arXiv subjects

Peng Li

Publications and source records attributed to Peng Li.

At least 145 records · Page 8Linked to original sources

arXiv2025

SyncHuman: Synchronizing 2D and 3D Generative Models for Single-view Human Reconstruction

Photorealistic 3D full-body human reconstruction from a single image is a critical yet challenging task for applications in films and video games due to inherent ambiguities and severe self-occlusions. While recent approaches leverage SMPL estimation and SMPL-conditioned image generative models to hallucinate novel views, they suffer from inaccurate 3D priors estimated from SMPL meshes and have difficulty in handling difficult human poses and reconstructing fine details. In this paper, we propose SyncHuman, a novel framework that combines 2D multiview generative model and 3D native generative model for the first time, enabling high-quality clothed human mesh reconstruction from single-view images even under challenging human poses. Multiview generative model excels at capturing fine 2D details but struggles with structural consistency, whereas 3D native generative model generates coarse yet structurally consistent 3D shapes. By integrating the complementary strengths of these two approaches, we develop a more effective generation framework. Specifically, we first jointly fine-tune the multiview generative model and the 3D native generative model with proposed pixel-aligned 2D-3D synchronization attention to produce geometrically aligned 3D shapes and 2D multiview images. To further improve details, we introduce a feature injection mechanism that lifts fine details from 2D multiview images onto the aligned 3D shapes, enabling accurate and high-fidelity reconstruction. Extensive experiments demonstrate that SyncHuman achieves robust and photo-realistic 3D human reconstruction, even for images with challenging poses. Our method outperforms baseline methods in geometric accuracy and visual fidelity, demonstrating a promising direction for future 3D generation models.

cs.CV↗

arXiv2025

Enhancing Large Language Model Reasoning via Selective Critical Token Fine-Tuning

Large language models (LLMs) primarily rely on supervised fine-tuning (SFT) as a key method to adapt pre-trained models to domain-specific tasks such as mathematical reasoning. However, standard SFT uniformly penalizes all tokens, neglecting that only a small subset of critical tokens determines reasoning correctness. This uniform supervision often causes reduced output diversity and limited generalization. We propose Critical Token Fine-tuning (CFT), a simple yet effective approach that updates only tokens identified as functionally indispensable via counterfactual perturbations. By focusing gradient signals on these decisive reasoning steps while preserving the diversity of non-critical tokens, CFT can enhance both generation and diversity. Extensive experiments on five models across three families (Qwen, OLMo, LLaMA) and eleven mathematical reasoning benchmarks show that CFT, despite fine-tuning on less than 12% of tokens, consistently outperforms standard SFT. Moreover, CFT enables test-time scaling through improved sampling diversity and provides a stronger initialization for reinforcement learning, sustaining performance gains in later training stages while maintaining higher entropy for better exploration. These results highlight CFT as a practical and general framework for efficient and robust LLM fine-tuning.

cs.CL↗

arXiv2025

Federated Fine-Tuning of Sparsely-Activated Large Language Models on Resource-Constrained Devices

Federated fine-tuning of Mixture-of-Experts (MoE)-based large language models (LLMs) is challenging due to their massive computational requirements and the resource constraints of participants. Existing working attempts to fill this gap through model quantization, computation offloading, or expert pruning. However, they cannot achieve desired performance due to impractical system assumptions and a lack of consideration for MoE-specific characteristics. In this paper, we propose FLUX, a system designed to enable federated fine-tuning of MoE-based LLMs across participants with constrained computing resources (e.g., consumer-grade GPUs), aiming to minimize time-to-accuracy. FLUX introduces three key innovations: (1) quantization-based local profiling to estimate expert activation with minimal overhead, (2) adaptive layer-aware expert merging to reduce resource consumption while preserving accuracy, and (3) dynamic expert role assignment using an exploration-exploitation strategy to balance tuning and non-tuning experts. Extensive experiments on LLaMA-MoE and DeepSeek-MoE with multiple benchmark datasets demonstrate that FLUX significantly outperforms existing methods, achieving up to 4.75X speedup in time-to-accuracy.

