Yulin Wang

EAAI Journal 2026 Journal Article

A fault evolution knowledge-driven adversarial meta-learning method for few-shot tool state recognition under variable working conditions

• Chen Yin
• Yining Dong
• Jianliang He
• Yulin Wang

The health status of cutting tools is vital to ensuring workpiece quality and operational safety. While most existing research on tool condition monitoring focuses on progressive wear, such as estimating wear volume or remaining useful life, the more critical and abrupt failure mode of tool breakage has received less attention. In real-world manufacturing, stringent safety protocols lead to a severe scarcity of breakage data, presenting a typical few-shot learning challenge for tool state recognition (TSR). To tackle this issue, we propose a novel fault evolution knowledge-driven adversarial meta-learning (FEK-AML) method for few-shot TSR, where fault evolution knowledge is creatively formulated and integrated into the proposed adversarial meta-learning method, resulting in a physics-guided learning framework. Specifically, a feature extraction network is first trained using domain adversarial training to learn domain-invariant features while capturing the fault evolution knowledge. Subsequently, a metric-based meta-learning network is designed to transfer this knowledge for effective TSR under few-shot conditions. Milling experiments are performed on cutting tools in healthy, worn, and broken states under various working conditions. A series of TSR tasks is constructed, with only one fault sample per class available in the target domain. Comparative results show that FEK-AML effectively mines fault evolution knowledge and outperforms existing approaches in recognizing tool states under extremely limited data conditions, confirming its potential for reliable deployment in CNC monitoring systems to achieve accurate and robust TSR.

YNIMG Journal 2026 Journal Article

Corrigendum to “Opposite changes in morphometric similarity of medial reward and lateral non-reward orbitofrontal cortex circuits in obesity” [NeuroImage, 290 (2024) 120574]

• Debo Dong
• Ximei Chen
• Wei Li
• Xiao Gao
• Yulin Wang
• Feng Zhou
• Simon B. Eickhoff
• Hong Chen

EAAI Journal 2026 Journal Article

Real-time Vehicle-Induced Response Identification via crowdsourced labeling for high-frequency unlabeled sensor data

• Zhixin Qi
• Yulin Wang
• Zhiyang Liu
• Zemin Chao
• Zejiao Dong
• Hongzhi Wang

Since the high sampling frequency of highway structural health monitoring data brings a heavy burden on data storage and processing, it is urgent to identify the vehicle-induced responses in real time and filter the remaining meaningless data. However, existing approaches either fail to construct a high-quality training data for identification or obtain an unsatisfactory speed far from real-time requirements. Motivated by this, we propose RVIRI (Real-time Vehicle-Induced Response Identification), a novel framework that integrates similarity-driven crowdsourced labeling, statistical feature engineering, and lightweight identification to enable efficient and accurate response identification from high-frequency unlabeled sensor data. To reduce the costs of crowdsourcing and guarantee the high quality of training data, we define a vehicle-induced response crowdsourced labeling problem and develop a greedy algorithm with (ln δ +2)-approximation ratio. Our proposed method not only decreases the crowdsourced by discovering similar patterns, but also enhances the training data with theoretical guarantees. Experimental results on real highway structural health monitoring data demonstrate that our proposed approach achieves an effective identification throughput of 77. 24 GB(GigaByte)/s and outperforms commonly-used methods in identification effectiveness.

NeurIPS Conference 2025 Conference Paper

GeoLLaVA-8K: Scaling Remote-Sensing Multimodal Large Language Models to 8K Resolution

• Fengxiang Wang
• Mingshuo Chen
• Yueying Li
• Di Wang
• Haotian Wang
• Zonghao Guo
• Zefan Wang
• Shan Boqi

