arXiv subjects

Pu Jian

Publications and source records attributed to Pu Jian.

At least 19 recordsLinked to original sources

arXiv2026

SceneActBench: Can Agents Act on the 3D Scenes They See?

Vision-language model (VLM) agents increasingly use tools to act on 3D scenes rather than only describe them. Existing 3D benchmarks score textual responses or single-object operations, leaving agent action on complete multi-object 3D scenes under evaluated. We present SceneActBench, a benchmark for visually conditioned action across five 3D tasks under a unified agent-environment loop. Given PNG images or sampled video frames and, where applicable, supplied 3D assets, an agent acts on a 3D environment. We evaluate each final output against hidden ground truth with task-specific geometric metrics. SceneActBench comprises five tasks built from 210 source instances, yielding 520 task cases including paired input conditions. Every task runs through one fixed agent loop to keep the comparison fair. Across eleven proprietary VLM configurations, Overall scores span 38.6-50.2, and none performs consistently well across tasks. We further analyse where and how failures manifest.

cs.AI↗

arXiv2026

AMR-SD: Asymmetric Meta-Reflective Self-Distillation for Token-Level Credit Assignment

The alignment of Large Language Models (LLMs) for complex reasoning heavily relies on Reinforcement Learning with Verifiable Rewards (RLVR). However, standard algorithms like GRPO apply sequence-level rewards uniformly to all tokens, creating a severe credit-assignment bottleneck. While on-policy self-distillation attempts to resolve this by conditioning a self-teacher on privileged contexts, direct exposure to raw oracle solutions often induces over-conditioned teacher distributions, implicit answer leakage, and late-stage training collapse. To overcome these limitations, we propose Asymmetric Meta-Reflective Self-Distillation (AMR-SD). Instead of conditioning directly on raw reference traces, AMR-SD inserts a reflection bottleneck: it compresses diagnostic signals -- from verifier outcomes, peer rollouts, or reference feedback -- into concise, self-generated Socratic hints and critiques. Furthermore, we introduce Causal Information Gain (CIG) with an asymmetric, ReLU-gated threshold to translate these reflections into sparse, highly precise token-level advantage modulations. Combined with temporal annealing, this mechanism preserves the base environmental reward while filtering out distributional noise. Experiments across scientific, mathematical, and tool-use benchmarks demonstrate that AMR-SD significantly outperforms existing baselines, achieving robust long-horizon stability and successfully preventing late-stage collapse.

cs.AI↗

arXiv2025

Look Again, Think Slowly: Enhancing Visual Reflection in Vision-Language Models

Recent advances in text-only "slow-thinking" reasoning have prompted efforts to transfer this capability to vision-language models (VLMs), for training visual reasoning models (\textbf{VRMs}). owever, such transfer faces critical challenges: Effective "slow thinking" in VRMs requires \textbf{visual reflection}, the ability to check the reasoning process based on visual information. Through quantitative analysis, we observe that current VRMs exhibit limited visual reflection, as their attention to visual information diminishes rapidly with longer generated responses. To address this challenge, we propose a new VRM \textbf{Reflection-V}, which enhances visual reflection based on reasoning data construction for cold-start and reward design for reinforcement learning (RL). Firstly, we construct vision-centered reasoning data by leveraging an agent that interacts between VLMs and reasoning LLMs, enabling cold-start learning of visual reflection patterns. Secondly, a visual attention based reward model is employed during RL to encourage reasoning based on visual information. Therefore, \textbf{Reflection-V} demonstrates significant improvements across multiple visual reasoning benchmarks. Furthermore, \textbf{Reflection-V} maintains a stronger and more consistent reliance on visual information during visual reasoning, indicating effective enhancement in visual reflection capabilities.

cs.CV↗

arXiv2025

Teaching Vision-Language Models to Ask: Resolving Ambiguity in Visual Questions

In visual question answering (VQA) context, users often pose ambiguous questions to visual language models (VLMs) due to varying expression habits. Existing research addresses such ambiguities primarily by rephrasing questions. These approaches neglect the inherently interactive nature of user interactions with VLMs, where ambiguities can be clarified through user feedback. However, research on interactive clarification faces two major challenges: (1) Benchmarks are absent to assess VLMs' capacity for resolving ambiguities through interaction; (2) VLMs are trained to prefer answering rather than asking, preventing them from seeking clarification. To overcome these challenges, we introduce \textbf{ClearVQA} benchmark, which targets three common categories of ambiguity in VQA context, and encompasses various VQA scenarios.

