arXiv subjects

Shuang Chen

Publications and source records attributed to Shuang Chen.

At least 55 records · Page 3Linked to original sources

arXiv2026

Taking a Deep Breath: Enhancing Language Modeling of Large Language Models with Sentinel Tokens

Large language models (LLMs) have shown promising efficacy across various tasks, becoming powerful tools in numerous aspects of human life. However, Transformer-based LLMs suffer a performance degradation when modeling long-term contexts due to they discard some information to reduce computational overhead. In this work, we propose a simple yet effective method to enable LLMs to take a deep breath, encouraging them to summarize information contained within discrete text chunks. Specifically, we segment the text into multiple chunks and insert special token at the end of each chunk. We then modify the attention mask to integrate the chunk's information into the corresponding token. This facilitates LLMs to interpret information not only from historical individual tokens but also from the token, aggregating the chunk's semantic information. Experiments on language modeling and out-of-domain downstream tasks validate the superiority of our approach.

cs.CL↗

arXiv2026

Vision-DeepResearch Benchmark: Rethinking Visual and Textual Search for Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) have advanced VQA and now support Vision-DeepResearch systems that use search engines for complex visual-textual fact-finding. However, evaluating these visual and textual search abilities is still difficult, and existing benchmarks have two major limitations. First, existing benchmarks are not visual search-centric: answers that should require visual search are often leaked through cross-textual cues in the text questions or can be inferred from the prior world knowledge in current MLLMs. Second, overly idealized evaluation scenario: On the image-search side, the required information can often be obtained via near-exact matching against the full image, while the text-search side is overly direct and insufficiently challenging. To address these issues, we construct the Vision-DeepResearch benchmark (VDR-Bench) comprising 2,000 VQA instances. All questions are created via a careful, multi-stage curation pipeline and rigorous expert review, designed to assess the behavior of Vision-DeepResearch systems under realistic real-world conditions. Moreover, to address the insufficient visual retrieval capabilities of current MLLMs, we propose a simple multi-round cropped-search workflow. This strategy is shown to effectively improve model performance in realistic visual retrieval scenarios. Overall, our results provide practical guidance for the design of future multimodal deep-research systems. The code will be released in https://github.com/Osilly/Vision-DeepResearch.

cs.CV↗

arXiv2026

Supervise-assisted Multi-modality Fusion Diffusion Model for PET Restoration

Positron emission tomography (PET) offers powerful functional imaging but involves radiation exposure. Efforts to reduce this exposure by lowering the radiotracer dose or scan time can degrade image quality. While using magnetic resonance (MR) images with clearer anatomical information to restore standard-dose PET (SPET) from low-dose PET (LPET) is a promising approach, it faces challenges with the inconsistencies in the structure and texture of multi-modality fusion, as well as the mismatch in out-of-distribution (OOD) data. In this paper, we propose a supervise-assisted multi-modality fusion diffusion model (MFdiff) for addressing these challenges for high-quality PET restoration. Firstly, to fully utilize auxiliary MR images without introducing extraneous details in the restored image, a multi-modality feature fusion module is designed to learn an optimized fusion feature. Secondly, using the fusion feature as an additional condition, high-quality SPET images are iteratively generated based on the diffusion model. Furthermore, we introduce a two-stage supervise-assisted learning strategy that harnesses both generalized priors from simulated in-distribution datasets and specific priors tailored to in-vivo OOD data. Experiments demonstrate that the proposed MFdiff effectively restores high-quality SPET images from multi-modality inputs and outperforms state-of-the-art methods both qualitatively and quantitatively.

cs.CV↗

arXiv2026

Advancing Multimodal Reasoning: From Optimized Cold Start to Staged Reinforcement Learning

Inspired by the remarkable reasoning capabilities of Deepseek-R1 in complex textual tasks, many works attempt to incentivize similar capabilities in Multimodal Large Language Models (MLLMs) by directly applying reinforcement learning (RL). However, they still struggle to activate complex reasoning. In this paper, rather than examining multimodal RL in isolation, we delve into current training pipelines and identify three crucial phenomena: 1) Effective cold start initialization is critical for enhancing MLLM reasoning. Intriguingly, we find that initializing with carefully selected text data alone can lead to performance surpassing many recent multimodal reasoning models, even before multimodal RL. 2) Standard GRPO applied to multimodal RL suffers from gradient stagnation, which degrades training stability and performance. 3) Subsequent text-only RL training, following the multimodal RL phase, further enhances multimodal reasoning. This staged training approach effectively balances perceptual grounding and cognitive reasoning development. By incorporating the above insights and addressing multimodal RL issues, we introduce ReVisual-R1, achieving a new state-of-the-art among open-source 7B MLLMs on challenging benchmarks including MathVerse, MathVision, WeMath, LogicVista, DynaMath, and challenging AIME2024 and AIME2025.