Ultra-high-resolution (UHR) remote sensing (RS) imagery offers valuable data for Earth observation but pose challenges for existing multimodal foundation models due to two key bottlenecks: (1) limited availability of UHR training data, and (2) token explosion caused by the large image size. To address data scarcity, we introduce **SuperRS-VQA** (avg. 8, 376$\times$8, 376) and **HighRS-VQA** (avg. 2, 000$\times$1, 912), the highest-resolution vision-language datasets in RS to date, covering 22 real-world dialogue tasks. To mitigate token explosion, our pilot studies reveal significant redundancy in RS images: crucial information is concentrated in a small subset of object-centric tokens, while pruning background tokens (e. g. , ocean or forest) can even improve performance. Motivated by these findings, we propose two strategies: *Background Token Pruning* and *Anchored Token Selection*, to reduce the memory footprint while preserving key semantics. Integrating these techniques, we introduce **GeoLLaVA-8K**, the first RS-focused multimodal large language model capable of handling inputs up to 8K$\times$8K resolution, built on the LLaVA framework. Trained on SuperRS-VQA and HighRS-VQA, GeoLLaVA-8K sets a new state-of-the-art on the XLRS-Bench. Datasets and code were released at https: //github. com/MiliLab/GeoLLaVA-8K.

NeurIPS Conference 2025 Conference Paper

RoMA: Scaling up Mamba-based Foundation Models for Remote Sensing

• Fengxiang Wang
• Yulin Wang
• Mingshuo Chen
• Haotian Wang
• Hongzhen Wang
• Haiyan Zhao
• Yangang Sun
• Shuo Wang

Recent advances in self-supervised learning for Vision Transformers (ViTs) have fueled breakthroughs in remote sensing (RS) foundation models. However, the quadratic complexity of self-attention poses a significant barrier to scalability, particularly for large models and high-resolution images. While the linear-complexity Mamba architecture offers a promising alternative, existing RS applications of Mamba remain limited to supervised tasks on small, domain-specific datasets. To address these challenges, we propose RoMA, a framework that enables scalable self-supervised pretraining of Mamba-based RS foundation models using large-scale, diverse, unlabeled data. RoMA enhances scalability for high-resolution images through a tailored auto-regressive learning strategy, incorporating two key innovations: 1) a rotation-aware pretraining mechanism combining adaptive cropping with angular embeddings to handle sparsely distributed objects with arbitrary orientations, and 2) multi-scale token prediction objectives that address the extreme variations in object scales inherent to RS imagery. Systematic empirical studies validate that Mamba adheres to RS data and parameter scaling laws, with performance scaling reliably as model and data size increase. Furthermore, experiments across scene classification, object detection, and semantic segmentation tasks demonstrate that RoMA-pretrained Mamba models consistently outperform ViT-based counterparts in both accuracy and computational efficiency. The source code and pretrained models have be released at https: //github. com/MiliLab/RoMA.

EAAI Journal 2024 Journal Article

Deep convolutional neural network based on self-distillation for tool wear recognition

• Yi Pan
• Ling Hao
• Jianliang He
• Kun Ding
• Qiang Yu
• Yulin Wang

In the ever-evolving Industry 4. 0 landscape, traditional manufacturing constantly merges with cutting-edge AI technology. In manufacturing, Computer Numerical Control (CNC) machine tools heavily depend on cutting tools. Most existing researches on tool fault diagnosis center on machines customized for visual monitoring, overlooking the specific challenges posed by unmodifiable production line equipment. Mobile inspection robots, though versatile, are constrained by resources, limiting the use of large deep learning models. In view of this, we aim to introduce a tailored tool wear recognition solution for resource-constrained mobile inspection robots. To achieve this goal, a lightweight deep learning model with a unique single-stage self-distillation approach is presented. This method efficiently transfers knowledge from deep networks to shallow networks, optimizing performance while minimizing training costs. From our results, we achieve a promising top-1 accuracy of 97. 66% after 50 training epochs within 1 h and 20 min of training time. This surpasses the student model obtained through offline distillation by 4. 30%, saving 59. 39% of training time, and improves recognition accuracy by 2. 43% compared to online distillation, while reducing training time by 44. 83%. Compared with the state-of-the-art method TinyVit, 12. 09% of training time is saved with an accuracy improvement of 1. 48%. In conclusion, our model manifests a desirable balance between efficiency and accuracy, rendering it highly suitable for deployment on mobile terminals. Furthermore, ablation experiments confirm the robustness of our method’s internal structure in influencing final performance. This research addresses tool fault diagnosis in unmodifiable production line tools and is expected to provide a practical, efficient solution with broad implications for Industry 4. 0.

