Searcharxiv⌕ Search

arXiv subjects

Tao Chen

Publications and source records attributed to Tao Chen.

At least 91 records · Page 5Linked to original sources
arXiv2026

FRISM: Fine-Grained Reasoning Injection via Subspace-Level Model Merging for Vision-Language Models

Efficiently enhancing the reasoning capabilities of Vision-Language Models (VLMs) by merging them with Large Reasoning Models (LRMs) has emerged as a promising direction. However, existing methods typically operate at a coarse-grained layer level, which often leads to a trade-off between injecting reasoning capabilities and preserving visual capabilities. To address this limitation, we propose FRISM (Fine-grained Reasoning Injection via Subspace-level model Merging), a fine-grained reasoning injection framework based on subspace-level model merging. Observing that different SVD subspaces contribute differently to reasoning and perception, FRISM decomposes LRM task vectors via Singular Value Decomposition (SVD) and adaptively tunes the scaling coefficients of each subspace through learning to realize fine-grained reasoning injection. Furthermore, we introduce a label-free self-distillation learning strategy with dual-objective optimization using common vision-language perception datasets. Extensive experiments demonstrate that FRISM effectively improves reasoning capabilities while largely preserving the model's visual capabilities by consistently achieving strong performance across diverse visual-language reasoning benchmarks.

cs.CV↗
arXiv2026

FCMBench-Video: Benchmarking Document Video Intelligence

Document understanding is a critical capability in financial credit review, onboarding, and remote verification, where both decision accuracy and evidence traceability matter. Compared with static document images, document videos present a temporally redundant and sequentially unfolding evidence stream, require evidence integration across frames, and preserve acquisition-process cues relevant to authenticity-sensitive and anti-fraud review. We introduce FCMBench-Video, a benchmark for document-video intelligence that evaluates document perception, temporal grounding, and evidence-grounded reasoning under realistic capture conditions. For privacy-compliant yet realistic data at scale, we organize construction as an atomic-acquisition and composition workflow that records reusable single-document clips, applies controlled degradations, and assembles long-form multi-document videos with prescribed temporal spans. FCMBench-Video is built from 495 atomic videos composed into 1,200 long-form videos paired with 11,322 expert-annotated question--answer instances, covering 28 document types over 20s--60s duration tiers and 5,960 Chinese / 5,362 English instances. Evaluations on nine recent Video-MLLMs show that FCMBench-Video provides meaningful separation across systems and capabilities: counting is the most duration-sensitive task, Cross-Document Validation and Evidence-Grounded Selection probe higher-level evidence integration, and Visual Prompt Injection provides a complementary robustness dimension. The overall score distribution is broad and approximately bell-shaped, indicating a benchmark that is neither saturated nor dominated by trivial cases. Together, these results position FCMBench-Video as a reproducible benchmark for tracking Video-MLLM progress on document-video understanding and probing capability boundaries in authenticity-sensitive credit-domain applications.

cs.CV↗
arXiv2026

Meromorphic functions whose action on their Julia sets is Non-Ergodic

Nevanlinna functions are meromorphic functions with a finite number of asymptotic values and no critical values. In [KK2] it was proved that if the orbits of all the asymptotic values accumulate on a compact set on which the function acts as a repeller, then the function acts ergodically on its Julia set. In [CJK4] we proved the action of the function on its Julia set is still ergodic if some, but not all of the asymptotic values land on infinity, and the remaining ones land on a compact repeller. In this paper, we complete the characterization of ergodicity for Nevanlinna functions by proving that if all the asymptotic values land on infinity, then the Julia set is the whole sphere and the action of the map there is non-ergodic.

math.DS↗
arXiv2026

Conjecture and Inquiry: Quantifying Software Performance Requirements via Interactive Retrieval-Augmented Preference Elicitation

Since software performance requirements are documented in natural language, quantifying them into mathematical forms is essential for software engineering. Yet, the vagueness in performance requirements and uncertainty of human cognition have caused highly uncertain ambiguity in the interpretations, rendering their automated quantification an unaddressed and challenging problem. In this paper, we formalize the problem and propose IRAP, an approach that quantifies performance requirements into mathematical functions via interactive retrieval-augmented preference elicitation. IRAP differs from the others in that it explicitly derives from problem-specific knowledge to retrieve and reason the preferences, which also guides the progressive interaction with stakeholders, while reducing the cognitive overhead. Experiment results against 10 state-of-the-art methods on four real-world datasets demonstrate the superiority of IRAP on all cases with up to 40x improvements under as few as five rounds of interactions.

cs.SE↗
arXiv2026

Can Multimodal Large Language Models Truly Understand Small Objects?