cs.LG↗

arXiv2026

Innovator-VL: A Multimodal Large Language Model for Scientific Discovery

We present Innovator-VL, a scientific multimodal large language model designed to advance understanding and reasoning across diverse scientific domains while maintaining excellent performance on general vision tasks. Contrary to the trend of relying on massive domain-specific pretraining and opaque pipelines, our work demonstrates that principled training design and transparent methodology can yield strong scientific intelligence with substantially reduced data requirements. (i) First, we provide a fully transparent, end-to-end reproducible training pipeline, covering data collection, cleaning, preprocessing, supervised fine-tuning, reinforcement learning, and evaluation, along with detailed optimization recipes. This facilitates systematic extension by the community. (ii) Second, Innovator-VL exhibits remarkable data efficiency, achieving competitive performance on various scientific tasks using fewer than five million curated samples without large-scale pretraining. These results highlight that effective reasoning can be achieved through principled data selection rather than indiscriminate scaling. (iii) Third, Innovator-VL demonstrates strong generalization, achieving competitive performance on general vision, multimodal reasoning, and scientific benchmarks. This indicates that scientific alignment can be integrated into a unified model without compromising general-purpose capabilities. Our practices suggest that efficient, reproducible, and high-performing scientific multimodal models can be built even without large-scale data, providing a practical foundation for future research.

cs.CV↗

arXiv2026

Exploring the Potentials of Spiking Neural Networks for Image Deraining

Biologically plausible and energy-efficient frameworks such as Spiking Neural Networks (SNNs) have not been sufficiently explored in low-level vision tasks. Taking image deraining as an example, this study addresses the representation of the inherent high-pass characteristics of spiking neurons, specifically in image deraining and innovatively proposes the Visual LIF (VLIF) neuron, overcoming the obstacle of lacking spatial contextual understanding present in traditional spiking neurons. To tackle the limitation of frequency-domain saturation inherent in conventional spiking neurons, we leverage the proposed VLIF to introduce the Spiking Decomposition and Enhancement Module and the lightweight Spiking Multi-scale Unit for hierarchical multi-scale representation learning. Extensive experiments across five benchmark deraining datasets demonstrate that our approach significantly outperforms state-of-the-art SNN-based deraining methods, achieving this superior performance with only 13\% of their energy consumption. These findings establish a solid foundation for deploying SNNs in high-performance, energy-efficient low-level vision tasks.

cs.CV↗

arXiv2026

Democratizing planetary-scale analysis: An ultra-lightweight Earth embedding database for accurate and flexible global land monitoring

The rapid evolution of satellite-borne Earth Observation (EO) systems has revolutionized terrestrial monitoring, yielding petabyte-scale archives. However, the immense computational and storage requirements for global-scale analysis often preclude widespread use, hindering planetary-scale studies. To address these barriers, we present Embedded Seamless Data (ESD), an ultra-lightweight, 30-m global Earth embedding database spanning the 25-year period from 2000 to 2024. By transforming high-dimensional, multi-sensor observations from the Landsat series (5, 7, 8, and 9) and MODIS Terra into information-dense, quantized latent vectors, ESD distills essential geophysical and semantic features into a unified latent space. Utilizing the ESDNet architecture and Finite Scalar Quantization (FSQ), the dataset achieves a transformative ~340-fold reduction in data volume compared to raw archives. This compression allows the entire global land surface for a single year to be encapsulated within approximately 2.4 TB, enabling decadal-scale global analysis on standard local workstations. Rigorous validation demonstrates high reconstructive fidelity (MAE: 0.0130; RMSE: 0.0179; CC: 0.8543). By condensing the annual phenological cycle into 12 temporal steps, the embeddings provide inherent denoising and a semantically organized space that outperforms raw reflectance in land-cover classification, achieving 79.74% accuracy (vs. 76.92% for raw fusion). With robust few-shot learning capabilities and longitudinal consistency, ESD provides a versatile foundation for democratizing planetary-scale research and advancing next-generation geospatial artificial intelligence.