cs.CV↗

arXiv2025

TableRAG: A Retrieval Augmented Generation Framework for Heterogeneous Document Reasoning

Retrieval-Augmented Generation (RAG) has demonstrated considerable effectiveness in open-domain question answering. However, when applied to heterogeneous documents, comprising both textual and tabular components, existing RAG approaches exhibit critical limitations. The prevailing practice of flattening tables and chunking strategies disrupts the intrinsic tabular structure, leads to information loss, and undermines the reasoning capabilities of LLMs in multi-hop, global queries. To address these challenges, we propose TableRAG, an SQL-based framework that unifies textual understanding and complex manipulations over tabular data. TableRAG iteratively operates in four steps: context-sensitive query decomposition, text retrieval, SQL programming and execution, and compositional intermediate answer generation. We also develop HeteQA, a novel benchmark designed to evaluate the multi-hop heterogeneous reasoning capabilities. Experimental results demonstrate that TableRAG consistently outperforms existing baselines on both public datasets and our HeteQA, establishing a new state-of-the-art for heterogeneous document question answering. We release TableRAG at https://github.com/yxh-y/TableRAG/tree/main.

cs.CL↗

arXiv2025

CROP: Contextual Region-Oriented Visual Token Pruning

Current VLM-based VQA methods often process entire images, leading to excessive visual tokens that include redundant information irrelevant to the posed question. This abundance of unnecessary image details creates numerous visual tokens, drastically increasing memory and computational requirements in VLMs. To address this, we propose Contextual Region-Oriented Visual Token Pruning (CROP), a novel framework to compress visual tokens through a two-step process: Localization and Pruning. Specifically, CROP first employs an efficient model to identify the contextual region relevant to the input query. Subsequently, two distinct strategies are introduced for pruning: (1) Pre-LLM Compression (PLC), which adaptively compresses different image regions with varying ratios, and (2) Inner-LLM Pruning (ILP), a training-free method that prunes tokens within early LLM layers guided by the identified contextual region. Extensive experiments on a wide range of VQA tasks demonstrate that CROP significantly outperforms existing visual token pruning methods and achieves state-of-the-art performance.

cs.CV↗

arXiv2025

KTAE: A Model-Free Algorithm to Key-Tokens Advantage Estimation in Mathematical Reasoning

Recent advances have demonstrated that integrating reinforcement learning with rule-based rewards can significantly enhance the reasoning capabilities of large language models, even without supervised fine-tuning. However, prevalent reinforcement learning algorithms such as GRPO and its variants like DAPO, suffer from a coarse granularity issue when computing the advantage. Specifically, they compute rollout-level advantages that assign identical values to every token within a sequence, failing to capture token-specific contributions and hindering effective learning. To address this limitation, we propose Key-token Advantage Estimation (KTAE) - a novel algorithm that estimates fine-grained, token-level advantages without introducing additional models. KTAE leverages the correctness of sampled rollouts and applies statistical analysis to quantify the importance of individual tokens within a sequence to the final outcome. This quantified token-level importance is then combined with the rollout-level advantage to obtain a more fine-grained token-level advantage estimation. Empirical results show that models trained with GRPO+KTAE and DAPO+KTAE outperform baseline methods across five mathematical reasoning benchmarks. Notably, they achieve higher accuracy with shorter responses and even surpass R1-Distill-Qwen-1.5B using the same base model.

cs.AI↗

arXiv2025

Towards Scientific Intelligence: A Survey of LLM-based Scientific Agents

As scientific research becomes increasingly complex, innovative tools are needed to manage vast data, facilitate interdisciplinary collaboration, and accelerate discovery. Large language models (LLMs) are now evolving into LLM-based scientific agents that automate critical tasks ranging from hypothesis generation and experiment design to data analysis and simulation. Unlike general-purpose LLMs, these specialized agents integrate domain-specific knowledge, advanced tool sets, and robust validation mechanisms, enabling them to handle complex data types, ensure reproducibility, and drive scientific breakthroughs. This survey provides a focused review of the architectures, design, benchmarks, applications, and ethical considerations surrounding LLM-based scientific agents. We highlight why they differ from general agents and the ways in which they advance research across various scientific fields. By examining their development and challenges, this survey offers a comprehensive roadmap for researchers and practitioners to harness these agents for more efficient, reliable, and ethically sound scientific discovery.

