arXiv subjects

Zekai Wang

Publications and source records attributed to Zekai Wang.

At least 19 recordsLinked to original sources

arXiv2026

Online Reasoning Calibration: Test-Time Training Enables Generalizable Conformal LLM Reasoning

While test-time scaling has enabled large language models to solve highly difficult tasks, state-of-the-art results come at exorbitant compute costs. These inefficiencies can be attributed to the miscalibration of post-trained language models, and the lack of calibration in popular sampling techniques. Here, we present Online Reasoning Calibration (ORCA), a framework for calibrating the sampling process that draws upon conformal prediction and test-time training. Specifically, we introduce a meta-learning procedure that updates the calibration module for each input. This allows us to provide valid confidence estimates under distributional shift, e.g. in thought patterns that occur across different stages of reasoning, or in prompt distributions between model development and deployment. ORCA not only provides theoretical guarantees on conformal risks, but also empirically shows higher efficiency and generalization across different reasoning tasks. At risk level $δ=0.1$, ORCA improves Qwen2.5-32B efficiency on in-distribution tasks with savings up to 47.5% with supervised labels and 40.7% with self-consistency labels. Under zero-shot out-of-domain settings, it improves MATH-500 savings from 24.8% of the static calibration baseline to 67.0% while maintaining a low empirical error rate, and the same trend holds across model families and downstream benchmarks. Our code is publicly available at https://github.com/wzekai99/ORCA.

cs.LG↗

arXiv2026

T-Rex: Tactile-Reactive Dexterous Manipulation

The ability to react dynamically to tactile signals has long been considered crucial to agile human-level dexterity. Yet contemporary learning-based Vision-Language-Action (VLA) models for robotic manipulation generally either overlook the tactile modality or are limited to encoders with static cues, due in part to the scarcity of diverse training data and standardized evaluation, architectural constraints in current VLA models, and limitations of static tactile encoders. In this paper, we push the frontier of tactile-reactive manipulation by addressing all of these limitations. We propose a large-scale, 100-hour tactile-rich dataset collected via a novel, data-efficient recipe that prioritizes elementary motor primitives. To effectively exploit naturally high-frequency touch signals without sacrificing the existing capabilities of existing VLAs, we introduce a variable-rate Mixture-of-Transformers (MoT) architecture equipped with a novel temporal tactile VQ-VAE encoder. We demonstrate the effectiveness of tactile-reactive policies on 12 manipulation tasks requiring delicate force control and deformable object manipulation, achieving over 30% higher average success rate than the strongest baseline.

cs.RO↗

arXiv2026

P-K-GCN: Physics-augmented Koopman-enhanced Graph Convolutional Network for Deep Spatiotemporal Super-resolution

High-fidelity simulation of spatiotemporal dynamics is computationally prohibitive, necessitating efficient super-resolution techniques to reconstruct high-resolution data from coarse-grained inputs. Traditional data-driven methods often lack physical constraints, and simple physics-informed learning struggles with irregular spatial geometries and intricately evolving temporal dynamics. To tackle these challenges, we propose a Physics-augmented Koopman-enhanced Graph Convolutional Network (P-K-GCN) for spatiotemporal super-resolution on irregular geometries. Specifically, a continuous spline-based GCN is first designed to extract spatial dependencies directly from coarse graph, and Koopman operator theory is incorporated to project the nonlinear dynamics into a compact latent space where temporal progression is linearized. Second, we augment the optimization objective with a physics-based loss to force the data-driven reconstructions to adhere to physical laws for improving predictive fidelity and robustness. Finally, we provide a rigorous theoretical analysis, establishing that the physics augmentation and Koopman regularization mathematically guarantees a reduction in super-resolution error by diminishing Rademacher complexity and tightening generalization bounds. We evaluate our framework on reconstructing spatially high-resolution cardiac electrodynamics across a 3D heart geometry from sparse low-resolution measurements. Numerical experiments demonstrate that our method achieves superior accuracy compared to baseline models.