cs.CV↗

arXiv2026

Wetland mapping from sparse annotations with satellite image time series and temporal-aware segment anything model

Accurate wetland mapping is essential for ecosystem monitoring, yet dense pixel-level annotation is prohibitively expensive and practical applications usually rely on sparse point labels, under which existing deep learning models perform poorly, while strong seasonal and inter-annual wetland dynamics further render single-date imagery inadequate and lead to significant mapping errors; although foundation models such as SAM show promising generalization from point prompts, they are inherently designed for static images and fail to model temporal information, resulting in fragmented masks in heterogeneous wetlands. To overcome these limitations, we propose WetSAM, a SAM-based framework that integrates satellite image time series for wetland mapping from sparse point supervision through a dual-branch design, where a temporally prompted branch extends SAM with hierarchical adapters and dynamic temporal aggregation to disentangle wetland characteristics from phenological variability, and a spatial branch employs a temporally constrained region-growing strategy to generate reliable dense pseudo-labels, while a bidirectional consistency regularization jointly optimizes both branches. Extensive experiments across eight global regions of approximately 5,000 km2 each demonstrate that WetSAM substantially outperforms state-of-the-art methods, achieving an average F1-score of 85.58%, and delivering accurate and structurally consistent wetland segmentation with minimal labeling effort, highlighting its strong generalization capability and potential for scalable, low-cost, high-resolution wetland mapping.

cs.CV↗

arXiv2025

IPCV: Information-Preserving Compression for MLLM Visual Encoders

Multimodal Large Language Models (MLLMs) deliver strong vision-language performance but at high computational cost, driven by numerous visual tokens processed by the Vision Transformer (ViT) encoder. Existing token pruning strategies are inadequate: LLM-stage token pruning overlooks the ViT's overhead, while conventional ViT token pruning, without language guidance, risks discarding textually critical visual cues and introduces feature distortions amplified by the ViT's bidirectional attention. To meet these challenges, we propose IPCV, a training-free, information-preserving compression framework for MLLM visual encoders. IPCV enables aggressive token pruning inside the ViT via Neighbor-Guided Reconstruction (NGR) that temporarily reconstructs pruned tokens to participate in attention with minimal overhead, then fully restores them before passing to the LLM. Besides, we introduce Attention Stabilization (AS) to further alleviate the negative influence from token pruning by approximating the K/V of pruned tokens. It can be directly applied to previous LLM-side token pruning methods to enhance their performance. Extensive experiments show that IPCV substantially reduces end-to-end computation and outperforms state-of-the-art training-free token compression methods across diverse image and video benchmarks. Our code is available at https://github.com/Perkzi/IPCV.

cs.CV↗

arXiv2025

CaFlow: Enhancing Long-Term Action Quality Assessment with Causal Counterfactual Flow

Action Quality Assessment (AQA) predicts fine-grained execution scores from action videos and is widely applied in sports, rehabilitation, and skill evaluation. Long-term AQA, as in figure skating or rhythmic gymnastics, is especially challenging since it requires modeling extended temporal dynamics while remaining robust to contextual confounders. Existing approaches either depend on costly annotations or rely on unidirectional temporal modeling, making them vulnerable to spurious correlations and unstable long-term representations. To this end, we propose CaFlow, a unified framework that integrates counterfactual de-confounding with bidirectional time-conditioned flow. The Causal Counterfactual Regularization (CCR) module disentangles causal and confounding features in a self-supervised manner and enforces causal robustness through counterfactual interventions, while the BiT-Flow module models forward and backward dynamics with a cycle-consistency constraint to produce smoother and more coherent representations. Extensive experiments on multiple long-term AQA benchmarks demonstrate that CaFlow achieves state-of-the-art performance. Code is available at https://github.com/Harrison21/CaFlow

cs.CV↗

arXiv2025

Sequential dichotomies and uniformities for a neutral equation

Sequential dichotomies of general delay equations are not uniform, which was proved two decades ago. This however reminds whether the countably infinite many dichotomies of a neutral equation have the sequential uniformity. In this paper, considering a scalar neutral equation, we give a negative answer and prove that the series of the projections of dichotomies is divergent.

math.DS↗

arXiv2025

Harli: SLO-Aware Co-location of LLM Inference and PEFT-based Finetuning on Model-as-a-Service Platforms

