arXiv subjects

Tianyu Yang

Publications and source records attributed to Tianyu Yang.

At least 55 records · Page 3Linked to original sources

arXiv2025

Explainable AI-Generated Image Detection RewardBench

Conventional, classification-based AI-generated image detection methods cannot explain why an image is considered real or AI-generated in a way a human expert would, which reduces the trustworthiness and persuasiveness of these detection tools for real-world applications. Leveraging Multimodal Large Language Models (MLLMs) has recently become a trending solution to this issue. Further, to evaluate the quality of generated explanations, a common approach is to adopt an "MLLM as a judge" methodology to evaluate explanations generated by other MLLMs. However, how well those MLLMs perform when judging explanations for AI-generated image detection generated by themselves or other MLLMs has not been well studied. We therefore propose \textbf{XAIGID-RewardBench}, the first benchmark designed to evaluate the ability of current MLLMs to judge the quality of explanations about whether an image is real or AI-generated. The benchmark consists of approximately 3,000 annotated triplets sourced from various image generation models and MLLMs as policy models (detectors) to assess the capabilities of current MLLMs as reward models (judges). Our results show that the current best reward model scored 88.76\% on this benchmark (while human inter-annotator agreement reaches 98.30\%), demonstrating that a visible gap remains between the reasoning abilities of today's MLLMs and human-level performance. In addition, we provide an analysis of common pitfalls that these models frequently encounter. Code and benchmark are available at https://github.com/RewardBench/XAIGID-RewardBench.

cs.CV↗

arXiv2025

ALDEN: Reinforcement Learning for Active Navigation and Evidence Gathering in Long Documents

Vision-language models (VLMs) excel at interpreting text-rich images but struggle with long, visually complex documents that demand analysis and integration of information spread across multiple pages. Existing approaches typically rely on fixed reasoning templates or rigid pipelines, which force VLMs into a passive role and hinder both efficiency and generalization. We present Active Long-DocumEnt Navigation (ALDEN), a multi-turn reinforcement learning framework that fine-tunes VLMs as interactive agents capable of actively navigating long, visually rich documents. ALDEN introduces a novel fetch action that directly accesses the page by index, complementing the classic search action and better exploiting document structure. For dense process supervision and efficient training, we propose a rule-based cross-level reward that provides both turn- and token-level signals. To address the empirically observed training instability caused by numerous visual tokens from long documents, we further propose a visual-semantic anchoring mechanism that applies a dual-path KL-divergence constraint to stabilize visual and textual representations separately during training. Trained on a corpus constructed from three open-source datasets, ALDEN achieves state-of-the-art performance on five long-document benchmarks. Overall, ALDEN marks a step beyond passive document reading toward agents that autonomously navigate and reason across long, visually rich documents, offering a robust path to more accurate and efficient long-document understanding.

cs.AI↗

arXiv2025

Near-Field 3D Localization and MIMO Channel Estimation with Sub-Connected Planar Arrays

This paper investigates the design of channel estimation and 3D localization algorithms in a challenging scenario, where a sub-connected planar extremely large-scale multiple-input multiple-output (XL-MIMO) communicates with multi-antenna users. In the near field, the uplink MIMO channel is of full column rank and therefore can not be estimated effectively by applying existing codebooks that are designed for the far-field case or for the near-field case but limited to single antenna users. To solve this problem, we propose a three-stage algorithm aided by orthogonal matching pursuit (OMP) and sparse Bayesian learning (SBL). Specifically, we firstly partition the XL-MIMO into subarrays and use OMP to solve the compressed sensing (CS) problem about subarray channel estimation with the Discrete Fourier Transform (DFT)-based dictionary matrix. Secondly, exploiting the estimated subarray channels and employing one-dimensional multiple signal classification (MUSIC), we estimate the central location of the user array under the Least Squares (LS) criterion. Finally, we utilize the estimated central location to construct a refined location-aided dictionary matrix and obtain the MIMO channel estimation using SBL. Results exhibit the significant superiority of the proposed algorithm compared with several benchmarks, in terms of both the pilot overhead and estimation accuracy.

eess.SP↗

arXiv2025

Near-Field Imaging by Exploiting Frequency Correlation in Wireless Communication Networks

