arXiv subjects

Jiawei Li

Publications and source records attributed to Jiawei Li.

At least 91 records · Page 5Linked to original sources

arXiv2026

Collaborative Belief Reasoning with LLMs for Efficient Multi-Agent Collaboration

Effective real-world multi-agent collaboration requires not only accurate planning but also the ability to reason about collaborators' intents--a crucial capability for avoiding miscoordination and redundant communication under partial observable environments. Due to their strong planning and reasoning capabilities, large language models (LLMs) have emerged as promising autonomous agents for collaborative task solving. However, existing collaboration frameworks for LLMs overlook their reasoning potential for dynamic intent inference, and thus produce inconsistent plans and redundant communication, reducing collaboration efficiency. To bridge this gap, we propose CoBel-World, a novel framework that equips LLM agents with a Collaborative Belief World--an internal representation jointly modeling the physical environment and collaborators' mental states. CoBel-World enables agents to parse external open-world knowledge into structured beliefs via a symbolic belief representation module, and perform zero-shot Bayesian-style belief updates through LLM reasoning. This allows agents to proactively detect potential miscoordination (e.g., conflicting plans) and communicate adaptively. Evaluated on challenging embodied benchmarks (i.e., TDW-MAT and C-WAH), CoBel-World significantly reduces communication costs by 64-79% and improves task completion efficiency by 4-28% compared to the strongest baseline. Our results show that explicit, intent-aware belief modeling is essential for efficient and human-like collaboration in LLM-based multi-agent systems.

cs.AI↗

arXiv2026

MiMo-V2-Flash Technical Report

We present MiMo-V2-Flash, a Mixture-of-Experts (MoE) model with 309B total parameters and 15B active parameters, designed for fast, strong reasoning and agentic capabilities. MiMo-V2-Flash adopts a hybrid attention architecture that interleaves Sliding Window Attention (SWA) with global attention, with a 128-token sliding window under a 5:1 hybrid ratio. The model is pre-trained on 27 trillion tokens with Multi-Token Prediction (MTP), employing a native 32k context length and subsequently extended to 256k. To efficiently scale post-training compute, MiMo-V2-Flash introduces a novel Multi-Teacher On-Policy Distillation (MOPD) paradigm. In this framework, domain-specialized teachers (e.g., trained via large-scale reinforcement learning) provide dense and token-level reward, enabling the student model to perfectly master teacher expertise. MiMo-V2-Flash rivals top-tier open-weight models such as DeepSeek-V3.2 and Kimi-K2, despite using only 1/2 and 1/3 of their total parameters, respectively. During inference, by repurposing MTP as a draft model for speculative decoding, MiMo-V2-Flash achieves up to 3.6 acceptance length and 2.6x decoding speedup with three MTP layers. We open-source both the model weights and the three-layer MTP weights to foster open research and community collaboration.

cs.CL↗

arXiv2025

Splatwizard: A Benchmark Toolkit for 3D Gaussian Splatting Compression

The recent advent of 3D Gaussian Splatting (3DGS) has marked a significant breakthrough in real-time novel view synthesis. However, the rapid proliferation of 3DGS-based algorithms has created a pressing need for standardized and comprehensive evaluation tools, especially for compression task. Existing benchmarks often lack the specific metrics necessary to holistically assess the unique characteristics of different methods, such as rendering speed, rate distortion trade-offs memory efficiency, and geometric accuracy. To address this gap, we introduce Splatwizard, a unified benchmark toolkit designed specifically for benchmarking 3DGS compression models. Splatwizard provides an easy-to-use framework to implement new 3DGS compression model and utilize state-of-the-art techniques proposed by previous work. Besides, an integrated pipeline that automates the calculation of key performance indicators, including image-based quality metrics, chamfer distance of reconstruct mesh, rendering frame rates, and computational resource consumption is included in the framework as well. Code is available at https://github.com/splatwizard/splatwizard