cs.DC↗

arXiv2025

Unveiling Over-Memorization in Finetuning LLMs for Reasoning Tasks

The pretrained large language models (LLMs) are finetuned with labeled data for better instruction following ability and alignment with human values. In this paper, we study the learning dynamics of LLM finetuning on reasoning tasks and reveal the uncovered over-memorization phenomenon during a specific stage of LLM finetuning. At this stage, the LLMs have excessively memorized training data and exhibit high test perplexity while maintaining good test accuracy. We explore the conditions that contribute to over-memorization and discover that this issue is prevalent across various tasks, models, and fine-tuning methods, with prolonged training and large learning rates exacerbating the problem. Although models with over-memorization demonstrate comparable test accuracy to normal models, they suffer from reduced robustness, poor out-of-distribution generalization, and decreased generation diversity. In light of our findings on over-memorization, we offer recommendations for checkpoint selection and propose techniques such as checkpoint merging and memorization-aware reweighting to mitigate this effect.

cs.CL↗

arXiv2025

MUCAR: Benchmarking Multilingual Cross-Modal Ambiguity Resolution for Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) have demonstrated significant advances across numerous vision-language tasks. MLLMs have shown promising capability in aligning visual and textual modalities, allowing them to process image-text pairs with clear and explicit meanings. However, resolving the inherent ambiguities present in real-world language and visual contexts remains a challenge. Existing multimodal benchmarks typically overlook linguistic and visual ambiguities, relying mainly on unimodal context for disambiguation and thus failing to exploit the mutual clarification potential between modalities. To bridge this gap, we introduce MUCAR, a novel and challenging benchmark designed explicitly for evaluating multimodal ambiguity resolution across multilingual and cross-modal scenarios. MUCAR includes first a multilingual dataset where ambiguous textual expressions are uniquely resolved by corresponding visual contexts, and second a dual-ambiguity dataset that systematically pairs ambiguous images with ambiguous textual contexts, with each combination carefully constructed to yield a single, clear interpretation through mutual disambiguation. Extensive evaluations involving 19 state-of-the-art multimodal models--encompassing both open-source and proprietary architectures--reveal substantial gaps compared to human-level performance, highlighting the need for future research into more sophisticated cross-modal ambiguity comprehension methods, further pushing the boundaries of multimodal reasoning.

cs.CL↗

arXiv2025

Inference-Time Scaling for Generalist Reward Modeling

Reinforcement learning (RL) has been widely adopted in post-training for large language models (LLMs) at scale. Recently, the incentivization of reasoning capabilities in LLMs from RL indicates that $\textit{proper learning methods could enable effective inference-time scalability}$. A key challenge of RL is to obtain accurate reward signals for LLMs in various domains beyond verifiable questions or artificial rules. In this work, we investigate how to improve reward modeling (RM) with more inference compute for general queries, i.e. the $\textbf{inference-time scalability of generalist RM}$, and further, how to improve the effectiveness of performance-compute scaling with proper learning methods. For the RM approach, we adopt pointwise generative reward modeling (GRM) to enable flexibility for different input types and potential for inference-time scaling. For the learning method, we propose Self-Principled Critique Tuning (SPCT) to foster scalable reward generation behaviors in GRMs through online RL, to generate principles adaptively and critiques accurately, resulting in $\textbf{DeepSeek-GRM}$ models. Furthermore, for effective inference-time scaling, we use parallel sampling to expand compute usage, and introduce a meta RM to guide voting process for better scaling performance. Empirically, we show that SPCT significantly improves the quality and scalability of GRMs, outperforming existing methods and models in various RM benchmarks without severe biases, and could achieve better performance compared to training-time scaling. DeepSeek-GRM still meets challenges in some tasks, which we believe can be addressed by future efforts in generalist reward systems. The models are released at Hugging Face and ModelScope.

cs.CL↗

arXiv2025

Jet Reconstruction with Mamba Networks in Collider Events

We introduce a novel end-to-end framework for jet reconstruction in high-energy collider events, leveraging the efficiency and long-range modeling capabilities of the Mamba architecture. Our model unifies instance segmentation, classification, and kinematic regression into a single multi-task learning system, enabling a sophisticated multi-level reconstruction that simultaneously identifies primary heavy jets ($t$, $H$, $W/Z$) and their constituent sub-jets. To facilitate supervised learning for this complex task, we develop a novel method for assigning final-state hadrons to their ancestor colored partons using a Mixed-Integer Linear Programming solver, which generates high-fidelity ground-truth labels. The model achieves high classification accuracy, with an Average Precision score of 0.569 for $W/Z$-jets and 0.568 for $b$-jets, and shows exceptional precision in kinematic reconstruction. Furthermore, we show that the model not only maintains stable performance in high-pileup environments but also successfully reconstructs the mass peaks of beyond the standard model particles. This work presents a powerful and versatile new tool for comprehensive event reconstruction at the LHC.