NeurIPS Conference 2024 Conference Paper

DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution

• Yang Yue
• Yulin Wang
• Bingyi Kang
• Yizeng Han
• Shenzhi Wang
• Shiji Song
• Jiashi Feng
• Gao Huang

Multimodal Large Language Models (MLLMs) have demonstrated remarkable comprehension and reasoning capabilities with complex language and visual data. These advances have spurred the vision of establishing a generalist robotic MLLM proficient in understanding complex human instructions and accomplishing various embodied tasks, whose feasibility has been recently verified~\cite{rt-2, rt-x}. However, developing MLLMs for real-world robots is challenging due to the typically limited computation and memory capacities available on robotic platforms. In contrast, the inference of MLLMs usually incorporates storing billions of parameters and performing tremendous computation, imposing significant hardware demands. In our paper, we seek to address this challenge by leveraging an intriguing observation: relatively easier situations make up the bulk of the procedure of controlling robots to fulfill diverse tasks, and they generally require far smaller models to obtain the correct robotic actions. Motivated by this observation, we propose a \emph{DynamicEarly-Exit for Robotic MLLM} (DeeR) framework that automatically adjusts the size of the activated MLLM based on each situation at hand. The approach leverages a multi-exit architecture in MLLMs, which allows the model to cease processing once a proper size of the model has been activated for a specific situation, thus avoiding further redundant computation. Additionally, we develop novel algorithms that establish early-termination criteria for DeeR, conditioned on predefined demands such as average computational cost (\emph{i. e. }, power consumption), as well as peak computational consumption (\emph{i. e. }, latency) and GPU memory usage. These enhancements ensure that DeeR operates efficiently under varying resource constraints while maintaining competitive performance. Moreover, we design a tailored training method for integrating temporal information on top of such multi-exit architectures to predict actions reasonably. On the CALVIN robot manipulation benchmark, DeeR demonstrates significant reductions in computational costs by 5. 2-6. 5x and GPU memory by 2x without compromising performance. Code and checkpoints are available at https: //github. com/yueyang130/DeeR-VLA.

NeurIPS Conference 2024 Conference Paper

ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis

• Zanlin Ni
• Yulin Wang
• Renping Zhou
• Yizeng Han
• Jiayi Guo
• Zhiyuan Liu
• Yuan Yao
• Gao Huang

Recently, token-based generation approaches have demonstrated their effectiveness in synthesizing visual content. As a representative example, non-autoregressive Transformers (NATs) can generate decent-quality images in just a few steps. NATs perform generation in a progressive manner, where the latent tokens of a resulting image are incrementally revealed step-by-step. At each step, the unrevealed image regions are padded with [MASK] tokens and inferred by NAT, with the most reliable predictions preserved as newly revealed, visible tokens. In this paper, we delve into understanding the mechanisms behind the effectiveness of NATs and uncover two important interaction patterns that naturally emerge from NAT’s paradigm: Spatially (within a step), although [MASK] and visible tokens are processed uniformly by NATs, the interactions between them are highly asymmetric. In specific, [MASK] tokens mainly gather information for decoding. On the contrary, visible tokens tend to primarily provide information, and their deep representations can be built only upon themselves. Temporally (across steps), the interactions between adjacent generation steps mostly concentrate on updating the representations of a few critical tokens, while the computation for the majority of tokens is generally repetitive. Driven by these findings, we propose EfficientNAT (ENAT), a NAT model that explicitly encourages these critical interactions inherent in NATs. At the spatial level, we disentangle the computations of visible and [MASK] tokens by encoding visible tokens independently, while decoding [MASK] tokens conditioned on the fully encoded visible tokens. At the temporal level, we prioritize the computation of the critical tokens at each step, while maximally reusing previously computed token representations to supplement necessary information. ENAT improves the performance of NATs notably with significantly reduced computational cost. Experiments on ImageNet-256 2 & 512 2 and MS-COCO validate the effectiveness of ENAT. Code and pre-trained models will be released at https: //github. com/LeapLabTHU/ENAT.