cs.CV↗

arXiv2025

MiMo-Audio: Audio Language Models are Few-Shot Learners

Existing audio language models typically rely on task-specific fine-tuning to accomplish particular audio tasks. In contrast, humans are able to generalize to new audio tasks with only a few examples or simple instructions. GPT-3 has shown that scaling next-token prediction pretraining enables strong generalization capabilities in text, and we believe this paradigm is equally applicable to the audio domain. By scaling MiMo-Audio's pretraining data to over one hundred million of hours, we observe the emergence of few-shot learning capabilities across a diverse set of audio tasks. We develop a systematic evaluation of these capabilities and find that MiMo-Audio-7B-Base achieves SOTA performance on both speech intelligence and audio understanding benchmarks among open-source models. Beyond standard metrics, MiMo-Audio-7B-Base generalizes to tasks absent from its training data, such as voice conversion, style transfer, and speech editing. MiMo-Audio-7B-Base also demonstrates powerful speech continuation capabilities, capable of generating highly realistic talk shows, recitations, livestreaming and debates. At the post-training stage, we curate a diverse instruction-tuning corpus and introduce thinking mechanisms into both audio understanding and generation. MiMo-Audio-7B-Instruct achieves open-source SOTA on audio understanding benchmarks (MMSU, MMAU, MMAR, MMAU-Pro), spoken dialogue benchmarks (Big Bench Audio, MultiChallenge Audio) and instruct-TTS evaluations, approaching or surpassing closed-source models. Model checkpoints and full evaluation suite are available at https://github.com/XiaomiMiMo/MiMo-Audio.

cs.CL↗

arXiv2025

One Perturbation is Enough: On Generating Universal Adversarial Perturbations against Vision-Language Pre-training Models

Vision-Language Pre-training (VLP) models have exhibited unprecedented capability in many applications by taking full advantage of the multimodal alignment. However, previous studies have shown they are vulnerable to maliciously crafted adversarial samples. Despite recent success, these methods are generally instance-specific and require generating perturbations for each input sample. In this paper, we reveal that VLP models are also vulnerable to the instance-agnostic universal adversarial perturbation (UAP). Specifically, we design a novel Contrastive-training Perturbation Generator with Cross-modal conditions (C-PGC) to achieve the attack. In light that the pivotal multimodal alignment is achieved through the advanced contrastive learning technique, we devise to turn this powerful weapon against themselves, i.e., employ a malicious version of contrastive learning to train the C-PGC based on our carefully crafted positive and negative image-text pairs for essentially destroying the alignment relationship learned by VLP models. Besides, C-PGC fully utilizes the characteristics of Vision-and-Language (V+L) scenarios by incorporating both unimodal and cross-modal information as effective guidance. Extensive experiments show that C-PGC successfully forces adversarial samples to move away from their original area in the VLP model's feature space, thus essentially enhancing attacks across various victim models and V+L tasks. The GitHub repository is available at https://github.com/ffhibnese/CPGC_VLP_Universal_Attacks.

cs.CV↗

arXiv2025

The Query Complexity of Local Search and Brouwer in Rounds

We consider the query complexity of finding a local minimum of a function defined on a graph. This abstract problem is fundamental to many optimization tasks, such as finding a local minimum of the loss function when training deep neural networks. In such applications, each query is an expensive loss evaluation, making it crucial to parallelize computations. This motivates our study of local search where at most $k$ rounds of interaction (aka adaptivity) with the oracle are allowed. We focus on the $d$-dimensional grid $\{1, 2, \ldots, n \}^d$, where the dimension $d \geq 2$ is a constant. Our main contribution is to give algorithms and lower bounds that characterize the query complexity of finding a local minimum in $k$ rounds, when $k$ is constant and polynomial in $n$, respectively. Our proof technique for lower bounding the query complexity in rounds may be of independent interest as an alternative to the classical relational adversary method of Aaronson from the fully adaptive setting. The local search analysis also enables us to characterize the query complexity of computing a Brouwer fixed point in rounds.

