Searcharxiv⌕ Search

arXiv subjects

Hao Wang

Publications and source records attributed to Hao Wang.

At least 451 records · Page 25Linked to original sources
arXiv2026

Partially Recentralization Softmax Loss for Vision-Language Models Robustness

As Large Language Models make a breakthrough in natural language processing tasks (NLP), multimodal technique becomes extremely popular. However, it has been shown that multimodal NLP are vulnerable to adversarial attacks, where the outputs of a model can be dramatically changed by a perturbation to the input. While several defense techniques have been proposed both in computer vision and NLP models, the multimodal robustness of models have not been fully explored. In this paper, we study the adversarial robustness provided by modifying loss function of pre-trained multimodal models, by restricting top K softmax outputs. Based on the evaluation and scoring, our experiments show that after a fine-tuning, adversarial robustness of pre-trained models can be significantly improved, against popular attacks. Further research should be studying, such as output diversity, generalization and the robustness-performance trade-off of this kind of loss functions. Our code will be available after this paper is accepted

cs.CL↗
arXiv2026

LongStream: Long-Sequence Streaming Autoregressive Visual Geometry

Long-sequence streaming 3D reconstruction remains a significant open challenge. Existing autoregressive models often fail when processing long sequences because they anchor poses to the first frame, leading to attention decay, scale drift, and extrapolation errors. We introduce LongStream, a novel gauge-decoupled streaming visual geometry model for metric-scale scene reconstruction across thousands of frames under a strictly online, future-invisible setting. Our approach is threefold. First, we discard the first-frame anchor and predict keyframe-relative poses. This reformulates long-range extrapolation into a constant-difficulty local task. Second, we introduce orthogonal scale learning. This method fully disentangles geometry from scale estimation to suppress drift. Finally, we identify attention bias issues in Transformers, including attention-sink reliance and long-term KV-cache saturation. We propose cache-consistent training combined with periodic cache refresh. This approach suppresses attention biases and contamination over ultra-long sequences and reduces the gap between training and inference. Experiments show that LongStream achieves state-of-the-art performance, enabling stable, metric-scale reconstruction over kilometer-scale sequences at 18 FPS. Project Page: https://3dagentworld.github.io/longstream/

cs.CV↗
arXiv2026

Learning Actionable Manipulation Recovery via Counterfactual Failure Synthesis

While recent foundation models have significantly advanced robotic manipulation, these systems still struggle to autonomously recover from execution errors. Current failure-learning paradigms rely on either costly and unsafe real-world data collection or simulator-based perturbations, which introduce a severe sim-to-real gap. Furthermore, existing visual analyzers predominantly output coarse, binary diagnoses rather than the executable, trajectory-level corrections required for actual recovery. To bridge the gap between failure diagnosis and actionable recovery, we introduce Dream2Fix, a framework that synthesizes photorealistic, counterfactual failure rollouts directly from successful real-world demonstrations. By perturbing actions within a generative world model, Dream2Fix creates paired failure-correction data without relying on simulators. To ensure the generated data is physically viable for robot learning, we implement a structured verification mechanism that strictly filters rollouts for task validity, visual coherence, and kinematic safety. This engine produces a high-fidelity dataset of over 120k paired samples. Using this dataset, we fine-tune a vision-language model to jointly predict failure types and precise recovery trajectories, mapping visual anomalies directly to corrective actions. Extensive real-world robotic experiments show our approach achieves state-of-the-art correction accuracy, improving from 19.7% to 81.3% over prior baselines, and successfully enables zero-shot closed-loop failure recovery in physical deployments.

cs.RO↗
arXiv2026

VistaWise: Building Cost-Effective Agent with Cross-Modal Knowledge Graph for Minecraft

Large language models (LLMs) have shown significant promise in embodied decision-making tasks within virtual open-world environments. Nonetheless, their performance is hindered by the absence of domain-specific knowledge. Methods that finetune on large-scale domain-specific data entail prohibitive development costs. This paper introduces VistaWise, a cost-effective agent framework that integrates cross-modal domain knowledge and finetunes a dedicated object detection model for visual analysis. It reduces the requirement for domain-specific training data from millions of samples to a few hundred. VistaWise integrates visual information and textual dependencies into a cross-modal knowledge graph (KG), enabling a comprehensive and accurate understanding of multimodal environments. We also equip the agent with a retrieval-based pooling strategy to extract task-related information from the KG, and a desktop-level skill library to support direct operation of the Minecraft desktop client via mouse and keyboard inputs. Experimental results demonstrate that VistaWise achieves state-of-the-art performance across various open-world tasks, highlighting its effectiveness in reducing development costs while enhancing agent performance.

