a Semantic-Guided Local Cross-Attention module that integrates semantic features from vision foundation models into geometric descriptors while preserving local structural continuity. Furthermore, we adapt conditional flow matching as a time-conditioned feature transport regularizer that promotes spatially coherent point-wise recovery. Experimental results on multiple benchmarks demonstrate that SGMatch achieves competitive performance across near-isometric settings and consistent improvements under non-isometric deformations and topological noise.

cs.CV↗

arXiv2026

Divide and Conquer: Decoupled Representation Alignment for Multimodal World Models

Emerging multi-modal world models attempt to jointly generate videos across diverse modalities (e.g., RGB, depth, and mask), yet they fail to fully exploit the rich priors of existing foundation models. We propose $M^2$-REPA, the first representation alignment method tailored for multi-modal video generation. Our key insight is that foundation models trained on different modality spaces naturally capture distinct domain-specific priors, acting as complementary "experts." Specifically, we first decouple modality-specific features from the diffusion model's intermediate representations, then align each with its corresponding expert foundation model. To this end, we design two synergistic objectives: a multi-modal representation alignment loss that enforces feature-to-expert matching, and a modality-specific decoupling regularization that encourages complementarity across different modalities. This design enables joint optimization, fully exploiting priors from multiple foundation models. Extensive experiments demonstrate that our method significantly outperforms baselines in visual quality and long-term consistency.

cs.CV↗

arXiv2026

Forget to Improve: On-Device LLM-Agent Continual Learning via Budget-Curated Memory

On-device language-model agents improve by accumulating experience in retrieved memory rather than by updating weights. This memory is hard-bounded and exposed: it consumes RAM and energy, reaches peers through a thin uplink, and becomes an attack surface because it is writable by what the agent reads. Existing systems each cover one part of this problem: agentic memories grow without a budget, on-device methods keep entries by success alone, and poisoning is studied mainly as an attack rather than as a memory-governance problem. We propose \sys{}, a single net-value-per-byte score that governs an agent's experience-memory lifecycle. The main idea is to let the budget act as the curator: each entry is scored as value minus harm, per byte, so one ruler decides what to keep, share, and trust. \sys{} makes three decisions: (1) \textbf{KEEP} evicts low-value bytes under the RAM and energy budget; (2) \textbf{SHARE} sends an insight only when its value exceeds its uplink cost; and (3) \textbf{TRUST} gates a peer entry by provenance. On language-model-agent task-drift benchmarks and a real heterogeneous Jetson testbed with two robot-arm nodes and a hub, \sys{} reduces memory by $2.7\times$ and uplink by $2.4\times$, drives injection success from 0.75 to zero, and raises accuracy on cases corrupted by poison or stale memory. Curating by net value reduces footprint, energy, uplink, and injection success together without reducing accuracy. In this setting, forgetting by net value improves the agent rather than weakening it.

cs.LG↗

arXiv2026

Bridging What the Model Thinks and How It Speaks: Expressive Speech Generation via Self-Aware Intent-Realization Alignment

Speech Language Models (SLMs) exhibit strong semantic understanding, yet often fail to translate this capacity into expressive acoustic realization, producing speech with flattened prosody and misaligned emotion. We identify this mismatch as the semantic understanding-acoustic realization gap. Existing approaches typically rely on externally specified proxies, such as emotion labels or style prompts, which require annotations and struggle to capture dynamically evolving expressive intent throughout dialogue. To overcome these limitations, we propose SASLM (Self-Aware Speech Language Model), a proxy-free framework that bridges what the model thinks and how it speaks through self-aware intent-realization alignment: (1) Intent-Aware Bridging self-distills expressive intent from the model's own evolving semantic generation states via a Variational Information Bottleneck (VIB), thereby guiding expressive speech realization without external expressive supervision; while (2) Realization-Aware Alignment reflectively aligns generated acoustics with intended expression through self-reward optimization, progressively improving intent-realization consistency during speech generation. Despite using only 3B parameters and 800 hours of expressive speech data, SASLM achieves state-of-the-art performance on EchoMind among open-source systems, surpassing models over 10 times larger and approaching commercial systems.

cs.CL↗

arXiv2026

SCALE: Sensitivity-Aware Federated Unlearning with Information Freshness Optimization for Mobile Edge Computing

Federated Unlearning (FU) is emerging as a powerful tool that enables the selective removal of client data to effectively address data contamination and meet strict privacy regulations in mobile edge computing (MEC) systems. Although FU has recently drawn attention in the AI community, existing approaches suffer from low unlearning precision and lack temporal information reflection, which results in suboptimal forgetting performance. To address these issues, we propose SCALE, a dual-level unlearning framework combining historical contribution analysis with information freshness-aware adaptive sparsification. Our framework first employs a historical contribution-based layer sensitivity analysis to identify layers most influenced by target clients, then performs fine-grained unlearning through adaptive sparsification at the weight sub-group level to balance information freshness with forgetting effectiveness. Through theoretical analysis, the proposed framework demonstrates the convergence properties and acceleration advantages. Our experiments and testbed results demonstrate superior unlearning effectiveness compared to state-of-the-art baselines, with significantly improved forgetting performance.

cs.NI↗

arXiv2026

OmniThoughtVis: A Scalable Distillation Pipeline for Deployable Multimodal Reasoning Models

Recent multimodal large language models (MLLMs) have shown strong chain-of-thought (CoT) reasoning ability on vision-language tasks, but their direct deployment in real-world systems is often limited by latency and resource constraints. In practice, smaller MLLMs are preferred for online serving, yet their reasoning performance is bottlenecked by the lack of large-scale, high-quality multimodal CoT supervision. In this paper, we present OmniThoughtVis, a scalable data curation and distillation pipeline for transferring multimodal reasoning capabilities from high-capacity teacher models to smaller, deployment-oriented MLLMs. Starting from a diverse open-source seed pool, our pipeline generates structured CoT traces and performs joint annotation of reasoning difficulty, answer quality, and semantic task tags. To maintain data quality at scale, we combine rule-based filtering, difficulty-aware selection, and tag-based diversity sampling, resulting in a curated corpus of 1.8M samples that supports controllable subset construction for downstream training. We use OmniThoughtVis to distill Qwen3-VL models from 2B to 8B parameters and evaluate them on nine multimodal reasoning benchmarks. The resulting distilled models show consistent gains across model scales, including improvements of up to +16.8 points on MathVerse and +5.6 points on MMMU-Pro for the 4B model. Notably, the distilled 4B model matches or surpasses the undistilled 8B baseline on several tasks, highlighting the practical value of scalable reasoning distillation for deployment-oriented MLLMs.

cs.CL↗