hep-ph↗

arXiv2025

YuE: Scaling Open Foundation Models for Long-Form Music Generation

We tackle the task of long-form music generation--particularly the challenging \textbf{lyrics-to-song} problem--by introducing YuE, a family of open foundation models based on the LLaMA2 architecture. Specifically, YuE scales to trillions of tokens and generates up to five minutes of music while maintaining lyrical alignment, coherent musical structure, and engaging vocal melodies with appropriate accompaniment. It achieves this through (1) track-decoupled next-token prediction to overcome dense mixture signals, (2) structural progressive conditioning for long-context lyrical alignment, and (3) a multitask, multiphase pre-training recipe to converge and generalize. In addition, we redesign the in-context learning technique for music generation, enabling versatile style transfer (e.g., converting Japanese city pop into an English rap while preserving the original accompaniment) and bidirectional generation. Through extensive evaluation, we demonstrate that YuE matches or even surpasses some of the proprietary systems in musicality and vocal agility. In addition, fine-tuning YuE enables additional controls and enhanced support for tail languages. Furthermore, beyond generation, we show that YuE's learned representations can perform well on music understanding tasks, where the results of YuE match or exceed state-of-the-art methods on the MARBLE benchmark. Keywords: lyrics2song, song generation, long-form, foundation model, music generation

eess.AS↗

arXiv2025

PATIMT-Bench: A Multi-Scenario Benchmark for Position-Aware Text Image Machine Translation in Large Vision-Language Models

Text Image Machine Translation (TIMT) aims to translate texts embedded within an image into another language. Current TIMT studies primarily focus on providing translations for all the text within an image, while neglecting to provide bounding boxes and covering limited scenarios. In this work, we extend traditional TIMT into position-aware TIMT (PATIMT), aiming to support fine-grained and layoutpreserving translation, which holds great practical value but remains largely unexplored. This task comprises two key sub-tasks: regionspecific translation and full-image translation with grounding. To support existing models on PATIMT and conduct fair evaluation, we construct the PATIMT benchmark (PATIMTBench), which consists of 10 diverse real-world scenarios. Specifically, we introduce an Adaptive Image OCR Refinement Pipeline, which adaptively selects appropriate OCR tools based on scenario and refines the results of text-rich images. To ensure evaluation reliability, we further construct a test set, which contains 1,200 high-quality instances manually annotated and reviewed by human experts. After fine-tuning on our data, compact Large Vision-Language Models (LVLMs) achieve state-of-the-art performance on both sub-tasks. Experimental results also highlight the scalability and generalizability of our training data

cs.CV↗

arXiv2025

Bridging Centralized and Distributed Frameworks in Unknown Input Observer Design

State estimation for linear time-invariant systems with unknown inputs is a fundamental problem in various research domains. In this article, we establish conditions for the design of unknown input observers (UIOs) from a geometric approach perspective. Specifically, we derive a necessary and sufficient geometric condition for the existence of a centralized UIO. Compared to existing results, our condition offers a more general design framework, allowing designers the flexibility to estimate partial information of the system state. Furthermore, we extend the centralized UIO design to distributed settings. In contrast to existing distributed UIO approaches, which require each local node to satisfy the rank condition regarding the unknown input and output matrices, our method accommodates cases where a subset of nodes does not meet this requirement. This relaxation significantly broadens the range of practical applications. Simulation results are provided to demonstrate the effectiveness of the proposed design.

eess.SY↗

arXiv2025

Double Ionization to CO2 Produces Molecular Oxygen: A Roaming Mechanism

We report a combined experimental and theoretical study on the formation of O2+ by electronimpact double ionization and fragmentation of carbon dioxide (CO2) molecule. Through fragment ions and electron coincidence momentum imaging, we determine unambiguously the ionization mechanism by measuring the projectile energy loss in association with the C+ + O2+ channel. Further potential energy and trajectory calculations enable us to elucidate the dynamical details of this fragmentation process, in which a bond rearrangement pathway is found to proceed via the structural deformation to a triangular intermediate. Moreover, we demonstrate a new roaming pathway for formation of O2+ from CO22+, in which a frustrated C-O bond cleavage leaves the O atom without sufficient energy to escape. The O atom then wanders around varied configuration spaces of the flat potential energy regions and forms a C-O-O2+ intermediate prior to the final products C+ + O2+. Considering the large quantities of free electrons in interstellar space, the processes revealed here are expected to be significant and should be incorporated into atmospheric evolution models.

