arXiv subjects

Yixuan Li

Publications and source records attributed to Yixuan Li.

At least 127 records · Page 7Linked to original sources

arXiv2025

How Contaminated Is Your Benchmark? Quantifying Dataset Leakage in Large Language Models with Kernel Divergence

Dataset contamination, where evaluation datasets overlap with pre-training corpora, inflates performance metrics and undermines the reliability of model evaluations. Measuring dataset contamination thus becomes essential to ensure that performance evaluations genuinely reflect a model's ability to generalize to unseen data, rather than relying on memorized examples. To address this problem, we propose Kernel Divergence Score (KDS), a novel method that evaluates dataset contamination by computing the divergence between the kernel similarity matrix of sample embeddings, before and after fine-tuning on the benchmark dataset. Leveraging the insight that fine-tuning affects unseen samples more significantly than seen ones, KDS provides a reliable measure of contamination. Through extensive experiments on controlled contamination scenarios, KDS demonstrates a near-perfect correlation with contamination levels and outperforms existing baselines. Additionally, we perform comprehensive ablation studies to analyze the impact of key design choices, providing deeper insights into the components and effectiveness of KDS. These ablations highlight the importance of leveraging fine-grained kernel-based information and confirm the reliability of the proposed framework across diverse datasets and settings. Code is released in https://github.com/deeplearning-wisc/kernel-divergence-score.

cs.LG↗

arXiv2025

From Perception to Decision: Assessing the Role of Chart Types Affordances in High-Level Decision Tasks

Visualization design influences how people perceive data patterns, yet most research focuses on low-level analytic tasks, such as finding correlations. The extent to which these perceptual affordances translate to high-level decision-making in the real world remains underexplored. Through a case study of academic mentorship selection using bar charts and pie charts, we investigated whether chart types differentially influence how students evaluate faculty research profiles. Our crowdsourced experiment revealed only minimal differences in decision outcomes between chart types, suggesting that perceptual affordances established in controlled analytical tasks may not directly translate to high-level decision scenarios. These findings emphasize the importance of evaluating visualizations within real-world contexts and highlight the need to distinguish between perceptual and decision affordances when developing visualization guidelines.

cs.HC↗

arXiv2025

AutoDroid-V2: Boosting SLM-based GUI Agents via Code Generation

Large language models (LLMs) have brought exciting new advances to mobile UI agents, a long-standing research field that aims to complete arbitrary natural language tasks through mobile UI interactions. However, existing UI agents usually demand powerful large language models that are difficult to be deployed locally on end-users' devices, raising huge concerns about user privacy and centralized serving cost. Inspired by the remarkable coding abilities of recent small language models (SLMs), we propose to convert the UI task automation problem to a code generation problem, which can be effectively solved by an on-device SLM and efficiently executed with an on-device code interpreter. Unlike normal coding tasks that can be extensively pre-trained with public datasets, generating UI automation code is challenging due to the diversity, complexity, and variability of target apps. Therefore, we adopt a document-centered approach that automatically builds fine-grained API documentation for each app and generates diverse task samples based on this documentation. By guiding the agent with the synthetic documents and task samples, it learns to generate precise and efficient scripts to complete unseen tasks. Based on detailed comparisons with state-of-the-art mobile UI agents, our approach effectively improves the mobile task automation with significantly higher success rates and lower latency/token consumption. Code is open-sourced at https://github.com/MobileLLM/AutoDroid-V2.

cs.AI↗

arXiv2025

Exploration on Real World Assets and Tokenization

This study delves into the tokenization of real-world assets (RWAs) on the blockchain with the objective of augmenting liquidity and refining asset management practices. By conducting an exhaustive analysis of the technical procedures implicated and scrutinizing case studies of existing deployments, this research evaluates the advantages, hurdles, and prospective advancements of blockchain technology in reshaping conventional asset management paradigms.