In this work, we address the near-field imaging under a wideband wireless communication network by exploiting both the near-field channel of a uniform linear array (ULA) and the image correlation in the frequency domain. We first formulate the image recovery as a special multiple measurement vector (MMV) compressed sensing (CS) problem, where at various frequencies the sensing matrices can be different, and the image coefficients are correlated. To solve such an MMV problem with various sensing matrices and correlated coefficients, we propose a sparse Bayesian learning (SBL)-based solution to simultaneously estimate all image coefficients and their correlation on multiple frequencies. Moreover, to enhance estimation performance, we design two illumination patterns following two different criteria. From the CS perspective, the first design minimizes the total coherence of the sensing matrix to increase the mutual orthogonality of the basis vectors. Alternatively, to improve SNR, the second design maximizes the illumination power of the imaging area. Numerical results demonstrate the effectiveness of the proposed SBL-based method and the superiority of the illumination designs.

cs.IT↗

arXiv2025

Self-Improvement in Multimodal Large Language Models: A Survey

Recent advancements in self-improvement for Large Language Models (LLMs) have efficiently enhanced model capabilities without significantly increasing costs, particularly in terms of human effort. While this area is still relatively young, its extension to the multimodal domain holds immense potential for leveraging diverse data sources and developing more general self-improving models. This survey is the first to provide a comprehensive overview of self-improvement in Multimodal LLMs (MLLMs). We provide a structured overview of the current literature and discuss methods from three perspectives: 1) data collection, 2) data organization, and 3) model optimization, to facilitate the further development of self-improvement in MLLMs. We also include commonly used evaluations and downstream applications. Finally, we conclude by outlining open challenges and future research directions.

cs.CL↗

arXiv2025

Near-Field Integrated Imaging and Communication in Distributed MIMO Networks

In this work, we propose a general framework for wireless imaging in distributed MIMO wideband communication systems, considering multi-view non-isotropic targets and near-field propagation effects. For indoor scenarios where the objective is to image small-scale objects with high resolution, we propose a range migration algorithm (RMA)-based scheme using three kinds of array architectures: the full array, boundary array, and distributed boundary array. With non-isotropic near-field channels, we establish the Fourier transformation (FT)-based relationship between the imaging reflectivity and the distributed spatial-domain signals and discuss the corresponding theoretical properties. Next, for outdoor scenarios where the objective is to reconstruct the large-scale three-dimensional (3D) environment with coarse resolution, we propose a sparse Bayesian learning (SBL)-based algorithm to solve the multiple measurement vector (MMV) problem, which further addresses the non-isotropic reflectivity across different subcarriers. Numerical results demonstrate the effectiveness of the proposed algorithms in acquiring high-resolution small objects and accurately reconstructing large-scale environments.

eess.SP↗

arXiv2025

Charge density wave with suppressed long-range structural modulation in canted antiferromagnetic kagome FeGe

Kagome lattice can host abundant exotic quantum states such as superconductivity and charge density wave (CDW). Recently, successive orders of A-type antiferromagnetism (AFM), CDW and canted AFM have been manifested upon cooling in kagome FeGe. However, the mechanism of CDW and interaction with magnetism remains unclear. Here we investigate the evolution of CDW with temperature across the canted AFM by single-crystal x-ray diffraction, scanning tunneling microscope (STM) and resonant elastic x-ray scattering (REXS). Interestingly, CDW-induced superlattice reflections become weak after the canted AFM, although long-range CDW order is still detectable by STM and REXS. We uncover a novel long-range CDW order with suppressed structural modulation, likely due to the competition for the underlying crystal structure between CDW and canted AFM. Additionally, occupational modulations of Ge1 in the kagome plane and displacive modulations of all atoms were extracted. The results confirm Ge dimerization along the c axis and suggest a dynamic transformation between different CDW domains.

cond-mat.str-el↗

arXiv2025

Sensitive infrared surface photovoltage in quasi-equilibrium in a layered semiconductor at low-intensity low-temperature condition

Benefit to layer-dependent bandgap, van der Waals materials with surface photovoltaic effect (SPV) enable photodetection over a tunable wavelength range with low power consumption. However, sensitive SPV in the infrared region, especially in a quasi-steady illumination condition, is still elusive in layered semiconductors. Here, using angle-resolved photoemission spectroscopy, we report a sensitive SPV in quasi-equilibrium in NbSi0.5Te2, with photoresponsivity up to 2.4*10^6 V/(W*cm^(-2)) at low intensity low temperature condition (LILT). The sensitive SPV is further confirmed by observing the Dember effect, where the photogenerated carrier density is high enough and diffusion currents suppress SPV. Temperature-dependent measurements indicate that intrinsic carriers freezing at low temperature leads to the ultrahigh photoresponse, while a small amount of photon-generated carriers in quasi-equilibrium dominate the system. Our work not only provides a promising layered semiconductor for Infrared optoelectronic devices with strong infrared SPV at LILT, which has application potential in fields such as quantum information and deep-space exploration, but also paves a novel way to enhance light-matter interaction effect by freezing bulk carriers.

