arXiv subjects

Pan Zhang

Publications and source records attributed to Pan Zhang.

At least 55 records · Page 3Linked to original sources

arXiv2025

Tensor Network Markov Chain Monte Carlo: Efficient Sampling of Three-Dimensional Spin Glasses and Beyond

Sampling the three-dimensional (3D) spin glass -- i.e., generating equilibrium configurations of a 3D lattice with quenched random couplings -- is widely regarded as one of the central and long-standing open problems in statistical physics. The rugged energy landscape, pronounced critical slowing down, and intrinsic ergodicity breaking render standard Monte Carlo methods severely inefficient, particularly for large systems at low temperatures. In this work, we introduce the Tensor Network Markov Chain Monte Carlo (TNMCMC) approach to address the issue. It generates large-scale collective updates in MCMC using tensor networks on the 2D slices of the 3D lattice, greatly improving the autocorrelation time and offering orders-of-magnitude speed-ups over conventional MCMC in generating unbiased samples of the Boltzmann distribution. We conduct numerical experiments on 3D spin glasses up to system size $64\times 64\times 64$ using a single CPU, and show that TNMCMC dramatically suppresses critical slowing down in large disordered systems, which usually require a supercomputer to perform MCMC simulations. Furthermore, we apply our approach to the 3-state Potts model up to system size $64\times 64\times 64$ using a single CPU, and show that the TNMCMC approach efficiently traverses the exponential barriers of the strong first-order transition, whereas conventional MCMC fails. Our results reveal that TNMCMC opens a promising path toward tackling long-standing, formidable three-dimensional problems in statistical physics.

cond-mat.stat-mech↗

arXiv2025

BatchTNMC: Efficient sampling of two-dimensional spin glasses using tensor network Monte Carlo

Efficient sampling of two-dimensional statistical physics systems remains a central challenge in computational statistical physics. Traditional Markov chain Monte Carlo (MCMC) methods, including cluster algorithms, provide only partial solutions, as their efficiency collapses for large systems in the presence of frustration and quenched disorder. The recently proposed Tensor Network Monte Carlo (TNMC) method offers a promising alternative, yet its original implementation suffers from inefficiencies due to the lack of scalable parallel sampling. In this work, we introduce BatchTNMC, a GPU-optimized and parallelized implementation of TNMC tailored for large-scale simulations of two-dimensional spin glasses. By leveraging batch processing and parallel sampling across multiple disorder realizations, our implementation achieves speedups of up to five orders of magnitude compared with the original serial scheme. Benchmarking on two-dimensional spin glasses demonstrates dramatic gains in efficiency: for instance, on a single GPU, BatchTNMC concurrently produces 1000 uncorrelated and unbiased samples across 1000 disorder realizations on $1024\times 1024$ lattices in just 3.3 hours, with an acceptance probability of 37%. These results establish BatchTNMC as a scalable and powerful computational framework for the study of two-dimensional disordered spin glass systems.

cond-mat.stat-mech↗

arXiv2025

X-Prompt: Towards Universal In-Context Image Generation in Auto-Regressive Vision Language Foundation Models

In-context generation is a key component of large language models' (LLMs) open-task generalization capability. By leveraging a few examples as context, LLMs can perform both in-domain and out-of-domain tasks. Recent advancements in auto-regressive vision-language models (VLMs) built upon LLMs have showcased impressive performance in text-to-image generation. However, the potential of in-context learning for general image generation tasks remains largely unexplored. To address this, we introduce X-Prompt, a purely auto-regressive large-vision language model designed to deliver competitive performance across a wide range of both seen and unseen image generation tasks, all within a unified in-context learning framework. X-Prompt incorporates a specialized design that efficiently compresses valuable features from in-context examples, supporting longer in-context token sequences and improving its ability to generalize to unseen tasks. A unified training task for both text and image prediction enables X-Prompt to handle general image generation with enhanced task awareness from in-context examples. Extensive experiments validate the model's performance across diverse seen image generation tasks and its capacity to generalize to previously unseen tasks.