YNIMG Journal 2024 Journal Article

Opposite changes in morphometric similarity of medial reward and lateral non-reward orbitofrontal cortex circuits in obesity

• Debo Dong
• Ximei Chen
• Wei Li
• Xiao Gao
• Yulin Wang
• Feng Zhou
• Simon B. Eickhoff
• Hong Chen

Obesity has a profound impact on metabolic health thereby adversely affecting brain structure and function. However, the majority of previous studies used a single structural index to investigate the link between brain structure and body mass index (BMI), which hinders our understanding of structural covariance between regions in obesity. This study aimed to examine the relationship between macroscale cortical organization and BMI using novel morphometric similarity networks (MSNs). The individual MSNs were first constructed from individual eight multimodal cortical morphometric features between brain regions. Then the relationship between BMI and MSNs within the discovery sample of 434 participants was assessed. The key findings were further validated in an independent sample of 192 participants. We observed that the lateral non-reward orbitofrontal cortex (lOFC) exhibited decoupling (i.e., reduction in integration) in obesity, which was mainly manifested by its decoupling with the cognitive systems (i.e., DMN and FPN) while the medial reward orbitofrontal cortex (mOFC) showed de-differentiation (i.e., decrease in distinctiveness) in obesity, which was mainly represented by its de-differentiation with the cognitive and attention systems (i.e., DMN and VAN). Additionally, the lOFC showed de-differentiation with the visual system in obesity, while the mOFC showed decoupling with the visual system and hyper-coupling with the sensory-motor system in obesity. As an important first step in revealing the role of underlying structural covariance in body mass variability, the present study presents a novel mechanism that underlies the reward-control interaction imbalance in obesity, thus can inform future weight-management approaches.

STOC Conference 2024 Conference Paper

Sampling Proper Colorings on Line Graphs Using (1+o(1))Δ Colors

• Yulin Wang
• Chihao Zhang 0001
• Zihan Zhang

YNICL Journal 2022 Journal Article

Linking cerebellar functional gradients to transdiagnostic behavioral dimensions of psychopathology

• Debo Dong
• Xavier Guell
• Sarah Genon
• Yulin Wang
• Ji Chen
• Simon B. Eickhoff
• Dezhong Yao
• Cheng Luo

High co-morbidity and substantial overlap across psychiatric disorders encourage a transition in psychiatry research from categorical to dimensional approaches that integrate neuroscience and psychopathology. Converging evidence suggests that the cerebellum is involved in a wide range of cognitive functions and mental disorders. An important question thus centers on the extent to which cerebellar function can be linked to transdiagnostic dimensions of psychopathology. To address this question, we used a multivariate data-driven statistical technique (partial least squares) to identify latent dimensions linking human cerebellar connectome as assessed by functional MRI to a large set of clinical, cognitive, and trait measures across 198 participants, including healthy controls (n = 92) as well as patients diagnosed with attention-deficit/hyperactivity disorder (n = 35), bipolar disorder (n = 36), and schizophrenia (n = 35). Macroscale spatial gradients of connectivity at voxel level were used to characterize cerebellar connectome properties, which provide a low-dimensional representation of cerebellar connectivity, i.e., a sensorimotor-supramodal hierarchical organization. This multivariate analysis revealed significant correlated patterns of cerebellar connectivity gradients and behavioral measures that could be represented into four latent dimensions: general psychopathology, impulsivity and mood, internalizing symptoms and executive dysfunction. Each dimension was associated with a unique spatial pattern of cerebellar connectivity gradients across all participants. Multiple control analyses and 10-fold cross-validation confirmed the robustness and generalizability of the yielded four dimensions. These findings highlight the relevance of cerebellar connectivity as a necessity for the study and classification of transdiagnostic dimensions of psychopathology and call on researcher to pay more attention to the role of cerebellum in the dimensions of psychopathology, not just within the cerebral cortex.