physics.atm-clus↗

arXiv2025

ColorGS: High-fidelity Surgical Scene Reconstruction with Colored Gaussian Splatting

High-fidelity reconstruction of deformable tissues from endoscopic videos remains challenging due to the limitations of existing methods in capturing subtle color variations and modeling global deformations. While 3D Gaussian Splatting (3DGS) enables efficient dynamic reconstruction, its fixed per-Gaussian color assignment struggles with intricate textures, and linear deformation modeling fails to model consistent global deformation. To address these issues, we propose ColorGS, a novel framework that integrates spatially adaptive color encoding and enhanced deformation modeling for surgical scene reconstruction. First, we introduce Colored Gaussian Primitives, which employ dynamic anchors with learnable color parameters to adaptively encode spatially varying textures, significantly improving color expressiveness under complex lighting and tissue similarity. Second, we design an Enhanced Deformation Model (EDM) that combines time-aware Gaussian basis functions with learnable time-independent deformations, enabling precise capture of both localized tissue deformations and global motion consistency caused by surgical interactions. Extensive experiments on DaVinci robotic surgery videos and benchmark datasets (EndoNeRF, StereoMIS) demonstrate that ColorGS achieves state-of-the-art performance, attaining a PSNR of 39.85 (1.5 higher than prior 3DGS-based methods) and superior SSIM (97.25\%) while maintaining real-time rendering efficiency. Our work advances surgical scene reconstruction by balancing high fidelity with computational practicality, critical for intraoperative guidance and AR/VR applications.

cs.CV↗

arXiv2025

FormaRL: Enhancing Autoformalization with no Labeled Data

Autoformalization is one of the central tasks in formal verification, while its advancement remains hindered due to the data scarcity and the absence efficient methods. In this work we propose \textbf{FormaRL}, a simple yet efficient reinforcement learning framework for autoformalization which only requires a small amount of unlabeled data. FormaRL integrates syntax check from Lean compiler and consistency check from large language model to calculate the reward, and adopts GRPO algorithm to update the formalizer. We also curated a proof problem dataset from undergraduate-level math materials, named \textbf{uproof}, in the hope to facilitate the exploration of autoformalization and theorem proving in advanced math. Experiments show that FormaRL can increase the pass@1 autoformalization accuracy of Qwen2.5-Coder-7B-Instruct by 4 $\sim$ 6x (4.04\% $\to$ 26.15\% on ProofNet and 2.4\% $\to$ 9.6\% on uproof) with merely 859 unlabeled data. And on uproof our method also achieved a strong improvement in out-of-distribution performance compared to existing open-source state-of-the-art autoformalizers on both pass@1 accuracy (6.2\% $\to$ 9.6\%) and pass@16 accuracy (24.4\% $\to$ 33.6\%). Training code of FormaRL is open-sourced at https://github.com/THUNLP-MT/FormaRL.

cs.AI↗

arXiv2025

Transfer Learning for Minimum Operating Voltage Prediction in Advanced Technology Nodes: Leveraging Legacy Data and Silicon Odometer Sensing

Accurate prediction of chip performance is critical for ensuring energy efficiency and reliability in semiconductor manufacturing. However, developing minimum operating voltage ($V_{min}$) prediction models at advanced technology nodes is challenging due to limited training data and the complex relationship between process variations and $V_{min}$. To address these issues, we propose a novel transfer learning framework that leverages abundant legacy data from the 16nm technology node to enable accurate $V_{min}$ prediction at the advanced 5nm node. A key innovation of our approach is the integration of input features derived from on-chip silicon odometer sensor data, which provide fine-grained characterization of localized process variations -- an essential factor at the 5nm node -- resulting in significantly improved prediction accuracy.

cs.LG↗

arXiv2025

Advancing Language Multi-Agent Learning with Credit Re-Assignment for Interactive Environment Generalization