cond-mat.mtrl-sci↗

arXiv2025

Robust Utility-Preserving Text Anonymization Based on Large Language Models

Anonymizing text that contains sensitive information is crucial for a wide range of applications. Existing techniques face the emerging challenges of the re-identification ability of large language models (LLMs), which have shown advanced capability in memorizing detailed information and reasoning over dispersed pieces of patterns to draw conclusions. When defending against LLM-based re-identification, anonymization could jeopardize the utility of the resulting anonymized data in downstream tasks. In general, the interaction between anonymization and data utility requires a deeper understanding within the context of LLMs. In this paper, we propose a framework composed of three key LLM-based components: a privacy evaluator, a utility evaluator, and an optimization component, which work collaboratively to perform anonymization. Extensive experiments demonstrate that the proposed model outperforms existing baselines, showing robustness in reducing the risk of re-identification while preserving greater data utility in downstream tasks. We provide detailed studies on these core modules. To consider large-scale and real-time applications, we investigate the distillation of the anonymization capabilities into lightweight models. All of our code and datasets will be made publicly available at https://github.com/UKPLab/acl2025-rupta.

cs.CL↗

arXiv2025

Downlink CSIT under Compressed Feedback: Joint vs. Separate Source-Channel Coding

The acquisition of Downlink (DL) channel state information at the transmitter (CSIT) is known to be a challenging task in multiuser massive MIMO systems when uplink/downlink channel reciprocity does not hold (e.g., in frequency division duplexing systems). From a coding viewpoint, the DL channel state acquired at the users via DL training can be seen as an information source that must be conveyed to the base station via the UL communication channels. The transmission of a source through a channel can be accomplished either by separate or joint source-channel coding (SSCC or JSCC). In this work, using classical remote distortion-rate (DR) theory, we first provide a theoretical lower bound on the channel estimation mean-square-error (MSE) of both JSCC and SSCC-based feedback schemes, which however requires encoding of large blocks of successive channel states and thus cannot be used in practicesince it would incur in an extremely large feedback delay. We then focus on the relevant case of minimal (one slot) feedback delay and propose a practical JSCC-based feedback scheme that fully exploits the channel second-order statistics to optimize the dimension projection in the eigenspace. We analyze the large SNR behavior of the proposed JSCC-based scheme in terms of the quality scaling exponent (QSE). Given the second-order statistics of channel estimation of any feedback scheme, we further derive the closed-form of the lower bound to the ergodic sum-rate for DL data transmission under maximum ratio transmission and zero-forcing precoding. Via extensive numerical results, we show that our proposed JSCC-based scheme outperforms known JSCC, SSCC baseline and deep learning-based schemes and is able to approach the performance of the optimal DR scheme in the range of practical SNR.

cs.IT↗

arXiv2025

CLIPErase: Efficient Unlearning of Visual-Textual Associations in CLIP

Machine unlearning (MU) has gained significant attention as a means to remove specific data from trained models without requiring a full retraining process. While progress has been made in unimodal domains like text and image classification, unlearning in multimodal models remains relatively underexplored. In this work, we address the unique challenges of unlearning in CLIP, a prominent multimodal model that aligns visual and textual representations. We introduce CLIPErase, a novel approach that disentangles and selectively forgets both visual and textual associations, ensuring that unlearning does not compromise model performance. CLIPErase consists of three key modules: a Forgetting Module that disrupts the associations in the forget set, a Retention Module that preserves performance on the retain set, and a Consistency Module that maintains consistency with the original model. Extensive experiments on the CIFAR-100 and Flickr30K datasets across four CLIP downstream tasks demonstrate that CLIPErase effectively forgets designated associations in zero-shot tasks for multimodal samples, while preserving the model's performance on the retain set after unlearning.