cs.CV↗

arXiv2025

Dissipation-induced Half Quantized Conductance in One-dimensional Topological Systems

Quantized conductance from topologically protected edge states is a hallmark of two-dimensional topological phases. In contrast, edge states in one-dimensional (1D) topological systems cannot transmit current across the insulating bulk, rendering their topological nature invisible in transport. In this work, we investigate the transport properties of the Su-Schrieffer-Heeger model with gain and loss, and show that the zero-energy conductance exhibits qualitatively distinct behaviors between the topologically trivial and nontrivial phases, depending on the hybridization and dissipation strengths. Crucially, we analytically demonstrate that the conductance can become half-quantized in the topologically nontrivial phase, a feature absent in the trivial phase. We further show that the half quantization predominantly originates from transport channels involving gain/loss and edge states. Our results uncover a new mechanism for realizing quantized transport in 1D topological systems and highlight the nontrivial role of dissipation in enabling topological signatures in open quantum systems.

cond-mat.mes-hall↗

arXiv2025

SAM2Long: Enhancing SAM 2 for Long Video Segmentation with a Training-Free Memory Tree

The Segment Anything Model 2 (SAM 2) has emerged as a powerful foundation model for object segmentation in both images and videos, paving the way for various downstream video applications. The crucial design of SAM 2 for video segmentation is its memory module, which prompts object-aware memories from previous frames for current frame prediction. However, its greedy-selection memory design suffers from the "error accumulation" problem, where an errored or missed mask will cascade and influence the segmentation of the subsequent frames, which limits the performance of SAM 2 toward complex long-term videos. To this end, we introduce SAM2Long, an improved training-free video object segmentation strategy, which considers the segmentation uncertainty within each frame and chooses the video-level optimal results from multiple segmentation pathways in a constrained tree search manner. In practice, we maintain a fixed number of segmentation pathways throughout the video. For each frame, multiple masks are proposed based on the existing pathways, creating various candidate branches. We then select the same fixed number of branches with higher cumulative scores as the new pathways for the next frame. After processing the final frame, the pathway with the highest cumulative score is chosen as the final segmentation result. Benefiting from its heuristic search design, SAM2Long is robust toward occlusions and object reappearances, and can effectively segment and track objects for complex long-term videos. Notably, SAM2Long achieves an average improvement of 3.0 points across all 24 head-to-head comparisons, with gains of up to 5.3 points in J&F on long-term video object segmentation benchmarks such as SA-V and LVOS. The code is released at https://github.com/Mark12Ding/SAM2Long.

cs.CV↗

arXiv2025

ScaleCap: Inference-Time Scalable Image Captioning via Dual-Modality Debiasing

This paper presents ScaleCap, an inference-time scalable image captioning strategy that generates comprehensive and detailed image captions. The key challenges of high-quality image captioning lie in the inherent biases of LVLMs: multimodal bias resulting in imbalanced descriptive granularity, offering detailed accounts of some elements while merely skimming over others; linguistic bias leading to hallucinated descriptions of non-existent objects. To address these issues, we propose a scalable debiased captioning strategy, which continuously enriches and calibrates the caption with increased inference budget. Specifically, we propose two novel components: heuristic question answering and contrastive sentence rating. The former generates content-specific questions based on the image and answers them to progressively inject relevant information into the caption. The latter employs sentence-level offline contrastive decoding to effectively identify and eliminate hallucinations caused by linguistic biases. With increased inference cost, more heuristic questions are raised by ScaleCap to progressively capture additional visual details, generating captions that are more accurate, balanced, and informative. Extensive modality alignment experiments demonstrate the effectiveness of ScaleCap. Annotating 450K images with ScaleCap and using them for LVLM pretraining leads to consistent performance gains across 11 widely used benchmarks. Furthermore, ScaleCap showcases superb richness and fidelity of generated captions with two additional tasks: replacing images with captions in VQA task, and reconstructing images from captions to assess semantic coverage. Code is available at https://github.com/Cooperx521/ScaleCap.