cs.CR↗

arXiv2025

Challenges and Future Directions of Data-Centric AI Alignment

As AI systems become increasingly capable and influential, ensuring their alignment with human values, preferences, and goals has become a critical research focus. Current alignment methods primarily focus on designing algorithms and loss functions but often underestimate the crucial role of data. This paper advocates for a shift towards data-centric AI alignment, emphasizing the need to enhance the quality and representativeness of data used in aligning AI systems. In this position paper, we highlight key challenges associated with both human-based and AI-based feedback within the data-centric alignment framework. Through qualitative analysis, we identify multiple sources of unreliability in human feedback, as well as problems related to temporal drift, context dependence, and AI-based feedback failing to capture human values due to inherent model limitations. We propose future research directions, including improved feedback collection practices, robust data-cleaning methodologies, and rigorous feedback verification processes. We call for future research into these critical directions to ensure, addressing gaps that persist in understanding and improving data-centric alignment practices.

cs.CL↗

arXiv2025

Guided Tensor Lifting

Domain-specific languages (DSLs) for machine learning are revolutionizing the speed and efficiency of machine learning workloads as they enable users easy access to high-performance compiler optimizations and accelerators. However, to take advantage of these capabilities, a user must first translate their legacy code from the language it is currently written in, into the new DSL. The process of automatically lifting code into these DSLs has been identified by several recent works, which propose program synthesis as a solution. However, synthesis is expensive and struggles to scale without carefully designed and hard-wired heuristics. In this paper, we present an approach for lifting that combines an enumerative synthesis approach with a Large Language Model used to automatically learn the domain-specific heuristics for program lifting, in the form of a probabilistic grammar. Our approach outperforms the state-of-the-art tools in this area, despite only using learned heuristics.

cs.SE↗

arXiv2025

Pinching-Antenna Assisted Simultaneous Wireless Information and Power Transfer

This letter introduces a novel pinching-antenna-system (PASS) assisted simultaneous wireless information and power transfer (SWIPT), where multiple pinching antennas (PAs) are strategically activiated on a waveguide to facilitate information transmission to multiple information receivers (IRs) and power transfer to multiple energy receivers (ERs) simultaneously. Leveraging the single-waveguide architecture, non-orthogonal multiple access is employed to enable the superposed transmission of information signals, eliminating the need for dedicated energy carriers by concurrently serving both the IRs and the ERs. In this letter, an optimization problem is first formulated with an objective is to maximize the sum power received at the ERs via joint optimizing the IRs power allocation and the PAs positioning. Given the challenging non-convex nature of the formulated optimization problem, it is decoupled into two sub-problems which are solved alternatively. Specifically, the power allocation subproblem is recast a convex optimization problem, and hence can be solved efficiently. Furthermore, we propose a high-precision element-wise algorithm and a low-complexity linearly decreasing weight particle swarm optimization algorithm to solve the position optimization sub-problem. The numerical results demonstrate that PAs assisted SWIPT can achieve a remarkable performance gain compared to conventional system.

eess.SP↗

arXiv2025

Your Weak LLM is Secretly a Strong Teacher for Alignment

The burgeoning capabilities of large language models (LLMs) have underscored the need for alignment to ensure these models act in accordance with human values and intentions. Existing alignment frameworks present constraints either in the form of expensive human effort or high computational costs. This paper explores a promising middle ground, where we employ a weak LLM that is significantly less resource-intensive than top-tier models, yet offers more automation than purely human feedback. We present a systematic study to evaluate and understand weak LLM's ability to generate feedback for alignment. Our empirical findings demonstrate that weak LLMs can provide feedback that rivals or even exceeds that of fully human-annotated data. Our study indicates a minimized impact of model size on feedback efficacy, shedding light on a scalable and sustainable alignment strategy. To deepen our understanding of alignment under weak LLM feedback, we conduct a series of qualitative and quantitative analyses, offering novel insights into the quality discrepancies between human feedback vs. weak LLM feedback.

cs.CL↗

arXiv2025

II-NVM: Enhancing Map Accuracy and Consistency with Normal Vector-Assisted Mapping