cs.CV↗

arXiv2025

ReEx-SQL: Reasoning with Execution-Aware Reinforcement Learning for Text-to-SQL

In Text-to-SQL, execution feedback is essential for guiding large language models (LLMs) to reason accurately and generate reliable SQL queries. However, existing methods treat execution feedback solely as a post-hoc signal for correction or selection, failing to integrate it into the generation process. This limitation hinders their ability to address reasoning errors as they occur, ultimately reducing query accuracy and robustness. To address this issue, we propose ReEx-SQL (Reasoning with Execution-Aware Reinforcement Learning), a framework for Text-to-SQL that enables models to interact with the database during decoding and dynamically adjust their reasoning based on execution feedback. ReEx-SQL introduces an execution-aware reasoning paradigm that interleaves intermediate SQL execution into reasoning paths, facilitating context-sensitive revisions. It achieves this through structured prompts with markup tags and a stepwise rollout strategy that integrates execution feedback into each stage of generation. To supervise policy learning, we develop a composite reward function that includes an exploration reward, explicitly encouraging effective database interaction. Additionally, ReEx-SQL adopts a tree-based decoding strategy to support exploratory reasoning, enabling dynamic expansion of alternative reasoning paths. Notably, ReEx-SQL achieves 88.8% on Spider and 64.9% on BIRD at the 7B scale, surpassing the standard reasoning baseline by 2.7% and 2.6%, respectively. It also shows robustness, achieving 85.2% on Spider-Realistic with leading performance. In addition, its tree-structured decoding improves efficiency and performance over linear decoding, reducing inference time by 51.9% on the BIRD development set.

cs.CL↗

arXiv2025

DropMAE: Learning Representations via Masked Autoencoders with Spatial-Attention Dropout for Temporal Matching Tasks

This paper studies masked autoencoder (MAE) video pre-training for various temporal matching-based downstream tasks, i.e., object-level tracking tasks including video object tracking (VOT) and video object segmentation (VOS), self-supervised visual correspondence learning, dense tracking tasks including optical flow estimation and long-term point tracking, and 3D point cloud tracking. Specifically, our work explores to provide a general representation to boost the temporal matching ability in various downstream tracking tasks. To achieve this, we firstly find that a simple extension of MAE, which randomly masks out frame patches in videos and reconstruct the frame pixels, heavily relies on spatial cues while ignoring temporal relations for frame reconstruction, thus leading to sub-optimal temporal matching representations. To alleviate this, we propose DropMAE, which adaptively performs spatial-attention dropout in the frame reconstruction to facilitate temporal correspondence learning in videos. We obtain several important findings with DropMAE: 1) DropMAE is a strong and efficient temporal matching learner, which achieves better fine-tuning results on matching-based tasks than the ImageNet-based MAE with 2x faster pre-training speed. 2) DropMAE is effective for different tracking tasks, i.e., object-level matching tasks including VOT and VOS, dense tracking tasks including optical flow estimation and tracking any point (TAP), and even 3D tracking in the different modality of point cloud data. Since none exists, we build ViT-based trackers for different downstream tracking tasks, and our pre-trained DropMAE model can be directly loaded in these ViT-based trackers for fine-tuning without further modifications. Experiments on 6 downstream tracking tasks demonstrate the effectiveness of DropMAE as a general pre-trained representation for diverse tracking tasks.

cs.CV↗

arXiv2025

Holographic MIMO Multi-Cell Communications

Metamaterial antennas are appealing for next-generation wireless networks due to their simplified hardware and much-reduced size, power, and cost. This paper investigates the holographic multiple-input multiple-output (HMIMO)-aided multi-cell systems with practical per-radio frequency (RF) chain power constraints. With multiple antennas at both base stations (BSs) and users, we design the baseband digital precoder and the tuning response of HMIMO metamaterial elements to maximize the weighted sum user rate. Specifically, under the framework of block coordinate descent (BCD) and weighted minimum mean square error (WMMSE) techniques, we derive the low-complexity closed-form solution for baseband precoder without requiring bisection search and matrix inversion. Then, for the design of HMIMO metamaterial elements under binary tuning constraints, we first propose a low-complexity suboptimal algorithm with closed-form solutions by exploiting the hidden convexity (HC) in the quadratic problem and then further propose an accelerated sphere decoding (SD)-based algorithm which yields global optimal solution in the iteration. For HMIMO metamaterial element design under the Lorentzian-constrained phase model, we propose a maximization-minorization (MM) algorithm with closed-form solutions at each iteration step. Furthermore, in a simplified multiple-input single-output (MISO) scenario, we derive the scaling law of downlink single-to-noise (SNR) for HMIMO with binary and Lorentzian tuning constraints and theoretically compare it with conventional fully digital/hybrid arrays. Simulation results demonstrate the effectiveness of our algorithms compared to benchmarks and the benefits of HMIMO compared to conventional arrays.