cs.DS↗

arXiv2025

Total Search Problems in $\mathsf{ZPP}$

We initiate a systematic study of ${\sf TFZPP}$, the class of total ${\sf NP}$ search problems solvable by polynomial time randomized algorithms. ${\sf TFZPP}$ contains a variety of important search problems such as $\text{Bertrand-Chebyshev}$ (finding a prime between $N$ and $2N$), refuter problems for many circuit lower bounds, and $\text{Lossy-Code}$. The $\text{Lossy-Code}$ problem has found prominence due to its fundamental connections to derandomization, catalytic computing, and the metamathematics of complexity theory, among other areas. While ${\sf TFZPP}$ collapses to ${\sf FP}$ under standard derandomization assumptions in the white-box setting, we are able to separate ${\sf TFZPP}$ from the major ${\sf TFNP}$ subclasses in the black-box setting. In fact, we are able to separate it from every uniform ${\sf TFNP}$ class assuming that ${\sf NP}$ is not in quasi-polynomial time. To do so, we extend the connection between proof complexity and black-box ${\sf TFNP}$ to randomized proof systems and randomized reductions. Next, we turn to developing a taxonomy of ${\sf TFZPP}$ problems. We highlight a problem called $\text{Nephew}$, originating from an infinity axiom in set theory. We show that $\text{Nephew}$ is in $\mathsf{PWPP}\cap \mathsf{TFZPP}$ and conjecture that it is not reducible to $\text{Lossy-Code}$. Intriguingly, except for some artificial examples, most other black-box ${\sf TFZPP}$ problems that we are aware of reduce to $\text{Lossy-Code}$.

cs.CC↗

arXiv2025

Entropy-Guided Reasoning Compression

Large reasoning models have demonstrated remarkable performance on complex reasoning tasks, yet the excessive length of their chain-of-thought outputs remains a major practical bottleneck due to high computation cost and poor deployability. Existing compression methods have achieved partial success but overlook a crucial phenomenon in the training process -- the entropy conflict. During compression training, entropy decreases, leading to shorter reasoning but limited exploration, while accuracy-oriented objectives increase entropy, lengthening reasoning chains. This can cause the model to get stuck in a local dilemma. Our analysis further reveals the origin of the entropy conflict: many high-entropy tokens are logical connectors that receive larger gradients and are encouraged under the performance objective, while the compression objective simultaneously penalizes these potentially redundant connectors. This opposing pressure creates a direct source of entropy conflict. To address these issues, we adopt an entropy-guided training framework. As entropy descends, the model is guided toward efficient reasoning by encouraging concise thought steps; as entropy rises, exploration is reinforced under the compact reasoning mode to improve robustness. Experiments on six mathematical benchmarks show that our method compresses reasoning length to 20% of the original while maintaining or even surpassing baseline accuracy. Code and models will be released publicly.

cs.CL↗

arXiv2025

CODE-II: A large-scale dataset for artificial intelligence in ECG analysis

Data-driven methods for electrocardiogram (ECG) interpretation are rapidly progressing. Large datasets have enabled advances in artificial intelligence (AI) based ECG analysis, yet limitations in annotation quality, size, and scope remain major challenges. Here we present CODE-II, a large-scale real-world dataset of 2,735,269 12-lead ECGs from 2,093,807 adult patients collected by the Telehealth Network of Minas Gerais (TNMG), Brazil. Each exam was annotated using standardized diagnostic criteria and reviewed by cardiologists. A defining feature of CODE-II is a set of 66 clinically meaningful diagnostic classes, developed with cardiologist input and routinely used in telehealth practice. We additionally provide an open available subset: CODE-II-open, a public subset of 15,000 patients, and the CODE-II-test, a non-overlapping set of 8,475 exams reviewed by multiple cardiologists for blinded evaluation. A neural network pre-trained on CODE-II achieved superior transfer performance on external benchmarks (PTB-XL and CPSC 2018) and outperformed alternatives trained on larger datasets.

eess.SP↗

arXiv2025

FaSDiff: Balancing Perception and Semantics in Face Compression via Stable Diffusion Priors