SLAM technology plays a crucial role in indoor mapping and localization. A common challenge in indoor environments is the "double-sided mapping issue", where closely positioned walls, doors, and other surfaces are mistakenly identified as a single plane, significantly hindering map accuracy and consistency. To address this issue this paper introduces a SLAM approach that ensures accurate mapping using normal vector consistency. We enhance the voxel map structure to store both point cloud data and normal vector information, enabling the system to evaluate consistency during nearest neighbor searches and map updates. This process distinguishes between the front and back sides of surfaces, preventing incorrect point-to-plane constraints. Moreover, we implement an adaptive radius KD-tree search method that dynamically adjusts the search radius based on the local density of the point cloud, thereby enhancing the accuracy of normal vector calculations. To further improve realtime performance and storage efficiency, we incorporate a Least Recently Used (LRU) cache strategy, which facilitates efficient incremental updates of the voxel map. The code is released as open-source and validated in both simulated environments and real indoor scenarios. Experimental results demonstrate that this approach effectively resolves the "double-sided mapping issue" and significantly improves mapping precision. Additionally, we have developed and open-sourced the first simulation and real world dataset specifically tailored for the "double-sided mapping issue".

cs.RO↗

arXiv2025

Dynamic Initialization for LiDAR-inertial SLAM

The accuracy of the initial state, including initial velocity, gravity direction, and IMU biases, is critical for the initialization of LiDAR-inertial SLAM systems. Inaccurate initial values can reduce initialization speed or lead to failure. When the system faces urgent tasks, robust and fast initialization is required while the robot is moving, such as during the swift assessment of rescue environments after natural disasters, bomb disposal, and restarting LiDAR-inertial SLAM in rescue missions. However, existing initialization methods usually require the platform to remain stationary, which is ineffective when the robot is in motion. To address this issue, this paper introduces a robust and fast dynamic initialization method for LiDAR-inertial systems (D-LI-Init). This method iteratively aligns LiDAR-based odometry with IMU measurements to achieve system initialization. To enhance the reliability of the LiDAR odometry module, the LiDAR and gyroscope are tightly integrated within the ESIKF framework. The gyroscope compensates for rotational distortion in the point cloud. Translational distortion compensation occurs during the iterative update phase, resulting in the output of LiDAR-gyroscope odometry. The proposed method can initialize the system no matter the robot is moving or stationary. Experiments on public datasets and real-world environments demonstrate that the D-LI-Init algorithm can effectively serve various platforms, including vehicles, handheld devices, and UAVs. D-LI-Init completes dynamic initialization regardless of specific motion patterns. To benefit the research community, we have open-sourced our code and test datasets on GitHub.

cs.RO↗

arXiv2025

Enabling Auditory Large Language Models for Automatic Speech Quality Evaluation

Speech quality assessment typically requires evaluating audio from multiple aspects, such as mean opinion score (MOS) and speaker similarity (SIM) \etc., which can be challenging to cover using one small model designed for a single task. In this paper, we propose leveraging recently introduced auditory large language models (LLMs) for automatic speech quality assessment. By employing task-specific prompts, auditory LLMs are finetuned to predict MOS, SIM and A/B testing results, which are commonly used for evaluating text-to-speech systems. Additionally, the finetuned auditory LLM is able to generate natural language descriptions assessing aspects like noisiness, distortion, discontinuity, and overall quality, providing more interpretable outputs. Extensive experiments have been performed on the NISQA, BVCC, SOMOS and VoxSim speech quality datasets, using open-source auditory LLMs such as SALMONN, Qwen-Audio, and Qwen2-Audio. For the natural language descriptions task, a commercial model Google Gemini 1.5 Pro is also evaluated. The results demonstrate that auditory LLMs achieve competitive performance compared to state-of-the-art task-specific small models in predicting MOS and SIM, while also delivering promising results in A/B testing and natural language descriptions. Our data processing scripts and finetuned model checkpoints can be found at https://github.com/bytedance/SALMONN.