cs.CV↗

arXiv2025

Correcting a noisy quantum computer using a quantum computer

Quantum computers require error correction to achieve universal quantum computing. However, current decoding of quantum error-correcting codes relies on classical computation, which is slower than quantum operations in superconducting qubits. This discrepancy makes the practical implementation of real-time quantum error correction challenging. In this work, we propose a decoding scheme that leverages the operations of the quantum circuit itself. Given a noisy quantum circuit $A$, we train a decoding quantum circuit $B$ using syndrome measurements to identify the logical operators needed to correct errors in circuit $A$. The trained quantum circuit $B$ can be deployed on quantum devices, such as superconducting qubits, to perform real-time decoding and error correction. Our approach is applicable to general quantum codes with multiple logical qubits and operates efficiently under various noise conditions, and the decoding speed matches the speed of the quantum circuits being corrected. We have conducted numerical experiments using surface codes up to distance 7 under circuit-level noise, demonstrating performance on par with the classical minimum-weight perfect matching algorithm. Interestingly, our method reveals that the traditionally classical task of decoding error-correcting codes can be accomplished without classical devices or measurements. This insight paves the way for the development of self-correcting quantum computers.

quant-ph↗

arXiv2025

Towards Storage-Efficient Visual Document Retrieval: An Empirical Study on Reducing Patch-Level Embeddings

Despite the strong performance of ColPali/ColQwen2 in Visualized Document Retrieval (VDR), it encodes each page into multiple patch-level embeddings and leads to excessive memory usage. This empirical study investigates methods to reduce patch embeddings per page at minimum performance degradation. We evaluate two token-reduction strategies: token pruning and token merging. Regarding token pruning, we surprisingly observe that a simple random strategy outperforms other sophisticated pruning methods, though still far from satisfactory. Further analysis reveals that pruning is inherently unsuitable for VDR as it requires removing certain page embeddings without query-specific information. Turning to token merging (more suitable for VDR), we search for the optimal combinations of merging strategy across three dimensions and develop Light-ColPali/ColQwen2. It maintains 98.2% of retrieval performance with only 11.8% of original memory usage, and preserves 94.6% effectiveness at 2.8% memory footprint. We expect our empirical findings and resulting Light-ColPali/ColQwen2 offer valuable insights and establish a competitive baseline for future research towards efficient VDR.

cs.IR↗

arXiv2025

SongComposer: A Large Language Model for Lyric and Melody Generation in Song Composition

Creating lyrics and melodies for the vocal track in a symbolic format, known as song composition, demands expert musical knowledge of melody, an advanced understanding of lyrics, and precise alignment between them. Despite achievements in sub-tasks such as lyric generation, lyric-to-melody, and melody-to-lyric, etc, a unified model for song composition has not yet been achieved. In this paper, we introduce SongComposer, a pioneering step towards a unified song composition model that can readily create symbolic lyrics and melodies following instructions. SongComposer is a music-specialized large language model (LLM) that, for the first time, integrates the capability of simultaneously composing lyrics and melodies into LLMs by leveraging three key innovations: 1) a flexible tuple format for word-level alignment of lyrics and melodies, 2) an extended tokenizer vocabulary for song notes, with scalar initialization based on musical knowledge to capture rhythm, and 3) a multi-stage pipeline that captures musical structure, starting with motif-level melody patterns and progressing to phrase-level structure for improved coherence. Extensive experiments demonstrate that SongComposer outperforms advanced LLMs, including GPT-4, in tasks such as lyric-to-melody generation, melody-to-lyric generation, song continuation, and text-to-song creation. Moreover, we will release SongCompose, a large-scale dataset for training, containing paired lyrics and melodies in Chinese and English.

cs.SD↗

arXiv2025

VideoRoPE: What Makes for Good Video Rotary Position Embedding?