With the increasing deployment of facial image data across a wide range of applications, efficient compression tailored to facial semantics has become critical for both storage and transmission. While recent learning-based face image compression methods have achieved promising results, they often suffer from degraded reconstruction quality at low bit rates. Directly applying diffusion-based generative priors to this task leads to suboptimal performance in downstream machine vision tasks, primarily due to poor preservation of high-frequency details. In this work, we propose FaSDiff (\textbf{Fa}cial Image Compression with a \textbf{S}table \textbf{Diff}usion Prior), a novel diffusion-driven compression framework designed to enhance both visual fidelity and semantic consistency. FaSDiff incorporates a high-frequency-sensitive compressor to capture fine-grained details and generate robust visual prompts for guiding the diffusion model. To address low-frequency degradation, we further introduce a hybrid low-frequency enhancement module that disentangles and preserves semantic structures, enabling stable modulation of the diffusion prior during reconstruction. By jointly optimizing perceptual quality and semantic preservation, FaSDiff effectively balances human visual fidelity and machine vision accuracy. Extensive experiments demonstrate that FaSDiff outperforms state-of-the-art methods in both perceptual metrics and downstream task performance.

cs.CV↗

arXiv2025

Large model retrieval enhancement framework for construction site risk identification

This study addresses construction site hazard identification by proposing a retrieval-augmented framework that enhances large language models (LLMs) without requiring fine-tuning. Current LLM-based approaches face limitations: image-text matching struggles with complex hazards, while instruction tuning lacks generalization and is resource-intensive. Our method dynamically integrates external knowledge and retrieved similar cases via prompt tuning, overcoming LLMs' limitations in domain knowledge and feature correlation. The framework comprises a case database, an image retrieval module, and an LLM-based reasoning module. Evaluated on real-site data, our approach boosted GLM-4V's accuracy to 50%, a 35.49% improvement over baselines, with consistent gains across hazard types. Ablation studies validated the effectiveness of our image retrieval strategy, showing the superiority of our LPIPS- and CLIP-based method. The proposed technique significantly improves identification accuracy and contextual understanding, demonstrating strong generalization and offering a practical path for intelligent safety risk detection in construction.

cs.AI↗

arXiv2025

Delta-Influence: Unlearning Poisons via Influence Functions

Addressing data integrity challenges, such as unlearning the effects of data poisoning after model training, is necessary for the reliable deployment of machine learning models. State-of-the-art influence functions, such as EK-FAC and TRAK, often fail to accurately attribute abnormal model behavior to the specific poisoned training data responsible for the data poisoning attack. In addition, traditional unlearning algorithms often struggle to effectively remove the influence of poisoned samples, particularly when only a few affected examples can be identified. To address these challenge, we introduce $Δ$-Influence, a novel approach that leverages influence functions to trace abnormal model behavior back to the responsible poisoned training data using as little as just one poisoned test example. $Δ$-Influence applies data transformations that sever the link between poisoned training data and compromised test points without significantly affecting clean data. This allows $Δ$-Influence to detect large negative shifts in influence scores following data transformations, a phenomenon we term as influence collapse, thereby accurately identifying poisoned training data. Unlearning this subset, e.g. through retraining, effectively eliminates the data poisoning. We validate our method across three vision-based poisoning attacks and three datasets, benchmarking against five detection algorithms and five unlearning strategies. We show that $Δ$-Influence consistently achieves the best unlearning across all settings, showing the promise of influence functions for corrective unlearning. Our code is publicly available at: https://github.com/Ruby-a07/delta-influence

cs.CV↗

arXiv2025

SwarmSys: Decentralized Swarm-Inspired Agents for Scalable and Adaptive Reasoning