eess.AS↗

arXiv2025

Classical Black Hole Probes of UV Scales

In the context of the Swampland program, black hole attractors have been employed to probe infinite distances in moduli space, where the EFT cutoff goes to zero in Planck units and UV effects become significant. In this paper, we take the perspective of the two-derivative action of string theoretic effective field theories and explore various families of extremal black hole solutions that probe infinite distance limits at their horizons. While these solutions do not include higher-order corrections in the EFT expansion, we find that, in many cases, the smallest BPS black holes in these families remarkably reproduce either the species scale or some other Kaluza-Klein scale. In highly supersymmetric cases, this match with UV scales even persists in the interior of moduli space. We even find that non-BPS black holes solutions in circle compactification of Type II string theories follow the species scale in decompactification limits. These observations suggests that the two-derivative action may encode information about relevant UV scales. We discuss the interplay of these results with emergence and UV/IR mixing in quantum gravity.

hep-th↗

arXiv2025

PostEdit: Posterior Sampling for Efficient Zero-Shot Image Editing

In the field of image editing, three core challenges persist: controllability, background preservation, and efficiency. Inversion-based methods rely on time-consuming optimization to preserve the features of the initial images, which results in low efficiency due to the requirement for extensive network inference. Conversely, inversion-free methods lack theoretical support for background similarity, as they circumvent the issue of maintaining initial features to achieve efficiency. As a consequence, none of these methods can achieve both high efficiency and background consistency. To tackle the challenges and the aforementioned disadvantages, we introduce PostEdit, a method that incorporates a posterior scheme to govern the diffusion sampling process. Specifically, a corresponding measurement term related to both the initial features and Langevin dynamics is introduced to optimize the estimated image generated by the given target prompt. Extensive experimental results indicate that the proposed PostEdit achieves state-of-the-art editing performance while accurately preserving unedited regions. Furthermore, the method is both inversion- and training-free, necessitating approximately 1.5 seconds and 18 GB of GPU memory to generate high-quality results.

cs.CV↗

arXiv2025

How Useful is Continued Pre-Training for Generative Unsupervised Domain Adaptation?

Recent breakthroughs in scale have enabled the emergence of powerful generative language models, and the ability to fine-tune these models on various tasks by casting them into prompts or instructions. In this landscape, the problem of Unsupervised Domain Adaptation (UDA), or the problem of leveraging knowledge from a labeled source domain to an unlabeled target domain, has been left behind, with recent UDA methods still addressing discriminative classification. In particular, two popular UDA approaches, involving Continued Pre-Training (CPT) and learning domain invariant representations, have been under-explored in the generative setting, signaling a gap. In this work, we evaluate the utility of CPT for generative UDA. We first perform an empirical evaluation to measure the trade-offs between CPT and strong methods promoting domain invariance. We further evaluate how well the benefits of CPT extend to different architectures, tuning methods and data regimes. We then motivate the use of CPT by studying to what degree it benefits classification performance on the target domain. Finally, we attempt to understand the mechanism behind which CPT improves classification performance on the unlabeled target domain. Our findings suggest that a implicitly learns the downstream task while predicting masked words informative to that task. Our work connects the body of UDA research with that of instruction tuning, enabling an initial step towards a wider applicability of modern language models.

cs.CL↗

arXiv2025

Characteristics Analysis of Autonomous Vehicle Pre-crash Scenarios

To date, hundreds of crashes have occurred in open road testing of automated vehicles (AVs), highlighting the need for improving AV reliability and safety. Pre-crash scenario typology classifies crashes based on vehicle dynamics and kinematics features. Building on this, characteristics analysis can identify similar features under comparable crashes, offering a more effective reflection of general crash patterns and providing more targeted recommendations for enhancing AV performance. However, current studies primarily concentrated on crashes among conventional human-driven vehicles, leaving a gap in research dedicated to in-depth AV crash analyses. In this paper, we analyzed the latest California AV collision reports and used the newly revised pre-crash scenario typology to identify pre-crash scenarios. We proposed a set of mapping rules for automatically extracting these AV pre-crash scenarios, successfully identifying 24 types with a 98.1% accuracy rate, and obtaining two key scenarios of AV crashes (i.e., rear-end scenarios and intersection scenarios) through detailed analysis. Association analyses of rear-end scenarios showed that the significant environmental influencing factors were traffic control type, location type, light, etc. For intersection scenarios prone to severe crashes with detailed descriptions, we employed causal analyses to obtain the significant causal factors: habitual violations and expectations of certain behavior. Optimization recommendations were then formulated, addressing both governmental oversight and AV manufacturers' potential improvements. The findings of this paper could guide government authorities to develop related regulations, help manufacturers design AV test scenarios, and identify potential shortcomings in control algorithms specific to various real-world scenarios, thereby optimizing AV systems effectively.