cs.AI↗
arXiv2026

FedMomentum: Preserving LoRA Training Momentum in Federated Fine-Tuning

Federated fine-tuning of large language models (LLMs) with low-rank adaptation (LoRA) offers a communication-efficient and privacy-preserving solution for task-specific adaptation. Naive aggregation of LoRA modules introduces noise due to mathematical incorrectness when averaging the downsampling and upsampling matrices independently. However, existing noise-free aggregation strategies inevitably compromise the structural expressiveness of LoRA, limiting its ability to retain client-specific adaptations by either improperly reconstructing the low-rank structure or excluding partially trainable components. We identify this problem as loss of training momentum, where LoRA updates fail to accumulate effectively across rounds, resulting in slower convergence and suboptimal performance. To address this, we propose FedMomentum, a novel framework that enables structured and momentum-preserving LoRA aggregation via singular value decomposition (SVD). Specifically, after aggregating low-rank updates in a mathematically correct manner, FedMomentum applies SVD to extract the dominant components that capture the main update directions. These components are used to reconstruct the LoRA modules with the same rank, while residual components can be retained and later merged into the backbone to preserve semantic information and ensure robustness. Extensive experiments across multiple tasks demonstrate that FedMomentum consistently outperforms prior state-of-the-art methods in convergence speed and final accuracy.

cs.LG↗
arXiv2026

Central Limits via Dilated Categories

The Central Limit Theorem (CLT) establishes that sufficiently large sequences of independent and identically distributed random variables converge in probability to a normal distribution. This makes the CLT a fundamental building block of statistical reasoning and, by extension, in reasoning about computing systems that are based on statistical inference such as probabilistic programing languages, programs with optimisation, and machine learning components. However, there is no general theory of CLT-like results currently, which forces practitioners to redo proofs without having a good handle on the essential ingredients of CLT-type results. In this paper, we introduce dilated seminorm-enriched category theory as a unifying framework for central limits, and we establish an abstract central limit theorem within that framework. We illustrate how a strengthened version of the classical CLT and the law of large numbers can be obtained as instances of our framework. Moreover, we derive from our framework a novel central limit theorem for symplectic manifolds, the CLT for observables, which finds applications in statistical mechanics.

cs.LO↗
arXiv2026

Weyl-Transition-Driven Giant Reversible Orbital Hall Conductivity

Orbital Hall conductivity (OHC) is a central ingredient of orbitronics, yet how to control it microscopically remains largely unexplored. Here we identify a general mechanism in which tilted Weyl crossings formed by orbitally distinct bands generate a strongly asymmetric orbital Berry curvature (OBC) distribution, whose imbalance survives Brillouin-zone integration and yields a sizable OHC already at zeroth order. Using first-principles calculations, we show that monolayer PtBi2 realizes this mechanism and hosts a giant OHC dominated by a type-II Weyl point. A small biaxial tensile strain drives a type-II $\rightarrow$ type-I $\rightarrow$ type-II Weyl transition, leading to a reversible sign change of the OHC through the evolution of the OBC imbalance. This process is governed by the chiral orbital texture of the crossing bands and is further assisted by a strain-induced first-order structural phase transition through bonding reconstruction and polarization change. Our results establish Weyl engineering of orbital quantum geometry as a powerful route to generating and reversibly controlling OHC in polar multi-orbital materials.

cond-mat.mtrl-sci↗
arXiv2026

BrainCast: A Spatio-Temporal Forecasting Model for Whole-Brain fMRI Time Series Prediction

Functional magnetic resonance imaging (fMRI) enables noninvasive investigation of brain function, while short clinical scan durations, arising from human and non-human factors, usually lead to reduced data quality and limited statistical power for neuroimaging research. In this paper, we propose BrainCast, a novel spatio-temporal forecasting framework specifically tailored for whole-brain fMRI time series forecasting, to extend informative fMRI time series without additional data acquisition. It formulates fMRI time series forecasting as a multivariate time series prediction task and jointly models temporal dynamics within regions of interest (ROIs) and spatial interactions across ROIs. Specifically, BrainCast integrates a Spatial Interaction Awareness module to characterize inter-ROI dependencies via embedding every ROI time series as a token, a Temporal Feature Refinement module to capture intrinsic neural dynamics within each ROI by enhancing both low- and high-energy temporal components of fMRI time series at the ROI level, and a Spatio-temporal Pattern Alignment module to combine spatial and temporal representations for producing informative whole-brain features. Experimental results on resting-state and task fMRI datasets from the Human Connectome Project demonstrate the superiority of BrainCast over state-of-the-art time series forecasting baselines. Moreover, fMRI time series extended by BrainCast improve downstream cognitive ability prediction, highlighting the clinical and neuroscientific impact brought by whole-brain fMRI time series forecasting in scenarios with restricted scan durations.