While Rotary Position Embedding (RoPE) and its variants are widely adopted for their long-context capabilities, the extension of the 1D RoPE to video, with its complex spatio-temporal structure, remains an open challenge. This work first introduces a comprehensive analysis that identifies four key characteristics essential for the effective adaptation of RoPE to video, which have not been fully considered in prior work. As part of our analysis, we introduce a challenging V-NIAH-D (Visual Needle-In-A-Haystack with Distractors) task, which adds periodic distractors into V-NIAH. The V-NIAH-D task demonstrates that previous RoPE variants, lacking appropriate temporal dimension allocation, are easily misled by distractors. Based on our analysis, we introduce \textbf{VideoRoPE}, with a \textit{3D structure} designed to preserve spatio-temporal relationships. VideoRoPE features \textit{low-frequency temporal allocation} to mitigate periodic oscillations, a \textit{diagonal layout} to maintain spatial symmetry, and \textit{adjustable temporal spacing} to decouple temporal and spatial indexing. VideoRoPE consistently surpasses previous RoPE variants, across diverse downstream tasks such as long video retrieval, video understanding, and video hallucination. Our code will be available at \href{https://github.com/Wiselnn570/VideoRoPE}{https://github.com/Wiselnn570/VideoRoPE}.

cs.CV↗

arXiv2025

SongGen: A Single Stage Auto-regressive Transformer for Text-to-Song Generation

Text-to-song generation, the task of creating vocals and accompaniment from textual inputs, poses significant challenges due to domain complexity and data scarcity. Existing approaches often employ multi-stage generation procedures, leading to cumbersome training and inference pipelines, as well as suboptimal overall generation quality due to error accumulation across stages. In this paper, we propose SongGen, a fully open-source, single-stage auto-regressive transformer designed for controllable song generation. The proposed model facilitates fine-grained control over diverse musical attributes, including lyrics and textual descriptions of instrumentation, genre, mood, and timbre, while also offering an optional three-second reference clip for voice cloning. Within a unified auto-regressive framework, SongGen supports two output modes: mixed mode, which generates a mixture of vocals and accompaniment directly, and dual-track mode, which synthesizes them separately for greater flexibility in downstream applications. We explore diverse token pattern strategies for each mode, leading to notable improvements and valuable insights. Furthermore, we design an automated data preprocessing pipeline with effective quality control. To foster community engagement and future research, we will release our model weights, training code, annotated data, and preprocessing pipeline. The code is available at https://github.com/LiuZH-19/SongGen.

cs.SD↗

arXiv2025

On the steadiness of symmetric solutions to higher order perturbations of KdV

We consider the traveling structure of symmetric solutions to the Rosenau-Kawahara-RLW equation and the perturbed R-KdV-RLW equation. Both equations are higher order perturbations of the classical KdV equation. For the Rosenau-Kawahara-RLW equation, we prove that classical and weak solutions with a priori symmetry must be traveling solutions. For the more complicated perturbed R-KdV-RLW equation, we classify all symmetric traveling solutions, and prove that there exists no nontrivial symmetric traveling solution of solitary type once dissipation or shoaling perturbations exist. This gives a new perspective for evaluating the suitableness of a model for water waves. In addition, this result illustrates the sharpness of the symmetry principle in [Int. Math. Res. Not. IMRN, 2009; Ehrnstrom, Holden \& Raynaud] for solitary waves.