cs.LG↗

arXiv2026

Contrastive Action-Image Pre-training for Visuomotor Control

Existing vision encoders for robotics face a fundamental bottleneck: robotic datasets lack the scale necessary for large-scale pre-training. Prior work circumvents this data scarcity by turning to internet-scale image and language data or egocentric human video. While these models show promise, neither paradigm learns from paired vision and action data, which downstream visuomotor control policies require. However, robot trajectories, the most direct source of this paired signal, are not available at pre-training scale, motivating us to extract action signals from abundant human video instead. To this end, we introduce CAIP (Contrastive Action-Image Pre-training), a vision encoder that treats human hand poses from large-scale egocentric video as a proxy for end-effector actions. By extracting 3D hand keypoints, a representation that aligns naturally with downstream robot action spaces, CAIP learns a unified action-image representation through a contrastive objective. Leveraging 32,041 hours of egocentric human video and only 88 hours of robotic manipulation data, CAIP outperforms state-of-the-art vision encoders including DINOv2, SigLIP, MVP, and R3M. Evaluated on a challenging real-world dexterous manipulation setup using Dexmate Vega and Sharpa Wave hands, CAIP yields performance gains of more than 30% on tasks involving folding, pouring, and fine-grained manipulation. Our results show that our method of contrastive action-centric pre-training yields a scalable path to achieving robust visual representations better suited for physical interaction.

cs.RO↗

arXiv2026

Nanbeige4.1-3B: A Small General Model that Reasons, Aligns, and Acts

We present Nanbeige4.1-3B, a unified generalist language model that simultaneously achieves strong agentic behavior, code generation, and general reasoning with only 3B parameters. To the best of our knowledge, it is the first open-source small language model (SLM) to achieve such versatility in a single model. To improve reasoning and preference alignment, we combine point-wise and pair-wise reward modeling, ensuring high-quality, human-aligned responses. For code generation, we design complexity-aware rewards in Reinforcement Learning, optimizing both correctness and efficiency. In deep search, we perform complex data synthesis and incorporate turn-level supervision during training. This enables stable long-horizon tool interactions, allowing Nanbeige4.1-3B to reliably execute up to 600 tool-call turns for complex problem-solving. Extensive experimental results show that Nanbeige4.1-3B significantly outperforms prior models of similar scale, such as Nanbeige4-3B-2511 and Qwen3-4B, even achieving superior performance compared to much larger models, such as Qwen3-30B-A3B. Our results demonstrate that small models can achieve both broad competence and strong specialization simultaneously, redefining the potential of 3B parameter models.

cs.AI↗

arXiv2026

A Step in Flux to Suppress Axion Isocurvature

The QCD axion in the pre-inflation scenario faces a stringent isocurvature constraint, which requires a relatively low Hubble scale during inflation. If the axion was heavier than the Hubble scale during inflation, its isocurvature is suppressed and the constraint disappears. We point out a novel mechanism for achieving this, relying on the topological nature of a BF-type (monodromy) mass for the axion. Such a mass term has an integer coefficient, so it could naturally have been very large during inflation and exactly zero by the time of the QCD phase transition. This integer can be viewed as a quantized flux, which is discharged in a first-order phase transition that proceeds by the nucleation of charged branes. This mechanism can be embedded in cosmology in several different ways, with tunneling during, at the end of, or after inflation. We provide a detailed case study of the scenario in which the tunneling event occurs during inflation. We also comment briefly on possible UV completions within extra-dimensional gauge theories and string theory. Intriguingly, the phase transition could be accompanied by the emergence of the chiral Standard Model field content from a non-chiral theory during inflation.

hep-ph↗

arXiv2026

Mask-Guided Multi-Task Network for Face Attribute Recognition

Face Attribute Recognition (FAR) plays a crucial role in applications such as person re-identification, face retrieval, and face editing. Conventional multi-task attribute recognition methods often process the entire feature map for feature extraction and attribute classification, which can produce redundant features due to reliance on global regions. To address these challenges, we propose a novel approach emphasizing the selection of specific feature regions for efficient feature learning. We introduce the Mask-Guided Multi-Task Network (MGMTN), which integrates Adaptive Mask Learning (AML) and Group-Global Feature Fusion (G2FF) to address the aforementioned limitations. Leveraging a pre-trained keypoint annotation model and a fully convolutional network, AML accurately localizes critical facial parts (e.g., eye and mouth groups) and generates group masks that delineate meaningful feature regions, thereby mitigating negative transfer from global region usage. Furthermore, G2FF combines group and global features to enhance FAR learning, enabling more precise attribute identification. Extensive experiments on two challenging facial attribute recognition datasets demonstrate the effectiveness of MGMTN in improving FAR performance.