cs.CV↗
arXiv2026

Efficient Personalized Reranking with Semi-Autoregressive Generation and Online Knowledge Distillation

Generative models offer a promising paradigm for the final stage reranking in multi-stage recommender systems, with the ability to capture inter-item dependencies within reranked lists. However, their practical deployment still faces two key challenges: (1) an inherent conflict between achieving high generation quality and ensuring low-latency inference, making it difficult to balance the two, and (2) insufficient interaction between user and item features in existing methods. To address these challenges, we propose a novel Personalized Semi-Autoregressive with online knowledge Distillation (PSAD) framework for reranking. In this framework, the teacher model adopts a semi-autoregressive generator to balance generation quality and efficiency, while its ranking knowledge is distilled online into a lightweight scoring network during joint training, enabling real-time and efficient inference. Furthermore, we propose a User Profile Network (UPN) that injects user intent and models interest dynamics, enabling deeper interactions between users and items. Extensive experiments conducted on three large-scale public datasets demonstrate that PSAD significantly outperforms state-of-the-art baselines in both ranking performance and inference efficiency.

cs.IR↗
arXiv2026

Exploring Diffusion Models' Corruption Stage in Few-Shot Fine-tuning and Mitigating with Bayesian Neural Networks

Few-shot fine-tuning of Diffusion Models (DMs) is a key advancement, significantly reducing training costs and enabling personalized AI applications. However, we explore the training dynamics of DMs and observe an unanticipated phenomenon: during the training process, image fidelity initially improves, then unexpectedly deteriorates with the emergence of noisy patterns, only to recover later with severe overfitting. We term the stage with generated noisy patterns as corruption stage. To understand this corruption stage, we begin by theoretically modeling the one-shot fine-tuning scenario, and then extend this modeling to more general cases. Through this modeling, we identify the primary cause of this corruption stage: a narrowed learning distribution inherent in the nature of few-shot fine-tuning. To tackle this, we apply Bayesian Neural Networks (BNNs) on DMs with variational inference to implicitly broaden the learned distribution, and present that the learning target of the BNNs can be naturally regarded as an expectation of the diffusion loss and a further regularization with the pretrained DMs. This approach is highly compatible with current few-shot fine-tuning methods in DMs and does not introduce any extra inference costs. Experimental results demonstrate that our method significantly mitigates corruption, and improves the fidelity, quality and diversity of the generated images in both object-driven and subject-driven generation tasks. Code is available at https://github.com/Nicholas0228/BNN-Finetuning-DMs.

cs.CV↗
arXiv2026

MultiGO++: Monocular 3D Clothed Human Reconstruction via Geometry-Texture Collaboration

Monocular 3D clothed human reconstruction aims to generate a complete and realistic textured 3D avatar from a single image. Existing methods are commonly trained under multi-view supervision with annotated geometric priors, and during inference, these priors are estimated by the pre-trained network from the monocular input. These methods are constrained by three key limitations: texturally by unavailability of training data, geometrically by inaccurate external priors, and systematically by biased single-modality supervision, all leading to suboptimal reconstruction. To address these issues, we propose a novel reconstruction framework, named MultiGO++, which achieves effective systematic geometry-texture collaboration. It consists of three core parts: (1) A multi-source texture synthesis strategy that constructs 15,000+ 3D textured human scans to improve the performance on texture quality estimation in challenge scenarios; (2) A region-aware shape extraction module that extracts and interacts features of each body region to obtain geometry information and a Fourier geometry encoder that mitigates the modality gap to achieve effective geometry learning; (3) A dual reconstruction U-Net that leverages geometry-texture collaborative features to refine and generate high-fidelity textured 3D human meshes. Extensive experiments on two benchmarks and many in-the-wild cases show the superiority of our method over state-of-the-art approaches.

cs.CV↗
arXiv2026

FSMLP: Modelling Channel Dependencies With Simplex Theory Based Multi-Layer Perceptions In Frequency Domain