math.AP↗

arXiv2025

InternLM-XComposer2.5-Reward: A Simple Yet Effective Multi-Modal Reward Model

Despite the promising performance of Large Vision Language Models (LVLMs) in visual understanding, they occasionally generate incorrect outputs. While reward models (RMs) with reinforcement learning or test-time scaling offer the potential for improving generation quality, a critical gap remains: publicly available multi-modal RMs for LVLMs are scarce, and the implementation details of proprietary models are often unclear. We bridge this gap with InternLM-XComposer2.5-Reward (IXC-2.5-Reward), a simple yet effective multi-modal reward model that aligns LVLMs with human preferences. To ensure the robustness and versatility of IXC-2.5-Reward, we set up a high-quality multi-modal preference corpus spanning text, image, and video inputs across diverse domains, such as instruction following, general understanding, text-rich documents, mathematical reasoning, and video understanding. IXC-2.5-Reward achieves excellent results on the latest multi-modal reward model benchmark and shows competitive performance on text-only reward model benchmarks. We further demonstrate three key applications of IXC-2.5-Reward: (1) Providing a supervisory signal for RL training. We integrate IXC-2.5-Reward with Proximal Policy Optimization (PPO) yields IXC-2.5-Chat, which shows consistent improvements in instruction following and multi-modal open-ended dialogue; (2) Selecting the best response from candidate responses for test-time scaling; and (3) Filtering outlier or noisy samples from existing image and video instruction tuning training data. To ensure reproducibility and facilitate further research, we have open-sourced all model weights and training recipes at https://github.com/InternLM/InternLM-XComposer/tree/main/InternLM-XComposer-2.5-Reward

cs.CV↗

arXiv2025

HiFlow: Training-free High-Resolution Image Generation with Flow-Aligned Guidance

Text-to-image (T2I) diffusion/flow models have drawn considerable attention recently due to their remarkable ability to deliver flexible visual creations. Still, high-resolution image synthesis presents formidable challenges due to the scarcity and complexity of high-resolution content. Recent approaches have investigated training-free strategies to enable high-resolution image synthesis with pre-trained models. However, these techniques often struggle with generating high-quality visuals and tend to exhibit artifacts or low-fidelity details, as they typically rely solely on the endpoint of the low-resolution sampling trajectory while neglecting intermediate states that are critical for preserving structure and synthesizing finer detail. To this end, we present HiFlow, a training-free and model-agnostic framework to unlock the resolution potential of pre-trained flow models. Specifically, HiFlow establishes a virtual reference flow within the high-resolution space that effectively captures the characteristics of low-resolution flow information, offering guidance for high-resolution generation through three key aspects: initialization alignment for low-frequency consistency, direction alignment for structure preservation, and acceleration alignment for detail fidelity. By leveraging such flow-aligned guidance, HiFlow substantially elevates the quality of high-resolution image synthesis of T2I models and demonstrates versatility across their personalized variants. Extensive experiments validate HiFlow's capability in achieving superior high-resolution image quality over state-of-the-art methods.

cs.CV↗

arXiv2025

MM-IFEngine: Towards Multimodal Instruction Following

The Instruction Following (IF) ability measures how well Multi-modal Large Language Models (MLLMs) understand exactly what users are telling them and whether they are doing it right. Existing multimodal instruction following training data is scarce, the benchmarks are simple with atomic instructions, and the evaluation strategies are imprecise for tasks demanding exact output constraints. To address this, we present MM-IFEngine, an effective pipeline to generate high-quality image-instruction pairs. Our MM-IFEngine pipeline yields large-scale, diverse, and high-quality training data MM-IFInstruct-23k, which is suitable for Supervised Fine-Tuning (SFT) and extended as MM-IFDPO-23k for Direct Preference Optimization (DPO). We further introduce MM-IFEval, a challenging and diverse multi-modal instruction-following benchmark that includes (1) both compose-level constraints for output responses and perception-level constraints tied to the input images, and (2) a comprehensive evaluation pipeline incorporating both rule-based assessment and judge model. We conduct SFT and DPO experiments and demonstrate that fine-tuning MLLMs on MM-IFInstruct-23k and MM-IFDPO-23k achieves notable gains on various IF benchmarks, such as MM-IFEval (+10.2$\%$), MIA (+7.6$\%$), and IFEval (+12.3$\%$). We have fully open-sourced the datasets (both SFT and DPO), evaluation code and training scripts at https://github.com/SYuan03/MM-IFEngine.

cs.CV↗

arXiv2025

Learning eigenstates of quantum many-body Hamiltonians within the symmetric subspaces using neural network quantum states