cs.CV↗

arXiv2026

FAR-AMTN: Attention Multi-Task Network for Face Attribute Recognition

To enhance the generalization performance of Multi-Task Networks (MTN) in Face Attribute Recognition (FAR), it is crucial to share relevant information across multiple related prediction tasks effectively. Traditional MTN methods create shared low-level modules and distinct high-level modules, causing an exponential increase in model parameters with the addition of tasks. This approach also limits feature interaction at the high level, hindering the exploration of semantic relations among attributes, thereby affecting generalization negatively. In response, this study introduces FAR-AMTN, a novel Attention Multi-Task Network for FAR. It incorporates a Weight-Shared Group-Specific Attention (WSGSA) module with shared parameters to minimize complexity while improving group feature representation. Furthermore, a Cross-Group Feature Fusion (CGFF) module is utilized to foster interactions between attribute groups, enhancing feature learning. A Dynamic Weighting Strategy (DWS) is also introduced for synchronized task convergence. Experiments on the CelebA and LFWA datasets demonstrate that the proposed FAR-AMTN demonstrates superior accuracy with significantly fewer parameters compared to existing models.

cs.CV↗

arXiv2025

From Generated Human Videos to Physically Plausible Robot Trajectories

Video generation models are rapidly improving in their ability to synthesize human actions in novel contexts, holding the potential to serve as high-level planners for contextual robot control. To realize this potential, a key research question remains open: how can a humanoid execute the human actions from generated videos in a zero-shot manner? This challenge arises because generated videos are often noisy and exhibit morphological distortions that make direct imitation difficult compared to real video. To address this, we introduce a two-stage pipeline. First, we lift video pixels into a 4D human representation and then retarget to the humanoid morphology. Second, we propose GenMimic-a physics-aware reinforcement learning policy conditioned on 3D keypoints, and trained with symmetry regularization and keypoint-weighted tracking rewards. As a result, GenMimic can mimic human actions from noisy, generated videos. We curate GenMimicBench, a synthetic human-motion dataset generated using two video generation models across a spectrum of actions and contexts, establishing a benchmark for assessing zero-shot generalization and policy robustness. Extensive experiments demonstrate improvements over strong baselines in simulation and confirm coherent, physically stable motion tracking on a Unitree G1 humanoid robot without fine-tuning. This work offers a promising path to realizing the potential of video generation models as high-level policies for robot control.

cs.RO↗

arXiv2025

Nanbeige4-3B Technical Report: Exploring the Frontier of Small Language Models

We present Nanbeige4-3B, a family of small-scale but high-performing language models. Pretrained on 23T high-quality tokens and finetuned on over 30 million diverse instructions, we extend the boundary of the scaling law for small language models. In pre-training, we design a Fine-Grained Warmup-Stable-Decay (FG-WSD) training scheduler, which progressively refines data mixtures across stages to boost model performance. In post-training, to improve the quality of the SFT data, we design a joint mechanism that integrates deliberative generation refinement and chain-of-thought reconstruction, yielding substantial gains on complex tasks. Following SFT, we employ our flagship reasoning model to distill Nanbeige4-3B through our proposed Dual Preference Distillation (DPD) method, which leads to further performance gains. Finally, a multi-stage reinforcement learning phase was applied, leveraging verifiable rewards and preference modeling to strengthen abilities on both reasoning and human alignment. Extensive evaluations show that Nanbeige4-3B not only significantly outperforms models of comparable parameter scale but also rivals much larger models across a wide range of benchmarks. The model checkpoints are available at https://huggingface.co/Nanbeige.

cs.CL↗

arXiv2025

reslife: Residual Lifetime Analysis Tool in R

Mean residual lifetime is an important measure utilized in various fields, including pharmaceutical companies, manufacturing companies, and insurance companies for survival analysis. However, the computation of mean residual lifetime can be laborious and challenging. To address this issue, the R package reslife has been developed, which enables efficient calculation of mean residual lifetime based on closed-form solution in a user-friendly manner. reslife offers the capability to utilize either the results of a flexsurv regression or user-provided parameters to compute mean residual lifetime. Furthermore, there are options to return median and percentile residual lifetime. If the user chooses to use the outputs of a flexsurv regression, there is an option to input a data frame with unobserved data. In this article, we present reslife, explain its underlying mathematical principles, illustrate its functioning, and provide examples on how to utilize the package. The aim is to facilitate the use of mean residual lifetime, making it more accessible and efficient for practitioners in various disciplines, particularly those involved in survival analysis within the pharmaceutical industry. This package has been approved and available on CRAN: https://cran.r-project.org/web/packages/reslife/index.html

stat.CO↗

arXiv2025

Unsupervised Identification and Replay-based Detection (UIRD) for New Category Anomaly Detection in ECG Signal