Multimodal Large Language Models (MLLMs) have shown promising potential in diverse understanding tasks, e.g., image and video analysis, math and physics olympiads. However, they remain blank and unexplored for Small Object Understanding (SOU) tasks. To fill this gap, we introduce SOUBench, the first and comprehensive benchmark for exploring the small objects understanding capability of existing MLLMs. Specifically, we first design an effective and automatic visual question-answer generation strategy, constructing a new SOU-VQA evaluation dataset, with 18,204 VQA pairs, six relevant sub-tasks, and three dominant scenarios (i.e., Driving, Aerial, and Underwater). Then, we conduct a comprehensive evaluation on 15 state-of-the-art MLLMs and reveal their weak capabilities in small object understanding. Furthermore, we develop SOU-Train, a multimodal training dataset with 11,226 VQA pairs, to improve the SOU capabilities of MLLMs. Through supervising fine-tuning of the latest MLLM, we demonstrate that SOU-Train can effectively enhance the latest MLLM's ability to understand small objects. Comprehensive experimental results demonstrate that, the proposed SOUBench, along with the SOU-VQA and SOU-Train datasets, provides a crucial empirical foundation to the community for further developing models with enhanced small object understanding capabilities. Datasets and Code: https://github.com/Hanfj-X/SOU.

cs.CV↗
arXiv2026

A transfer-learning-enhanced POD-FNN surrogate for rapid signal prediction and inverse fitting in thermoreflectance with patterned transducers

Patterned-transducer thermoreflectance enhances sensitivity to low-thermal-conductivity materials by suppressing lateral heat spreading in the metal transducer, but its wider use is limited by the cost of repeated high-fidelity forward evaluations in iterative fitting. Here, we develop a transfer-learning-enhanced POD-FNN surrogate for rapid phase prediction in patterned-transducer thermoreflectance, using patterned FDTR as a representative case. A validated COMSOL model is first constructed, and proper orthogonal decomposition is applied directly to the phase signals to build a compact reduced-order representation. A feedforward neural network is then trained to predict the POD coefficients from thermophysical and geometric parameters. Within the original parameter domain, the surrogate achieves mean and median RMSE values of 0.19 and 0.17 degrees, with a maximum RMSE below 0.47 degrees, while reducing the average prediction time per signal from 5.39 s to 0.01 s (about 534x). In inverse analysis, the fitting time for a representative case is reduced from about 18950 s to about 65 s with comparable accuracy. The framework is further applied to measured Al/SiO2 samples, yielding stable silica thermal conductivities of 1.44 +/- 0.088, 1.43 +/- 0.093, and 1.50 +/- 0.079 W/(m K) for conventional FDTR and patterned FDTR with pattern radii of 5.3 and 3.25 um, respectively. Transfer learning further improves performance in expanded parameter domains, with the TL-FR strategy giving the best overall results. Reducing the additional target-domain dataset from 6000 to 1000 samples also lowers the high-fidelity data-generation time from about 34179 s to about 5885 s. The proposed framework provides an accurate and efficient route for repeated forward evaluation, rapid inverse fitting, and cost-effective model updating in patterned thermoreflectance workflows.

physics.app-ph↗
arXiv2026

Rethinking Cross-Domain Evaluation for Face Forgery Detection with Semantic Fine-grained Alignment and Mixture-of-Experts