LLM-based agents have made significant advancements in interactive environments, such as mobile operations and web browsing, and other domains beyond computer using. Current multi-agent systems universally excel in performance, compared to single agents, but struggle with generalization across environments due to predefined roles and inadequate strategies for generalizing language agents. The challenge of achieving both strong performance and good generalization has hindered the progress of multi-agent systems for interactive environments. To address these issues, we propose CollabUIAgents, a multi-agent reinforcement learning framework with a novel multi-agent credit re-assignment (CR) strategy, assigning process rewards with LLMs rather than environment-specific rewards and learning with synthesized preference data, in order to foster generalizable, collaborative behaviors among the role-free agents' policies. Empirical results show that our framework improves both performance and cross-environment generalizability of multi-agent systems. Moreover, our 7B-parameter system achieves results on par with or exceed strong closed-source models, and the LLM that guides the CR. We also provide insights in using granular CR rewards effectively for environment generalization, and accommodating trained LLMs in multi-agent systems.

cs.CL↗

arXiv2025

Topological resonance behaviors of surface acoustic waves under a surface liquid-layer loading and sensing applications

In this work, topological resonance behaviors of surface acoustic waves (SAW) under a surface liquid-layer loading are investigated. By revealing influences of the liquid-layer loading on wave velocity of SAW and topological indices (Berry curvature and Chern number) of topological interface-modes, a topological resonance peak with a high Q-factor is obtained based on couplings of a topological interface-mode waveguide and a resonant cavity under a surface liquid-layer loading. The results show that the degree of spatial-inversion-symmetry breaking resulting from structure parameters has an obvious influences on the topological resonance Q-factor, while the influences of the thickness of the liquid-layer loading on that is weak. It is worth noting that the topological resonance frequency is significantly sensitive to the liquid parameters. Based on that, a novel topological-resonance SAW liquid-phase sensor is proposed. Furthermore, sensing performances of this kind of sensor are simulated, which are used to sensing the concentration of hemoglobin, albumin, NaCl and NaI in aqueous solutions, and high sensitivities and Q-factors are obtained. The results presented in this paper can provide an important basis for the realization of highly sensitive and stable SAW micro-liquid-sample biomedical sensors in the future.

physics.app-ph↗

arXiv2025

Khan-GCL: Kolmogorov-Arnold Network Based Graph Contrastive Learning with Hard Negatives

Graph contrastive learning (GCL) has demonstrated great promise for learning generalizable graph representations from unlabeled data. However, conventional GCL approaches face two critical limitations: (1) the restricted expressive capacity of multilayer perceptron (MLP) based encoders, and (2) suboptimal negative samples that either from random augmentations-failing to provide effective 'hard negatives'-or generated hard negatives without addressing the semantic distinctions crucial for discriminating graph data. To this end, we propose Khan-GCL, a novel framework that integrates the Kolmogorov-Arnold Network (KAN) into the GCL encoder architecture, substantially enhancing its representational capacity. Furthermore, we exploit the rich information embedded within KAN coefficient parameters to develop two novel critical feature identification techniques that enable the generation of semantically meaningful hard negative samples for each graph representation. These strategically constructed hard negatives guide the encoder to learn more discriminative features by emphasizing critical semantic differences between graphs. Extensive experiments demonstrate that our approach achieves state-of-the-art performance compared to existing GCL methods across a variety of datasets and tasks.

cs.LG↗

arXiv2025

Vectorial Symmetry Decoding with Single-Particle Precision via Room-Temperature Lanthanide Luminescence Polarimetry

Determining the local symmetry of luminescent centers in crystals is critical for understanding and controlling their optical transitions, yet current methods are limited by stringent experimental requirements and ambiguous symmetry assignments. Here, we develop a robust computational electromagnetics framework that directly connect the local symmetry and chirality of rare-earth-doped single crystals to the polarization states of their emitted light. This framework is experimentally validated through the precise determination of point and space group symmetries using high-resolution, polarization-resolved micro-photoluminescence (μ-PL) spectra. Unlike conventional approaches that usually rely on analyzing multiple transitions at cryogenic temperatures, our technique operates at room temperature, requires only a single optical transition, and enables accurate orientation of symmetry axes. This enables deterministic polarization control of nano-emitters by tailoring symmetry groups and selecting appropriate transition dipoles, eliminating the need for bulky or complex photonic structures. Additionally, we demonstrate the function of bio-sensing, via determining single particle orientations in complex cellular environments using minimal polarization measurements. These results pave the way for advances in energy transfer systems, ultra-bright rare-earth nanocrystals, nanophotonic materials, and real-time single-particle tracking in biological contexts.

physics.optics↗