The exploration of neural network quantum states has become widespread in the studies of complicated quantum many-body systems. However, achieving high precision remains challenging due to the exponential growth of Hilbert space size and the intricate sign structures. Utilizing symmetries of the physical system, we propose a method to evaluate and sample the variational ansatz within a symmetric subspace. This approach isolates different symmetry sectors, reducing the relevant Hilbert space size by a factor approximately proportional to the size of the symmetry group. It is inspired by exact diagonalization techniques and the work of Choo et al. in Phys. Rev. Lett. 121, 167204 (2018). We validate our method using the frustrated spin-1/2 $J_1$-$J_2$ antiferromagnetic Heisenberg chain and compare its performance to the case without symmetrization. The results indicate that our symmetric subspace approach achieves a substantial improvement over the full Hilbert space on optimizing the ansatz, reducing the energy error by orders of magnitude. We also compare the results on degenerate eigenstates with different quantum numbers, highlighting the advantage of operating within a smaller Hilbert subspace.

cond-mat.str-el↗

arXiv2025

Integrating Neural Networks and Tensor Networks for Computing Free Energy

Computing free energy is a fundamental problem in statistical physics. Recently, two distinct methods have been developed and have demonstrated remarkable success: the tensor-network-based contraction method and the neural-network-based variational method. Tensor networks are accu?rate, but their application is often limited to low-dimensional systems due to the high computational complexity in high-dimensional systems. The neural network method applies to systems with general topology. However, as a variational method, it is not as accurate as tensor networks. In this work, we propose an integrated approach, tensor-network-based variational autoregressive networks (TNVAN), that leverages the strengths of both tensor networks and neural networks: combining the variational autoregressive neural network's ability to compute an upper bound on free energy and perform unbiased sampling from the variational distribution with the tensor network's power to accurately compute the partition function for small sub-systems, resulting in a robust method for precisely estimating free energy. To evaluate the proposed approach, we conducted numerical experiments on spin glass systems with various topologies, including two-dimensional lattices, fully connected graphs, and random graphs. Our numerical results demonstrate the superior accuracy of our method compared to existing approaches. In particular, it effectively handles systems with long-range interactions and leverages GPU efficiency without requiring singular value decomposition, indicating great potential in tackling statistical mechanics problems and simulating high-dimensional complex systems through both tensor networks and neural networks.

cond-mat.stat-mech↗

arXiv2025

Prethermalization by Random Multipolar Driving on a 78-Qubit Superconducting Processor

Time-dependent drives hold the promise of realizing non-equilibrium many-body phenomena that are absent in undriven systems. Yet, drive-induced heating normally destabilizes the systems, which can be parametrically suppressed in the high-frequency regime by using periodic (Floquet) drives. It remains largely unknown to what extent highly controllable quantum simulators can suppress heating in non-periodically driven systems. Using the 78-qubit superconducting quantum processor, Chuang-tzu 2.0, we report the experimental observation of long-lived prethermal phases in many-body systems with tunable heating rates, driven by structured random protocols, characterized by $n$-multipolar temporal correlations. By measuring both the particle imbalance and subsystem entanglement entropy, we monitor the entire heating process over 1,000 driving cycles and observe the existence of the prethermal plateau. The prethermal lifetime is `doubly tunable': one way by driving frequency, the other by multipolar order; it grows algebraically with the frequency with the universal scaling exponent $2n{+}1$. Using quantum state tomography on different subsystems, we demonstrate a non-uniform spatial entanglement distribution and observe a crossover from area-law to volume-law entanglement scaling. With 78 qubits and 137 couplers in a 2D configuration, the entire far-from-equilibrium heating dynamics are beyond the reach of simulation using tensor-network numerical techniques. Our work highlights superconducting quantum processors as a powerful platform for exploring universal scaling laws and non-equilibrium phases of matter in driven systems in regimes where classical simulation faces formidable challenges.

quant-ph↗