Nowadays, visual data forgery detection plays an increasingly important role in social and economic security with the rapid development of generative models. Existing face forgery detectors still can't achieve satisfactory performance because of poor generalization ability across datasets. The key factor that led to this phenomenon is the lack of suitable metrics: the commonly used cross-dataset AUC metric fails to reveal an important issue where detection scores may shift significantly across data domains. To explicitly evaluate cross-domain score comparability, we propose \textbf{Cross-AUC}, an evaluation metric that can compute AUC across dataset pairs by contrasting real samples from one dataset with fake samples from another (and vice versa). It is interesting to find that evaluating representative detectors under the Cross-AUC metric reveals substantial performance drops, exposing an overlooked robustness problem. Besides, we also propose the novel framework \textbf{S}emantic \textbf{F}ine-grained \textbf{A}lignment and \textbf{M}ixture-of-Experts (\textbf{SFAM}), consisting of a patch-level image-text alignment module that enhances CLIP's sensitivity to manipulation artifacts, and the facial region mixture-of-experts module, which routes features from different facial regions to specialized experts for region-aware forgery analysis. Extensive qualitative and quantitative experiments on the public datasets prove that the proposed method achieves superior performance compared with the state-of-the-art methods with various suitable metrics.

cs.CV↗
arXiv2026

MLG-Stereo: ViT Based Stereo Matching with Multi-Stage Local-Global Enhancement

With the development of deep learning, ViT-based stereo matching methods have made significant progress due to their remarkable robustness and zero-shot ability. However, due to the limitations of ViTs in handling resolution sensitivity and their relative neglect of local information, the ability of ViT-based methods to predict details and handle arbitrary-resolution images is still weaker than that of CNN-based methods. To address these shortcomings, we propose MLG-Stereo, a systematic pipeline-level design that extends global modeling beyond the encoder stage. First, we propose a Multi-Granularity Feature Network to effectively balance global context and local geometric information, enabling comprehensive feature extraction from images of arbitrary resolution and bridging the gap between training and inference scales. Then, a Local-Global Cost Volume is constructed to capture both locally-correlated and global-aware matching information. Finally, a Local-Global Guided Recurrent Unit is introduced to iteratively optimize the disparity locally under the guidance of global information. Extensive experiments are conducted on multiple benchmark datasets, demonstrating that our MLG-Stereo exhibits highly competitive performance on the Middlebury and KITTI-2015 benchmarks compared to contemporaneous leading methods, and achieves outstanding results in the KITTI-2012 dataset.

cs.CV↗
arXiv2026

LSTM-MAS: A Long Short-Term Memory Inspired Multi-Agent System for Long-Context Understanding

Effectively processing long contexts remains a fundamental yet unsolved challenge for large language models (LLMs). Existing single-LLM-based methods primarily reduce the context window or optimize the attention mechanism, but they often encounter additional computational costs or constrained expanded context length. While multi-agent-based frameworks can mitigate these limitations, they remain susceptible to the accumulation of errors and the propagation of hallucinations. In this work, we draw inspiration from the Long Short-Term Memory (LSTM) architecture to design a Multi-Agent System called LSTM-MAS, emulating LSTM's hierarchical information flow and gated memory mechanisms for long-context understanding. Specifically, LSTM-MAS organizes agents in a chained architecture, where each node comprises a worker agent for segment-level comprehension, a filter agent for redundancy reduction, a judge agent for continuous error detection, and a manager agent for globally regulates information propagation and retention, analogous to LSTM and its input gate, forget gate, constant error carousel unit, and output gate. These novel designs enable controlled information transfer and selective long-term dependency modeling across textual segments, which can effectively avoid error accumulation and hallucination propagation. We conducted an extensive evaluation of our method. Compared with the previous best multi-agent approach, CoA, our model achieves improvements of 97.97%, 65.75%, 122.19%, 39.61% and 10.80% on Narrative QA, Qasper, HotpotQA, 2WikiMQA and MuSiQue, respectively.

cs.CL↗
arXiv2026

Human-Machine Co-Boosted Bug Report Identification with Mutualistic Neural Active Learning