cs.AI↗

arXiv2023

ChineseWebText: Large-scale High-quality Chinese Web Text Extracted with Effective Evaluation Model

During the development of large language models (LLMs), the scale and quality of the pre-training data play a crucial role in shaping LLMs' capabilities. To accelerate the research of LLMs, several large-scale datasets, such as C4 [1], Pile [2], RefinedWeb [3] and WanJuan [4], have been released to the public. However, most of the released corpus focus mainly on English, and there is still lack of complete tool-chain for extracting clean texts from web data. Furthermore, fine-grained information of the corpus, e.g. the quality of each text, is missing. To address these challenges, we propose in this paper a new complete tool-chain EvalWeb to extract Chinese clean texts from noisy web data. First, similar to previous work, manually crafted rules are employed to discard explicit noisy texts from the raw crawled web contents. Second, a well-designed evaluation model is leveraged to assess the remaining relatively clean data, and each text is assigned a specific quality score. Finally, we can easily utilize an appropriate threshold to select the high-quality pre-training data for Chinese. Using our proposed approach, we release the largest and latest large-scale high-quality Chinese web text ChineseWebText, which consists of 1.42 TB and each text is associated with a quality score, facilitating the LLM researchers to choose the data according to the desired quality thresholds. We also release a much cleaner subset of 600 GB Chinese data with the quality exceeding 90%.

cs.CL↗

arXiv2020

Pointing Error Compensation For Inter-Satellite Communication Using Multi-Plane Light Conversion Spatial Demultiplexer

In this work, we study the effect of beam deviation angle at the reception side and calculate the theoretical demultiplexed collected energy for up to 15 modes, investigating the influence of the ratio between incoming beam size and fundamental HG mode's waist. We show this approach greatly enhances the collection efficiency, tolerating tip-tilt error of more than 3 times compared to a Gaussian beam alone. We also show that, depending on wait size, a trade-off between collection efficiency at small angles and maximum acceptance angle can be achieved.

physics.optics↗

arXiv2019

Fabrication and Characterization of a Mode-selective 45-Mode Spatial Multiplexer based on Multi-Plane Light Conversion

Space Division Multiplexing (SMD) is a very attractive technique for addressing the ever-growing demands in transmission capacity by enabling the use of a new parameter \textemdash\ space \textemdash\ to increase the number of channels in multi-mode fibers. One key component to build a spatially multiplexed-based optical network is a spatial multiplexer and demultiplexer combining signals from multiple single-mode fibers into as many channels in a multi-mode fiber. In this article, we report the fabrication and characterization of a pair of 45-mode spatial multiplexer and demultiplexer saturating all the modes of a standard 50~$μ$m core graded-index (OM2) multi-mode fiber. The multiplexers are based on Multi-Plane Light Conversion (MPLC), a technique that enables the control of the transverse shape of the light by multiple reflections on specifically designed phase plates. We show that by using a separable variable basis of modes, such as Hermite-Gaussian (HG) modes, we are able to drastically reduce the number of reflections hence reducing the insertion losses and modal crosstalks. The multiplexers typically show an average 4~dB insertion loss and -28~dB cross-talk across the C band. Finally, we emphasize the use of this higher-order modes multiplexer to explore the propagation properties inside multi-mode fibers and more specifically the mode group crosstalks as well as the impact of fiber bending.

physics.app-ph↗

arXiv2017

80-Gbit/s 100-m Free-Space Optical Data Transmission Link via a Flying UAV Using Multiplexing of Orbital-Angular-Momentum Beams

We explore the use of orbital-angular-momentum (OAM)-multiplexing to increase the capacity of free-space data transmission to moving platforms, with an added potential benefit of decreasing the probability of data intercept. Specifically, we experimentally demonstrate and characterize the performance of an OAM-multiplexed, free-space optical (FSO) communications link between a ground station and a moving unmanned-aerial-vehicle (UAV). We achieve a total capacity of 80 Gbit/s up to 100-m-roundtrip link by multiplexing 2 OAM beams, each carrying a 40-Gbit/s quadrature-phase-shift-keying (QPSK) signal. Moreover, we investigate for static, hovering, and moving conditions the effects of channel impairments, including: tracking errors, propeller-induced airflows, power loss, intermodal crosstalk, and system bit error rate (BER). We find the following: (a) when the UAV hovers in the air, the power on the desired mode fluctuates by 2.1 dB, while the crosstalk to the other mode is -19 dB below the power on the desired mode; and (b) when the UAV moves in the air, the power fluctuation on the desired mode increases to 4.3 dB and the crosstalk to the other mode increases to -10 dB. Furthermore, the channel crosstalk decreases with an increase in OAM mode spacing.