Time series forecasting (TSF) plays a crucial role in various domains, including web data analysis, energy consumption prediction, and weather forecasting. While Multi-Layer Perceptrons (MLPs) are lightweight and effective for capturing temporal dependencies, they are prone to overfitting when used to model inter-channel dependencies. In this paper, we investigate the overfitting problem in channel-wise MLPs using Rademacher complexity theory, revealing that extreme values in time series data exacerbate this issue. To mitigate this issue, we introduce a novel Simplex-MLP layer, where the weights are constrained within a standard simplex. This strategy encourages the model to learn simpler patterns and thereby reducing overfitting to extreme values. Based on the Simplex-MLP layer, we propose a novel \textbf{F}requency \textbf{S}implex \textbf{MLP} (FSMLP) framework for time series forecasting, comprising of two kinds of modules: \textbf{S}implex \textbf{C}hannel-\textbf{W}ise MLP (SCWM) and \textbf{F}requency \textbf{T}emporal \textbf{M}LP (FTM). The SCWM effectively leverages the Simplex-MLP to capture inter-channel dependencies, while the FTM is a simple yet efficient temporal MLP designed to extract temporal information from the data. Our theoretical analysis shows that the upper bound of the Rademacher Complexity for Simplex-MLP is lower than that for standard MLPs. Moreover, we validate our proposed method on seven benchmark datasets, demonstrating significant improvements in forecasting accuracy and efficiency, while also showcasing superior scalability. Additionally, we demonstrate that Simplex-MLP can improve other methods that use channel-wise MLP to achieve less overfitting and improved performance. Code are available \href{https://github.com/FMLYD/FSMLP}{\textcolor{red}{here}}.

cs.LG↗
arXiv2026

$\ell_{1\text{-}2}$ Regularization for Sparse Optimization: Consistency and Global Convergence

The $\ell_{1\text{-}2}$ regularization method has a strong sparsity promoting capability in approaching sparse solutions of linear inverse problems and gained successful applications in various mathematics and applied science fields. This paper aims to investigate the consistency theory and global convergent algorithms for the $\ell_{1\text{-}2}$ regularization problem. In the theoretical aspect, we introduce a notion of restricted eigenvalue condition relative to the $\ell_{1\text{-}2}$ penalty, and employ it to establish an oracle property and a recovery bound for the global solution of the $\ell_{1\text{-}2}$ regularization problem. In the algorithmic aspect, we propose two types of iterative thresholding algorithms with the truncation technique and the continuation technique, respectively, to solve the $\ell_{1\text{-}2}$ regularization problem. Moreover, under the assumption of the well-known restricted isometry property, we establish the convergence of the proposed algorithms to the ground true sparse solution within a tolerance relevant to the noise level and the recovery bound. Preliminary numerical results show that our proposed algorithms can approach the ground true sparse solution and significantly enhance the sparsity recovery capability, compared with the popular sparse optimization algorithms in the literature.

math.OC↗
arXiv2026

Changing the Game: The Bounce-Bind Ising Machine

The Ising model, originally proposed a century ago, has become a cornerstone of combinatorial optimization in recent decades. However, Ising machines remain constrained by a fundamental hardware-speed trade-off. We introduce the Bounce-Bind Ising Machine (BBIM), a mechanism with a single tunable parameter that modulates spin dynamics without altering the energy landscape, building upon the classic golf-ball analogy but replacing it with a dynamic tennis ball/shot put system. The Bounce mode (accelerating escapes from local minima) and Bind mode (enabling rapid convergence) dynamically balance speed and quality. Benchmarked on dense MAX-CUT (edge density=0.5), BBIM achieves a peak speedup of 6.15 times at n=200. For sparse 3-Regular 3-XORSAT (second-order), the peak speedup reaches 27.3 times at n=160. Both results incur negligible additional hardware resource consumption. This work demonstrates a critical pathway to circumventing the hardware-speed bottleneck and its practical applicability to large-scale optimization hardware, validated on structurally distinct benchmarks.

cs.AR↗
arXiv2026

Emerging trends in Cislunar Space for Lunar Science Exploration and Space Robotics aiding Human Spaceflight Safety

In recent years, the Moon has emerged as an unparalleled extraterrestrial testbed for advancing cuttingedge technological and scientific research critical to enabling sustained human presence on its surface and supporting future interplanetary exploration. This study identifies and investigates two pivotal research domains with substantial transformative potential for accelerating humanity interplanetary aspirations. First is Lunar Science Exploration with Artificial Intelligence and Space Robotics which focusses on AI and Space Robotics redefining the frontiers of space exploration. Second being Space Robotics aid in manned spaceflight to the Moon serving as critical assets for pre-deployment infrastructure development, In-Situ Resource Utilization, surface operations support, and astronaut safety assurance. By integrating autonomy, machine learning, and realtime sensor fusion, space robotics not only augment human capabilities but also serve as force multipliers in achieving sustainable lunar exploration, paving the way for future crewed missions to Mars and beyond.