Bug reports, encompassing a wide range of bug types, are crucial for maintaining software quality. However, the increasing complexity and volume of bug reports pose a significant challenge in sole manual identification and assignment to the appropriate teams for resolution, as dealing with all the reports is time-consuming and resource-intensive. In this paper, we introduce a cross-project framework, dubbed Mutualistic Neural Active Learning (MNAL), designed for automated and more effective identification of bug reports from GitHub repositories boosted by human-machine collaboration. MNAL utilizes a neural language model that learns and generalizes reports across different projects, coupled with active learning to form neural active learning. A distinctive feature of MNAL is the purposely crafted mutualistic relation between the machine learners (neural language model) and human labelers (developers) when enriching the knowledge learned. That is, the most informative human-labeled reports and their corresponding pseudo-labeled ones are used to update the model while those reports that need to be labeled by developers are more readable and identifiable, thereby enhancing the human-machine teaming therein. We evaluate MNAL using a large scale dataset against the SOTA approaches, baselines, and different variants. The results indicate that MNAL achieves up to 95.8% and 196.0% effort reduction in terms of readability and identifiability during human labeling, respectively, while resulting in a better performance in bug report identification. Additionally, our MNAL is model-agnostic since it is capable of improving the model performance with various underlying neural language models. To further verify the efficacy of our approach, we conducted a qualitative case study involving 10 human participants, who rate MNAL as being more effective while saving more time and monetary resources.

cs.SE↗
arXiv2026

Deep Learning for Multi-Antenna Modulation Recognition of Radio Signals

Multi-antenna receiving systems have become a prevalent technical solution in communication systems. Meanwhile, deep learning has achieved significant progress in automatic modulation recognition tasks in single-antenna systems. However, the application of deep learning in multi-antenna modulation recognition (MAMR) tasks is still limited. In this paper, we propose an MAMR method namely MAMR-IQ to fully explore the diversity gain of a multi-antenna receiving system, which concatenates the raw received in-phase and quadrature (IQ) signals of multiple antennas and feeds them into a convolutional neural network. Simulation results show that the proposed MAMR-IQ method outperforms two existing deep learning-based MAMR methods which are based on direct voting (DV) and weight average (WA) in terms of both recognition accuracy and computational complexity. To address the problem of limited training data in few-shot scenarios, we further propose a data augmentation method that involves exchanging IQ sequences received by any two antennas to generate augmented samples. Simulation results show that with the proposed augmentation method, the recognition accuracy can be further improved.

eess.SP↗
arXiv2026

Small Model as Master Orchestrator: Learning Unified Agent-Tool Orchestration with Parallel Subtask Decomposition

Multi-agent systems (MAS) demonstrate clear advantages in tackling complex problems by coordinating diverse agents and external tools. However, most existing orchestration methods rely on static workflows or serial agent scheduling, and are further constrained by heterogeneous interface protocols between tools and agents. This leads to high system complexity and poor extensibility. To mitigate these issues, we propose Agent-as-Tool, a unified parallel orchestration paradigm that abstracts both agents and tools into a standardized, learnable action space with protocol normalization and explicit state feedback. Building on this paradigm, we train a lightweight orchestrator, ParaManager, which decouples planning decisions from subtask solving, enabling state-aware parallel subtask decomposition, delegation, and asynchronous execution. For training, we adopt a two-stage ParaManager training pipeline. It improves robustness by incorporating supervised fine-tuning (SFT) trajectories equipped with recovery mechanisms, and further applies reinforcement learning (RL) to achieve an optimal balance among task success, protocol compliance, diversity, and reasoning efficiency. Experiments show that ParaManager achieves strong performance across multiple benchmarks and exhibits robust generalization under unseen model pools.