Large language model (LLM) agents have shown remarkable reasoning abilities. However, existing multi-agent frameworks often rely on fixed roles or centralized control, limiting scalability and adaptability in long-horizon reasoning. We introduce SwarmSys, a closed-loop framework for distributed multi-agent reasoning inspired by swarm intelligence. Coordination in SwarmSys emerges through iterative interactions among three specialized roles, Explorers, Workers, and Validators, that continuously cycle through exploration, exploitation, and validation. To enable scalable and adaptive collaboration, we integrate adaptive agent and event profiles, embedding-based probabilistic matching, and a pheromone-inspired reinforcement mechanism, supporting dynamic task allocation and self-organizing convergence without global supervision. Across symbolic reasoning, research synthesis, and scientific programming tasks, SwarmSys consistently outperforms baselines, improving both accuracy and reasoning stability. These findings highlight swarm-inspired coordination as a promising paradigm for scalable, robust, and adaptive multi-agent reasoning, suggesting that coordination scaling may rival model scaling in advancing LLM intelligence.

cs.AI↗

arXiv2025

An all-sky 3D dust map Based on Gaia and LAMOST

We present a comprehensive 3D dust reddening map covering the entire Milky Way, constructed by combining reddening estimates based on LAMOST low-resolution spectra (E(B$-$V)$_{\rm LAMOST}$) with those derived from $Gaia$ XP spectra (E(B$-$V)$_{\rm XP}$), along with revised $Gaia$ distances. E(B$-$V)$_{\rm LAMOST}$ values of $\sim$ 4.6 million unique sources were obtained with the standard-pair analysis using LAMOST DR11 stellar parameters and synthesized $B/V$-band photometry from $Gaia$ XP spectra, showing a typical precision of $\sim$ 0.01 mag. The E(B$-$V)$_{\rm XP}$ from the catalog of \citet{zhang2023}, which was derived using forward modeling of $Gaia$ XP spectra, were cross-validated with E(B$-$V)$_{\rm LAMOST}$, leading to the selection of $\sim$ 150 million high-reliability measurements. The combined dataset achieves a median precision of $\sim$ 0.03 mag for E(B$-$V). To model the reddening -- distance relationship along various lines-of-sight, we implemented a parametric approach that accounts for contributions from the local bubble, diffuse interstellar-medium, and multiple potential molecular clouds. The sky was adaptively partitioned based on stellar density, resulting in angular resolutions ranging from 3.4$^{\prime}$ to 58$^{\prime}$, with about half of the sky having a resolution better than 6.9$^{\prime}$. The reddening precision of our 3D map for individual stars reaches $\sim$ 0.01 mag in most regions at $|b| > 20^\circ$, but degrades to 0.01-0.05 mag at $|b| < 20^\circ$. The map reaches a maximum distance of 3-5 kpc in high-extinction regions with $|b| < 5^\circ$, and extends to 10-15 kpc elsewhere. An interactive platform and Python package have been developed for utilization of the 3D dust map. Available online: https://nadc.china-vo.org/data/dustmaps/.

astro-ph.GA↗

arXiv2025

Improving Spatial Resolution of Background Oriented Schlieren Based on Directional Rays

The background-oriented Schlieren technique has emerged as a promising method for visualizing density gradients and performing quantitative measurements. However, an inherent constraint of BOS is the compromise between spatial resolution and measurement sensitivity, as the BOS camera typically remains focused on the background pattern. To overcome the resolution-sensitivity constraint, a new variant of BOS based on nominally directional rays has been proposed in this paper. Instead of utilizing diffusively reflective background patterns, a spherically concave mirror etched with random dots has been used to create a dotted background that reflects rays directionally. Combined with coaxial LED light illumination, we demonstrate that the current setup can improve the spatial resolution of canonical BOS without compromising measurement sensitivity. Moreover, the proposed setup decouples the requirement of a small lens aperture to achieve a large depth of field, thereby significantly alleviating the need for strong background light illumination in high-speed BOS applications. To demonstrate the effectiveness of the proposed method in improving the BOS spatial resolution, both synthetic BOS image generations and experiments on low- and high-speed jets are conducted. Results show that the proposed variant of BOS can be advantageous for measuring density-varying flows with a limited field of view.

physics.flu-dyn↗

arXiv2025

Probabilistic well-posedness of dispersive PDEs beyond variance blowup I: Benjamin-Bona-Mahony equation