cs.RO↗

arXiv2025

LayerPano3D: Layered 3D Panorama for Hyper-Immersive Scene Generation

3D immersive scene generation is a challenging yet critical task in computer vision and graphics. A desired virtual 3D scene should 1) exhibit omnidirectional view consistency, and 2) allow for free exploration in complex scene hierarchies. Existing methods either rely on successive scene expansion via inpainting or employ panorama representation to represent large FOV scene environments. However, the generated scene suffers from semantic drift during expansion and is unable to handle occlusion among scene hierarchies. To tackle these challenges, we introduce Layerpano3D, a novel framework for full-view, explorable panoramic 3D scene generation from a single text prompt. Our key insight is to decompose a reference 2D panorama into multiple layers at different depth levels, where each layer reveals the unseen space from the reference views via diffusion prior. Layerpano3D comprises multiple dedicated designs: 1) We introduce a new panorama dataset Upright360, comprising 9k high-quality and upright panorama images, and finetune the advanced Flux model on Upright360 for high-quality, upright and consistent panorama generation. 2) We pioneer the Layered 3D Panorama as underlying representation to manage complex scene hierarchies and lift it into 3D Gaussians to splat detailed 360-degree omnidirectional scenes with unconstrained viewing paths. Extensive experiments demonstrate that our framework generates state-of-the-art 3D panoramic scene in both full view consistency and immersive exploratory experience. We believe that Layerpano3D holds promise for advancing 3D panoramic scene creation with numerous applications.

cs.CV↗

arXiv2025

video-SALMONN-o1: Reasoning-enhanced Audio-visual Large Language Model

While recent advancements in reasoning optimization have significantly enhanced the capabilities of large language models (LLMs), existing efforts to improve reasoning have been limited to solving mathematical problems and focusing on visual graphical inputs, neglecting broader applications in general video understanding.This paper proposes video-SALMONN-o1, the first open-source reasoning-enhanced audio-visual LLM designed for general video understanding tasks. To enhance its reasoning abilities, we develop a reasoning-intensive dataset featuring challenging audio-visual questions with step-by-step solutions. We also propose process direct preference optimization (pDPO), which leverages contrastive step selection to achieve efficient step-level reward modelling tailored for multimodal inputs. Additionally, we introduce RivaBench, the first reasoning-intensive video understanding benchmark, featuring over 4,000 high-quality, expert-curated question-answer pairs across scenarios such as standup comedy, academic presentations, and synthetic video detection. video-SALMONN-o1 achieves 3-8% accuracy improvements over the LLaVA-OneVision baseline across different video reasoning benchmarks. Besides, pDPO achieves 6-8% improvements compared to the supervised fine-tuning model on RivaBench. Enhanced reasoning enables video-SALMONN-o1 zero-shot synthetic video detection capabilities.

cs.CV↗

arXiv2025

Process Reward Model with Q-Value Rankings

Process Reward Modeling (PRM) is critical for complex reasoning and decision-making tasks where the accuracy of intermediate steps significantly influences the overall outcome. Existing PRM approaches, primarily framed as classification problems, employ cross-entropy loss to independently evaluate each step's correctness. This method can lead to suboptimal reward distribution and does not adequately address the interdependencies among steps. To address these limitations, we introduce the Process Q-value Model (PQM), a novel framework that redefines PRM in the context of a Markov Decision Process. PQM optimizes Q-value rankings based on a novel comparative loss function, enhancing the model's ability to capture the intricate dynamics among sequential decisions. This approach provides a more granular and theoretically grounded methodology for process rewards. Our extensive empirical evaluations across various sampling policies, language model backbones, and multi-step reasoning benchmarks show that PQM outperforms classification-based PRMs. The effectiveness of the comparative loss function is highlighted in our comprehensive ablation studies, confirming PQM's practical efficacy and theoretical advantage.

cs.CL↗