cs.AI↗
arXiv2026

Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges

Reinforcement Learning from Human Feedback (RLHF) and related alignment paradigms have become central to steering large language models (LLMs) and multimodal large language models (MLLMs) toward human-preferred behaviors. However, these approaches introduce a systemic vulnerability: reward hacking, where models exploit imperfections in learned reward signals to maximize proxy objectives without fulfilling true task intent. As models scale and optimization intensifies, such exploitation manifests as verbosity bias, sycophancy, hallucinated justification, benchmark overfitting, and, in multimodal settings, perception--reasoning decoupling and evaluator manipulation. Recent evidence further suggests that seemingly benign shortcut behaviors can generalize into broader forms of misalignment, including deception and strategic gaming of oversight mechanisms. In this survey, we propose the Proxy Compression Hypothesis (PCH) as a unifying framework for understanding reward hacking. We formalize reward hacking as an emergent consequence of optimizing expressive policies against compressed reward representations of high-dimensional human objectives. Under this view, reward hacking arises from the interaction of objective compression, optimization amplification, and evaluator--policy co-adaptation. This perspective unifies empirical phenomena across RLHF, RLAIF, and RLVR regimes, and explains how local shortcut learning can generalize into broader forms of misalignment, including deception and strategic manipulation of oversight mechanisms. We further organize detection and mitigation strategies according to how they intervene on compression, amplification, or co-adaptation dynamics. By framing reward hacking as a structural instability of proxy-based alignment under scale, we highlight open challenges in scalable oversight, multimodal grounding, and agentic autonomy.

cs.LG↗
arXiv2026

Universal thermometry of solid-liquid interfacial thermal conductance

Solid-liquid interfacial thermal conductance (ITC) critically influences heat transport in microfluidic, electronic, and energy systems, yet most optical thermometry techniques are limited to specific metal-liquid interfaces. In this work, we introduce a universal broadband square-pulsed thermometry method that enables simultaneous quantification of ITC across a wide range of arbitrary solid-liquid interfaces, while also providing accurate measurements of nanoscale liquid-film thickness. To validate the method, we applied it to Al-water interfaces, yielding ITC values in the range of 50-55 MW m^(-2) K^(-1), consistent with prior studies. The technique also reveals markedly lower ITCs for glass-water (9.9 MW m^(-2) K^(-1)) and Si-water (5.7 MW m^(-2) K^(-1)), and further measurements on Al-silicone oil (~10 MW m^(-2) K^(-1)) and PMMA-silicone oil (~0.4 MW m^(-2) K^(-1)) extend the validation to highly viscous nonpolar liquids and polymer-liquid interfaces. These results highlight the capability of the method to capture thermal transport differences across diverse solid-liquid combinations. Further comparisons with acoustic/diffuse mismatch models and molecular dynamics simulations, together with theoretical analysis, highlight the influence of vibrational mismatch, wettability, and surface condition on interfacial thermal transport. This broadly applicable technique enables rapid, quantitative characterization of solid-liquid interfacial thermal transport, with broad implications for interfacial heat transfer science and technology.

physics.app-ph↗
arXiv2026

Simultaneous, Non-Contact Measurement of Liquid and Interfacial Thermal Properties via a Differential Square-Pulsed Source Method

Accurate characterization of heat transport across solid-liquid interfaces is essential for thermal management in micro and nanoscale systems. Yet existing techniques often require prior knowledge of liquid properties, which complicates the simultaneous resolution of interfacial and bulk behaviors, and lose sensitivity once interfacial conductance exceeds 100 MW m-2 K-1. Here we present a differential square pulsed source (DSPS) method that provides simultaneous, non-contact measurement of liquid thermal conductivity, volumetric heat capacity, and solid-liquid interfacial conductance without any predefined material parameters. Dual frequency excitation combined with in-situ substrate referencing enables property extraction from multilayer structures, and numerical simulations show a typical uncertainty of about 8 % in interfacial conductance, confirming robustness. The protocol is validated for a wide spectrum of liquids, including oils, lubricants, aqueous electrolytes, and pure water, with excellent agreement with literature values for bulk properties. Analysis of the data set clarifies how vibrational spectrum mismatch, ionic layering, and related interfacial phenomena govern heat transfer, and demonstrates that oleophilic hexadecyl silane modification of aluminum increases interfacial conductance by a factor of sixteen. The results reveal that conductance can be strongly tuned through surface wettability and chemical functionalization, offering direct guidelines for interface engineering. Because the approach is readily extendable to soft materials such as thermal interface gels, it promises broad applicability in emerging interface-dominated thermal technologies.