eess.SP↗

arXiv2025

Beyond Single-Value Metrics: Evaluating and Enhancing LLM Unlearning with Cognitive Diagnosis

Due to the widespread use of LLMs and the rising critical ethical and safety concerns, LLM unlearning methods have been developed to remove harmful knowledge and undesirable capabilities. In this context, evaluations are mostly based on single-value metrics such as QA accuracy. However, these metrics often fail to capture the nuanced retention of harmful knowledge components, making it difficult to assess the true effectiveness of unlearning. To address this issue, we propose UNCD (UNlearning evaluation via Cognitive Diagnosis), a novel framework that leverages Cognitive Diagnosis Modeling for fine-grained evaluation of LLM unlearning. Our dedicated benchmark, UNCD-Cyber, provides a detailed assessment of the removal of dangerous capabilities. Moreover, we introduce UNCD-Agent, which refines unlearning by diagnosing knowledge remnants and generating targeted unlearning data. Extensive experiments across eight unlearning methods and two base models demonstrate that UNCD not only enhances evaluation but also effectively facilitates the removal of harmful LLM abilities.

cs.LG↗

arXiv2025

Optimizing Feature Selection in Causal Inference: A Three-Stage Computational Framework for Unbiased Estimation

Feature selection is an important but challenging task in causal inference for obtaining unbiased estimates of causal quantities. Properly selected features in causal inference not only significantly reduce the time required to implement a matching algorithm but, more importantly, can also reduce the bias and variance when estimating causal quantities. When feature selection techniques are applied in causal inference, the crucial criterion is to select variables that, when used for matching, can achieve an unbiased and robust estimation of causal quantities. Recent research suggests that balancing only on treatment-associated variables introduces bias while balancing on spurious variables increases variance. To address this issue, we propose an enhanced three-stage framework that shows a significant improvement in selecting the desired subset of variables compared to the existing state-of-the-art feature selection framework for causal inference, resulting in lower bias and variance in estimating the causal quantity. We evaluated our proposed framework using a state-of-the-art synthetic data across various settings and observed superior performance within a feasible computation time, ensuring scalability for large-scale datasets. Finally, to demonstrate the applicability of our proposed methodology using large-scale real-world data, we evaluated an important US healthcare policy related to the opioid epidemic crisis: whether opioid use disorder has a causal relationship with suicidal behavior.

stat.ME↗

arXiv2024

SaSR-Net: Source-Aware Semantic Representation Network for Enhancing Audio-Visual Question Answering

Audio-Visual Question Answering (AVQA) is a challenging task that involves answering questions based on both auditory and visual information in videos. A significant challenge is interpreting complex multi-modal scenes, which include both visual objects and sound sources, and connecting them to the given question. In this paper, we introduce the Source-aware Semantic Representation Network (SaSR-Net), a novel model designed for AVQA. SaSR-Net utilizes source-wise learnable tokens to efficiently capture and align audio-visual elements with the corresponding question. It streamlines the fusion of audio and visual information using spatial and temporal attention mechanisms to identify answers in multi-modal scenes. Extensive experiments on the Music-AVQA and AVQA-Yang datasets show that SaSR-Net outperforms state-of-the-art AVQA methods.

cs.CV↗

arXiv2024

Cooperative Multistatic Target Detection in Cell-Free Communication Networks

In this work, we consider the target detection problem in a multistatic integrated sensing and communication (ISAC) scenario characterized by the cell-free MIMO communication network deployment, where multiple radio units (RUs) in the network cooperate with each other for the sensing task. By exploiting the angle resolution from multiple arrays deployed in the network and the delay resolution from the communication signals, i.e., orthogonal frequency division multiplexing (OFDM) signals, we formulate a cooperative sensing problem with coherent data fusion of multiple RUs' observations and propose a sparse Bayesian learning (SBL)-based method, where the global coordinates of target locations are directly detected. Intensive numerical results indicate promising target detection performance of the proposed SBL-based method. Additionally, a theoretical analysis of the considered cooperative multistatic sensing task is provided using the pairwise error probability (PEP) analysis, which can be used to provide design insights, e.g., illumination and beam patterns, for the considered problem.

cs.IT↗