Large language models (LLMs) are increasingly deployed under the Model-as-a-Service (MaaS) paradigm. To meet stringent quality-of-service (QoS) requirements, existing LLM serving systems disaggregate the prefill and decode phases of inference. However, decode instances often experience low GPU utilization due to their memory-bound nature and insufficient batching in dynamic workloads, leaving compute resources underutilized. We introduce Harli, a serving system that improves GPU utilization by co-locating parameter-efficient finetuning (PEFT) tasks with LLM decode instances. PEFT tasks are compute-bound and memory-efficient, making them ideal candidates for safe co-location. Specifically, Harli addresses key challenges--limited memory and unpredictable interference--using three components: a unified memory allocator for runtime memory reuse, a two-stage latency predictor for decode latency modeling, and a QoS-guaranteed throughput-maximizing scheduler for throughput maximization. Experimental results show that Harli improves the finetune throughput by 46.2% on average (up to 92.0%) over state-of-the-art serving systems, while maintaining strict QoS guarantees for inference decode.

cs.DC↗

arXiv2025

ARES: Multimodal Adaptive Reasoning via Difficulty-Aware Token-Level Entropy Shaping

Recent advances in multimodal large reasoning models (MLRMs) have substantially improved their ability to solve complex textual and visual tasks. However, these models tend to overthink on simple problems, producing unnecessarily lengthy reasoning traces, while under-exploring on challenging ones, leading to missed solutions. To address this imbalance, we propose ARES, a unified open-source framework for adaptive reasoning that dynamically allocates exploration effort based on task difficulty. Our approach is motivated by two key empirical findings: (i) while single-token entropy is noisy, high window-entropy (HWE) tokens (token-level entropies averaged under a sliding window) can reliably capture reasoning-critical moments; and (ii) reducing HWE usage benefits easy problems, while increasing it is essential for solving hard ones. Building on these insights, ARES introduces a two-stage training pipeline. In the Adaptive Cold-Start stage, we curate multimodal and textual data paired with reasoning traces of length proportional to problem difficulty, equipping the model with initial difficulty awareness. In the second stage, we develop Adaptive Entropy Policy Optimization (AEPO), which uses HWE tokens as exploration triggers to decide when to explore, and a hierarchical entropy reward with dynamic KL control to decide how much to explore. Extensive experiments demonstrate that ARES achieves superior performance and reasoning efficiency across diverse mathematical, logical, and multimodal benchmarks, while closing the gap to leading commercial systems under significantly lower inference costs.

cs.CL↗

arXiv2025

BcQLM: Efficient Vision-Language Understanding with Distilled Q-Gated Cross-Modal Fusion

As multimodal large language models (MLLMs) advance, their large-scale architectures pose challenges for deployment in resource-constrained environments. In the age of large models, where energy efficiency, computational scalability and environmental sustainability are paramount, the development of lightweight and high-performance models is critical for real-world applications. As such, we propose a lightweight MLLM framework for end-to-end visual question answering. Our proposed approach centres on BreezeCLIP, a compact yet powerful vision-language encoder optimised for efficient multimodal understanding. With only 1.2 billion parameters overall, our model significantly reduces computational cost while achieving performance comparable to standard-size MLLMs. Experiments conducted on multiple datasets further validate its effectiveness in balancing accuracy and efficiency. The modular and extensible design enables generalisation to broader multimodal tasks. The proposed lightweight vision-language framework is denoted as BcQLM (BreezeCLIP-enhanced Q-Gated Multimodal Language Model). It offers a promising path toward deployable MLLMs under practical hardware constraints. The source code is available at https://github.com/thico0224/BcQLM.

cs.CV↗

arXiv2025

Interleaving Reasoning for Better Text-to-Image Generation

Unified multimodal understanding and generation models recently have achieve significant improvement in image generation capability, yet a large gap remains in instruction following and detail preservation compared to systems that tightly couple comprehension with generation such as GPT-4o. Motivated by recent advances in interleaving reasoning, we explore whether such reasoning can further improve Text-to-Image (T2I) generation. We introduce Interleaving Reasoning Generation (IRG), a framework that alternates between text-based thinking and image synthesis: the model first produces a text-based thinking to guide an initial image, then reflects on the result to refine fine-grained details, visual quality, and aesthetics while preserving semantics. To train IRG effectively, we propose Interleaving Reasoning Generation Learning (IRGL), which targets two sub-goals: (1) strengthening the initial think-and-generate stage to establish core content and base quality, and (2) enabling high-quality textual reflection and faithful implementation of those refinements in a subsequent image. We curate IRGL-300K, a dataset organized into six decomposed learning modes that jointly cover learning text-based thinking, and full thinking-image trajectories. Starting from a unified foundation model that natively emits interleaved text-image outputs, our two-stage training first builds robust thinking and reflection, then efficiently tunes the IRG pipeline in the full thinking-image trajectory data. Extensive experiments show SoTA performance, yielding absolute gains of 5-10 points on GenEval, WISE, TIIF, GenAI-Bench, and OneIG-EN, alongside substantial improvements in visual quality and fine-grained fidelity. The code, model weights and datasets will be released in: https://github.com/Osilly/Interleaving-Reasoning-Generation .