physics.app-ph↗
arXiv2026

A Variable-Spot-Size and Multi-Frequency Square-Pulsed Source (SPS) Approach for Comprehensive Characterization of Anisotropic Thermal Transport Properties in Multilayered Thin Films

Multilayered thin-film structures are frequently encountered in industrial applications, where accurate thermal property characterization is essential for performance optimization. These films, typically ranging from nanometers to micrometers in thickness, often exhibit anisotropic thermal conductivity and non-bulk heat capacity, which are challenging to measure. In this study, we introduce a variable-spot-size and multi-frequency square-pulsed source (SPS) method for the simultaneous determination of anisotropic thermal conductivities, heat capacities, and interfacial thermal conductance in multilayered systems. By leveraging a broad modulation frequency range (1 Hz to 10 MHz) and tunable laser spot sizes, the SPS method enhances sensitivity to different thermal parameters across layers. We validate this approach on a silicon-on-insulator (SOI) sample comprising a 1.59 um Si layer, 1.03 um SiO2 layer, and a silicon substrate with a 122 nm aluminum (Al) transducer. The SPS method successfully extracts seven key thermal parameters, including the in-plane and cross-plane thermal conductivities and heat capacity of the Si film, the thermal conductivity and heat capacity of the SiO2 layer, the thermal conductivity of the substrate, and the interfacial thermal conductance between Al and Si. Temperature-dependent measurements from 80 to 500 K showed excellent agreement with literature values and first-principles predictions, confirming the method's accuracy and reliability. These results demonstrate the SPS method as a powerful tool for comprehensive thermal characterization of complex multilayered structures, with implications for both fundamental research and practical applications.

physics.app-ph↗
arXiv2026

Anisotropic Thermal Characterization of Suspended and Spin-Coated Polyimide Films Using a Square-Pulsed Source Method

Polyimide (PI) thin films are widely used in advanced technologies, yet accurate characterization of their thermal properties remains challenging, as evidenced by significant inconsistencies in reported data and an incomplete understanding of heat transfer mechanisms. In this study, we employ an optical Square-Pulsed Source (SPS) technique to simultaneously measure the in-plane and cross-plane thermal conductivities, as well as the volumetric heat capacity, of PI thin films. SPS is a pump-probe method that utilizes a square-wave-modulated pump laser to induce periodic heating and a probe laser to detect the thermoreflectance response. Thermal properties are extracted by analyzing amplitude signals across multiple modulation frequencies and laser spot sizes. Measurements were conducted on both suspended commercial PI films and spin-coated PI films on fused silica substrates. The results show that spin-coated films exhibit higher cross-plane thermal conductivity and lower anisotropy compared to suspended films, which we attribute to differences in molecular orientation and substrate interactions. These findings provide new physical insights into anisotropic heat transport in polymer thin films and demonstrate the SPS technique as a robust tool for probing microscale thermal phenomena in soft materials.

physics.app-ph↗
arXiv2026

Trust-region filter algorithms utilizing Hessian information for gray-box optimization

Optimizing industrial processes often involves gray-box models that couple algebraic glass-box equations with black-box components lacking analytic derivatives. Such systems challenge derivative-based solvers. The classical trust-region filter (TRF) algorithm provides a robust framework but requires extensive parameter tuning and numerous black-box evaluations. This work introduces four Hessian-informed TRF variants that use projected positive definite Hessians for automatic step scaling and minimal tuning, combined with both low-fidelity (linear, quadratic) and high-fidelity (Taylor series, Gaussian process) surrogates for local black-box approximation. Tested on 25 gray-box benchmarks and five engineering case studies, the new variants achieved up to order-of-magnitude reductions in iterations and black-box evaluations, with reduced sensitivity to tuning parameters relative to the classical TRF algorithm. High-fidelity surrogates solved 92%-100% of problems, compared with 72%-84% for low-fidelity surrogates. The developed TRF methods also outperformed classical derivative-free optimization solvers. Results show that new variants offer robust, scalable alternatives for gray-box optimization.

math.OC↗