physics.optics↗

arXiv2017

Quantum-limited measurements of distance fluctuations with a multimode detector

An experimental scheme is introduced to measure multiple parameters that are encoded in the phase quadrature of a light beam. Using a modal description and a spectrally-resolved homodyne detection, it is shown that all of the information is collected simultaneously, such that a single measurement allows extracting the value of multiple parameters \emph{post-facto}. With a femtosecond laser source, we apply this scheme to a measurement of the delay between two pulses with a shot-noise limited sensitivity as well as extracting the dispersion value of a dispersive medium.

physics.optics↗

arXiv2016

Absolute Measurement of Quantum-Limited Interferometric Displacements

A methodology is introduced that enables an absolute, quantum-limited measurement of sub-wavelength interferometric displacements. The technique utilizes a high-frequency optical path modulation within an interferometer operated in a homodyne configuration. All of the information necessary to fully characterize the resultant path displacement is contained within the relative strengths of the various harmonics of the phase modulation. The method, which is straightforward and readily implementable, allows a direct measurement of the theoretical Cramér-Rao limit of detection without any assumptions on the nature of the light source.

quant-ph↗

arXiv2014

Efficient and mode selective spatial mode multiplexer based on Multi-Plane Light Conversion

We designed and built a new type of spatial mode multiplexer, based on Multi-Plane Light Conversion (MPLC), with very low intrinsic loss and high mode selectivity. In this first demonstration we show that a typical 3-mode multiplexer achieves a mode selectivity better than -23 dB and a total insertion efficiency of -4.1 dB (optical coating improvements could increase efficiency to -2.4 dB), across the full C-band. Moreover this multiplexer is able to perform any mode conversion, and we demonstrate its performance for the first 6 eigenmodes of a few-mode fiber: LP$_{01}$, LP$_{11\mathrm{a}}$, LP$_{11\mathrm{b}}$, LP$_{02}$, LP$_{21\mathrm{a}}$ and LP$_{21\mathrm{b}}$.

physics.optics↗

arXiv2014

Precision measurements with photon-subtracted or photon-added Gaussian states

Photon-subtracted and photon-added Gaussian states are amongst the simplest non-Gaussian states that are experimentally available. It is generally believed that they are some of the best candidates to enhance sensitivity in parameter extraction. We derive here the quantum Cramér-Rao bound for such states and find that for large photon numbers photon-subtraction or -addition only leads to a small correction of the quantum Fisher information (QFI). On the other hand a divergence of the QFI appears for very small squeezing in the limit of vanishing photon number in the case of photon subtraction, implying an arbitrarily precise measurement with almost no light. However, at least for the standard and experimentally established preparation scheme, the decreasing success probability of the preparation in that limit exactly cancels the divergence, leading to finite sensitivity per square root of Hertz, when the duration of the preparation is taken into account.

quant-ph↗

arXiv2013

Quantum parameter estimation using general single-mode Gaussian states

We calculate the quantum Cramér--Rao bound for the sensitivity with which one or several parameters, encoded in a general single-mode Gaussian state, can be estimated. This includes in particular the interesting case of mixed Gaussian states. We apply the formula to the problems of estimating phase, purity, loss, amplitude, and squeezing. In the case of the simultaneous measurement of several parameters, we provide the full quantum Fisher information matrix. Our results unify previously known partial results, and constitute a complete solution to the problem of knowing the best possible sensitivity of measurements based on a single-mode Gaussian state.

quant-ph↗

arXiv2012

Real-time distance measurement immune from atmospheric parameters using optical frequency combs

We propose a direct and real-time ranging scheme using an optical frequency combs, able to compensate optically for index of refraction variations due to atmospheric parameters. This scheme could be useful for applications requiring stringent precision over a long distance in air, a situation where dispersion becomes the main limitation. The key ingredient is the use of a mode-locked laser as a precise source for multi-wavelength interferometry in a homodyne detection scheme. By shaping temporally the local oscillator, one can directly access the desired parameter (distance) while being insensitive to fluctuations induced by parameters of the environment such as pressure, temperature, humidity and CO$_2$ content.

quant-ph↗