cs.CV↗

arXiv2025

DEEP-SEA: Deep-Learning Enhancement for Environmental Perception in Submerged Aquatics

Continuous and reliable underwater monitoring is essential for assessing marine biodiversity, detecting ecological changes and supporting autonomous exploration in aquatic environments. Underwater monitoring platforms rely on mainly visual data for marine biodiversity analysis, ecological assessment and autonomous exploration. However, underwater environments present significant challenges due to light scattering, absorption and turbidity, which degrade image clarity and distort colour information, which makes accurate observation difficult. To address these challenges, we propose DEEP-SEA, a novel deep learning-based underwater image restoration model to enhance both low- and high-frequency information while preserving spatial structures. The proposed Dual-Frequency Enhanced Self-Attention Spatial and Frequency Modulator aims to adaptively refine feature representations in frequency domains and simultaneously spatial information for better structural preservation. Our comprehensive experiments on EUVP and LSUI datasets demonstrate the superiority over the state of the art in restoring fine-grained image detail and structural consistency. By effectively mitigating underwater visual degradation, DEEP-SEA has the potential to improve the reliability of underwater monitoring platforms for more accurate ecological observation, species identification and autonomous navigation.

cs.CV↗

arXiv2025

Implicit Search Intent Recognition using EEG and Eye Tracking: Novel Dataset and Cross-User Prediction

For machines to effectively assist humans in challenging visual search tasks, they must differentiate whether a human is simply glancing into a scene (navigational intent) or searching for a target object (informational intent). Previous research proposed combining electroencephalography (EEG) and eye-tracking measurements to recognize such search intents implicitly, i.e., without explicit user input. However, the applicability of these approaches to real-world scenarios suffers from two key limitations. First, previous work used fixed search times in the informational intent condition -- a stark contrast to visual search, which naturally terminates when the target is found. Second, methods incorporating EEG measurements addressed prediction scenarios that require ground truth training data from the target user, which is impractical in many use cases. We address these limitations by making the first publicly available EEG and eye-tracking dataset for navigational vs. informational intent recognition, where the user determines search times. We present the first method for cross-user prediction of search intents from EEG and eye-tracking recordings and reach 84.5% accuracy in leave-one-user-out evaluations -- comparable to within-user prediction accuracy (85.5%) but offering much greater flexibility

cs.HC↗

arXiv2025

Dynamic mapping from static labels: remote sensing dynamic sample generation with temporal-spectral embedding

Accurate remote sensing geographic mapping requires timely and representative samples. However, rapid land surface changes often render static samples obsolete within months, making manual sample updates labor-intensive and unsustainable. To address this challenge, we propose TasGen, a two-stage Temporal spectral-aware Automatic Sample Generation method for generating dynamic training samples from single-date static labels without human intervention. Land surface dynamics often manifest as anomalies in temporal-spectral sequences. %These anomalies are multivariate yet unified: temporal, spectral, or joint anomalies stem from different mechanisms and cannot be naively coupled, as this may obscure the nature of changes. Yet, any land surface state corresponds to a coherent temporal-spectral signature, which would be lost if the two dimensions are modeled separately. To effectively capture these dynamics, TasGen first disentangles temporal and spectral features to isolate their individual contributions, and then couples them to model their synergistic interactions. In the first stage, we introduce a hierarchical temporal-spectral variational autoencoder (HTS-VAE) with a dual-dimension embedding to learn low-dimensional latent patterns of normal samples by first disentangling and then jointly embedding temporal and spectral information. This temporal-spectral embedding enables robust anomaly detection by identifying deviations from learned joint patterns. In the second stage, a classifier trained on stable samples relabels change points across time to generate dynamic samples. To not only detect but also explain surface dynamics, we further propose an anomaly interpretation method based on Gibbs sampling, which attributes changes to specific spectral-temporal dimensions.

eess.IV↗