cs.RO↗
arXiv2026

FuXi-Linear: Unleashing the Power of Linear Attention in Long-term Time-aware Sequential Recommendation

Modern recommendation systems primarily rely on attention mechanisms with quadratic complexity, which limits their ability to handle long user sequences and slows down inference. While linear attention is a promising alternative, existing research faces three critical challenges: (1) temporal signals are often overlooked or integrated via naive coupling that causes mutual interference between temporal and semantic signals while neglecting behavioral periodicity; (2) insufficient positional information provided by existing linear frameworks; and (3) a primary focus on short sequences and shallow architectures. To address these issues, we propose FuXi-Linear, a linear-complexity model designed for efficient long-sequence recommendation. Our approach introduces two key components: (1) a Temporal Retention Channel that independently computes periodic attention weights using temporal data, preventing crosstalk between temporal and semantic signals; (2) a Linear Positional Channel that integrates positional information through learnable kernels within linear complexity. Moreover, we demonstrate that FuXi-Linear exhibits a robust power-law scaling property at a thousand-length scale, a characteristic largely unexplored in prior linear recommendation studies. Extensive experiments on sequences of several thousand tokens demonstrate that FuXi-Linear outperforms state-of-the-art models in recommendation quality, while achieving up to 10$\times$ speedup in the prefill stage and up to 21$\times$ speedup in the decode stage compared to competitive baselines. Our code has been released in a public repository https://github.com/USTC-StarTeam/fuxi-linear.

cs.IR↗
arXiv2026

RLHFless: Serverless Computing for Efficient RLHF

Reinforcement Learning from Human Feedback (RLHF) has been widely applied to Large Language Model (LLM) post-training to align model outputs with human preferences. Recent models, such as DeepSeek-R1, have also shown RLHF's potential to improve LLM reasoning on complex tasks. In RL, inference and training co-exist, creating dynamic resource demands throughout the workflow. Compared to traditional RL, RLHF further challenges training efficiency due to expanding model sizes and resource consumption. Several RLHF frameworks aim to balance flexible abstraction and efficient execution. However, they rely on serverful infrastructures, which struggle with fine-grained resource variability. As a result, during synchronous RLHF training, idle time between or within RL components often causes overhead and resource wastage. To address these issues, we present RLHFless, the first scalable training framework for synchronous RLHF, built on serverless computing environments. RLHFless adapts to dynamic resource demands throughout the RLHF pipeline, pre-computes shared prefixes to avoid repeated computation, and uses a cost-aware actor scaling strategy that accounts for response length variation to find sweet spots with lower cost and higher speed. In addition, RLHFless assigns workloads efficiently to reduce intra-function imbalance and idle time. Experiments on both physical testbeds and a large-scale simulated cluster show that RLHFless achieves up to 1.35x speedup and 44.8% cost reduction compared to the state-of-the-art baseline.

cs.AI↗
arXiv2026

SpikePingpong: Spike Vision-based Fast-Slow Pingpong Robot System

Learning to control high-speed objects in dynamic environments represents a fundamental challenge in robotics. Table tennis serves as an ideal testbed for advancing robotic capabilities in dynamic environments. This task presents two fundamental challenges: it requires a high-precision vision system capable of accurately predicting ball trajectories under complex dynamics, and it necessitates intelligent control strategies to ensure precise ball striking to target regions. High-speed object manipulation typically demands advanced visual perception hardware capable of capturing rapid motion with exceptional temporal resolution. Drawing inspiration from Kahneman's dual-system theory, where fast intuitive processing complements slower deliberate reasoning, there exists an opportunity to develop more robust perception architectures that can handle high-speed dynamics while maintaining accuracy. To this end, we present \textit{\textbf{SpikePingpong}}, a novel system that integrates spike-based vision with imitation learning for high-precision robotic table tennis. We develop a Fast-Slow system architecture where System 1 provides rapid ball detection and preliminary trajectory prediction with millisecond-level responses, while System 2 employs spike-oriented neural calibration for precise hittable position corrections. For strategic ball striking, we introduce Imitation-based Motion Planning And Control Technology, which learns optimal robotic arm striking policies through demonstration-based learning. Experimental results demonstrate that \textit{\textbf{SpikePingpong}} achieves a remarkable 92\% success rate for 30 cm accuracy zones and 70\% in the more challenging 20 cm precision targeting. This work demonstrates the potential of Fast-Slow architectures for advancing robotic capabilities in time-critical manipulation tasks.

cs.RO↗