In clinical practice, automatic analysis of electrocardiogram (ECG) is widely applied to identify irregular heart rhythms and other electrical anomalies of the heart, enabling timely intervention and potentially improving clinical outcomes. However, due to the limited samples in certain types of ECG signals, the class imbalance issues pose a challenge for ECG-based detection. In addition, as the volume of patient data grows, long-term storage of all historical data becomes increasingly burdensome as training samples to recognize new patterns and classify existing ECG signals accurately. Therefore, to enhance the performance of anomaly detection while addressing storage limitations, we propose a pseudo-replay based semi-supervised continual learning framework, which consists of two components: unsupervised identification and replay-based detection. For unsupervised identification, an unsupervised generative adversarial network (GAN)-based framework is integrated to detect novel patterns. Besides, instead of directly storing all historical data, a pseudo replay-based learning strategy is proposed which utilizes a generator to learn the data distribution for each individual task. When a new task arises, the generator synthesizes pseudo data representative of previous learnt classes, enabling the model to detect both the existed patterns and the newly presented anomalies. The effectiveness of the proposed framework is validated in four public ECG datasets, which leverages supervised classification problems for anomaly detection. The experimental results show that the developed approach is very promising in identifying novel anomalies while maintaining good performance on detecting existing ECG signals.

cs.LG↗

arXiv2025

Transformer with Koopman-Enhanced Graph Convolutional Network for Spatiotemporal Dynamics Forecasting

Spatiotemporal dynamics forecasting is inherently challenging, particularly in systems defined over irregular geometric domains, due to the need to jointly capture complex spatial correlations and nonlinear temporal dynamics. To tackle these challenges, we propose TK-GCN, a two-stage framework that integrates geometry-aware spatial encoding with long-range temporal modeling. In the first stage, a Koopman-enhanced Graph Convolutional Network (K-GCN) is developed to embed the high-dimensional dynamics distributed on spatially irregular domains into a latent space where the evolution of system states is approximately linear. By leveraging Koopman operator theory, this stage enhances the temporal consistency during the latent learning. In the second stage, a Transformer module is employed to model the temporal progression within the Koopman-encoded latent space. Through the self-attention mechanism, the Transformer captures long-range temporal dependencies, enabling accurate forecasting over extended horizons. We evaluate TK-GCN in spatiotemporal cardiac dynamics forecasting and benchmark its performance against several state-of-the-art baselines. Experimental results and ablation studies show that TK-GCN consistently delivers superior predictive accuracy across a range of forecast horizons, demonstrating its capability to effectively model complex spatial structures and nonlinear temporal dynamics.

cs.LG↗

arXiv2025

In-Context Learning Enables Robot Action Prediction in LLMs

Recently, Large Language Models (LLMs) have achieved remarkable success using in-context learning (ICL) in the language domain. However, leveraging the ICL capabilities within LLMs to directly predict robot actions remains largely unexplored. In this paper, we introduce RoboPrompt, a framework that enables off-the-shelf text-only LLMs to directly predict robot actions through ICL without training. Our approach first heuristically identifies keyframes that capture important moments from an episode. Next, we extract end-effector actions from these keyframes as well as the estimated initial object poses, and both are converted into textual descriptions. Finally, we construct a structured template to form ICL demonstrations from these textual descriptions and a task instruction. This enables an LLM to directly predict robot actions at test time. Through extensive experiments and analysis, RoboPrompt shows stronger performance over zero-shot and ICL baselines in simulated and real-world settings. Our project page is available at https://davidyyd.github.io/roboprompt.

cs.RO↗

arXiv2025

MUSE-Net: Missingness-aware mUlti-branching Self-attention Encoder for Irregular Longitudinal Electronic Health Records