We investigate a possible extension of probabilistic well-posedness theory of nonlinear dispersive PDEs with random initial data beyond variance blowup. As a model equation, we study the Benjamin-Bona-Mahony equation (BBM) with Gaussian random initial data. By introducing a suitable vanishing multiplicative renormalization constant on the initial data, we show that solutions to BBM with the renormalized Gaussian random initial data beyond variance blowup converge in law to a solution to the stochastic BBM forced by the derivative of a spatial white noise. By considering alternative renormalization, we show that solutions to the renormalized BBM with the frequency-truncated Gaussian initial data converges in law to a solution to the linear stochastic BBM with the full Gaussian initial data, forced by the derivative of a spatial white noise. This latter result holds for the Gaussian random initial data of arbitrarily low regularity. We also establish analogous results for the stochastic BBM forced by a fractional derivative of a space-time white noise.

math.AP↗

arXiv2025

INDS: Incremental Named Data Streaming for Real-Time Point Cloud Video

Real-time streaming of point cloud video, characterized by massive data volumes and high sensitivity to packet loss, remains a key challenge for immersive applications under dynamic network conditions. While connection-oriented protocols such as TCP and more modern alternatives like QUIC alleviate some transport-layer inefficiencies, including head-of-line blocking, they still retain a coarse-grained, segment-based delivery model and a centralized control loop that limit fine-grained adaptation and effective caching. We introduce INDS (Incremental Named Data Streaming), an adaptive streaming framework based on Information-Centric Networking (ICN) that rethinks delivery for hierarchical, layered media. INDS leverages the Octree structure of point cloud video and expressive content naming to support progressive, partial retrieval of enhancement layers based on consumer bandwidth and decoding capability. By combining time-windows with Group-of-Frames (GoF), INDS's naming scheme supports fine-grained in-network caching and facilitates efficient multi-user data reuse. INDS can be deployed as an overlay, remaining compatible with QUIC-based transport infrastructure as well as future Media-over-QUIC (MoQ) architectures, without requiring changes to underlying IP networks. Our prototype implementation shows up to 80% lower delay, 15-50% higher throughput, and 20-30% increased cache hit rates compared to state-of-the-art DASH-style systems. Together, these results establish INDS as a scalable, cache-friendly solution for real-time point cloud streaming under variable and lossy conditions, while its compatibility with MoQ overlays further positions it as a practical, forward-compatible architecture for emerging immersive media systems.

cs.MM↗

arXiv2025

MonoPartNeRF:Human Reconstruction from Monocular Video via Part-Based Neural Radiance Fields

In recent years, Neural Radiance Fields (NeRF) have achieved remarkable progress in dynamic human reconstruction and rendering. Part-based rendering paradigms, guided by human segmentation, allow for flexible parameter allocation based on structural complexity, thereby enhancing representational efficiency. However, existing methods still struggle with complex pose variations, often producing unnatural transitions at part boundaries and failing to reconstruct occluded regions accurately in monocular settings. We propose MonoPartNeRF, a novel framework for monocular dynamic human rendering that ensures smooth transitions and robust occlusion recovery. First, we build a bidirectional deformation model that combines rigid and non-rigid transformations to establish a continuous, reversible mapping between observation and canonical spaces. Sampling points are projected into a parameterized surface-time space (u, v, t) to better capture non-rigid motion. A consistency loss further suppresses deformation-induced artifacts and discontinuities. We introduce a part-based pose embedding mechanism that decomposes global pose vectors into local joint embeddings based on body regions. This is combined with keyframe pose retrieval and interpolation, along three orthogonal directions, to guide pose-aware feature sampling. A learnable appearance code is integrated via attention to model dynamic texture changes effectively. Experiments on the ZJU-MoCap and MonoCap datasets demonstrate that our method significantly outperforms prior approaches under complex pose and occlusion conditions, achieving superior joint alignment, texture fidelity, and structural continuity.

cs.CV↗