The era of big data has made vast amounts of clinical data readily available, particularly in the form of electronic health records (EHRs), which provides unprecedented opportunities for developing data-driven diagnostic tools to enhance clinical decision making. However, the application of EHRs in data-driven modeling faces challenges such as irregularly spaced multi-variate time series, issues of incompleteness, and data imbalance. Realizing the full data potential of EHRs hinges on the development of advanced analytical models. In this paper, we propose a novel Missingness-aware mUlti-branching Self-Attention Encoder (MUSE-Net) to cope with the challenges in modeling longitudinal EHRs for data-driven disease prediction. The proposed MUSE-Net is composed by four novel modules including: (1) a multi-task Gaussian process (MGP) with missing value masks for data imputation; (2) a multi-branching architecture to address the data imbalance problem; (3) a time-aware self-attention encoder to account for the irregularly spaced time interval in longitudinal EHRs; (4) interpretable multi-head attention mechanism that provides insights into the importance of different time points in disease prediction, allowing clinicians to trace model decisions. We evaluate the proposed MUSE-Net using both synthetic and real-world datasets. Experimental results show that our MUSE-Net outperforms existing methods that are widely used to investigate longitudinal signals.

cs.LG↗

arXiv2024

Exploring Large Language Model based Intelligent Agents: Definitions, Methods, and Prospects

Intelligent agents stand out as a potential path toward artificial general intelligence (AGI). Thus, researchers have dedicated significant effort to diverse implementations for them. Benefiting from recent progress in large language models (LLMs), LLM-based agents that use universal natural language as an interface exhibit robust generalization capabilities across various applications -- from serving as autonomous general-purpose task assistants to applications in coding, social, and economic domains, LLM-based agents offer extensive exploration opportunities. This paper surveys current research to provide an in-depth overview of LLM-based intelligent agents within single-agent and multi-agent systems. It covers their definitions, research frameworks, and foundational components such as their composition, cognitive and planning methods, tool utilization, and responses to environmental feedback. We also delve into the mechanisms of deploying LLM-based agents in multi-agent systems, including multi-role collaboration, message passing, and strategies to alleviate communication issues between agents. The discussions also shed light on popular datasets and application scenarios. We conclude by envisioning prospects for LLM-based agents, considering the evolving landscape of AI and natural language processing.

cs.AI↗

arXiv2023

Better Diffusion Models Further Improve Adversarial Training

It has been recognized that the data generated by the denoising diffusion probabilistic model (DDPM) improves adversarial training. After two years of rapid development in diffusion models, a question naturally arises: can better diffusion models further improve adversarial training? This paper gives an affirmative answer by employing the most recent diffusion model which has higher efficiency ($\sim 20$ sampling steps) and image quality (lower FID score) compared with DDPM. Our adversarially trained models achieve state-of-the-art performance on RobustBench using only generated data (no external datasets). Under the $\ell_\infty$-norm threat model with $ε=8/255$, our models achieve $70.69\%$ and $42.67\%$ robust accuracy on CIFAR-10 and CIFAR-100, respectively, i.e. improving upon previous state-of-the-art models by $+4.58\%$ and $+8.03\%$. Under the $\ell_2$-norm threat model with $ε=128/255$, our models achieve $84.86\%$ on CIFAR-10 ($+4.44\%$). These results also beat previous works that use external data. We also provide compelling results on the SVHN and TinyImageNet datasets. Our code is available at https://github.com/wzekai99/DM-Improves-AT.

cs.CV↗

arXiv2022

Hierarchical Deep Learning with Generative Adversarial Network for Automatic Cardiac Diagnosis from ECG Signals

Cardiac disease is the leading cause of death in the US. Accurate heart disease detection is of critical importance for timely medical treatment to save patients' lives. Routine use of electrocardiogram (ECG) is the most common method for physicians to assess the electrical activities of the heart and detect possible abnormal cardiac conditions. Fully utilizing the ECG data for reliable heart disease detection depends on developing effective analytical models. In this paper, we propose a two-level hierarchical deep learning framework with Generative Adversarial Network (GAN) for automatic diagnosis of ECG signals. The first-level model is composed of a Memory-Augmented Deep auto-Encoder with GAN (MadeGAN), which aims to differentiate abnormal signals from normal ECGs for anomaly detection. The second-level learning aims at robust multi-class classification for different arrhythmias identification, which is achieved by integrating the transfer learning technique to transfer knowledge from the first-level learning with the multi-branching architecture to handle the data-lacking and imbalanced data issue. We evaluate the performance of the proposed framework using real-world medical data from the MIT-BIH arrhythmia database. Experimental results show that our proposed model outperforms existing methods that are commonly used in current practice.

eess.SP↗