NeurIPS Conference 2021 Conference Paper

Not All Images are Worth 16x16 Words: Dynamic Transformers for Efficient Image Recognition

• Yulin Wang
• Rui Huang
• Shiji Song
• Zeyi Huang
• Gao Huang

Vision Transformers (ViT) have achieved remarkable success in large-scale image recognition. They split every 2D image into a fixed number of patches, each of which is treated as a token. Generally, representing an image with more tokens would lead to higher prediction accuracy, while it also results in drastically increased computational cost. To achieve a decent trade-off between accuracy and speed, the number of tokens is empirically set to 16x16 or 14x14. In this paper, we argue that every image has its own characteristics, and ideally the token number should be conditioned on each individual input. In fact, we have observed that there exist a considerable number of “easy” images which can be accurately predicted with a mere number of 4x4 tokens, while only a small fraction of “hard” ones need a finer representation. Inspired by this phenomenon, we propose a Dynamic Transformer to automatically configure a proper number of tokens for each input image. This is achieved by cascading multiple Transformers with increasing numbers of tokens, which are sequentially activated in an adaptive fashion at test time, i. e. , the inference is terminated once a sufficiently confident prediction is produced. We further design efficient feature reuse and relationship reuse mechanisms across different components of the Dynamic Transformer to reduce redundant computations. Extensive empirical results on ImageNet, CIFAR-10, and CIFAR-100 demonstrate that our method significantly outperforms the competitive baselines in terms of both theoretical computational efficiency and practical inference speed. Code and pre-trained models (based on PyTorch and MindSpore) are available at https: //github. com/blackfeather-wang/Dynamic-Vision-Transformer and https: //github. com/blackfeather-wang/Dynamic-Vision-Transformer-MindSpore.

NeurIPS Conference 2020 Conference Paper

Glance and Focus: a Dynamic Approach to Reducing Spatial Redundancy in Image Classification

• Yulin Wang
• Kangchen Lv
• Rui Huang
• Shiji Song
• Le Yang
• Gao Huang

The accuracy of deep convolutional neural networks (CNNs) generally improves when fueled with high resolution images. However, this often comes at a high computational cost and high memory footprint. Inspired by the fact that not all regions in an image are task-relevant, we propose a novel framework that performs efficient image classification by processing a sequence of relatively small inputs, which are strategically selected from the original image with reinforcement learning. Such a dynamic decision process naturally facilitates adaptive inference at test time, i. e. , it can be terminated once the model is sufficiently confident about its prediction and thus avoids further redundant computation. Notably, our framework is general and flexible as it is compatible with most of the state-of-the-art light-weighted CNNs (such as MobileNets, EfficientNets and RegNets), which can be conveniently deployed as the backbone feature extractor. Experiments on ImageNet show that our method consistently improves the computational efficiency of a wide variety of deep models. For example, it further reduces the average latency of the highly efficient MobileNet-V3 on an iPhone XS Max by 20% without sacrificing accuracy. Code and pre-trained models are available at https: //github. com/blackfeather-wang/GFNet-Pytorch.

NeurIPS Conference 2019 Conference Paper

Implicit Semantic Data Augmentation for Deep Networks

• Yulin Wang
• Xuran Pan
• Shiji Song
• Hong Zhang
• Gao Huang
• Cheng Wu

In this paper, we propose a novel implicit semantic data augmentation (ISDA) approach to complement traditional augmentation techniques like flipping, translation or rotation. Our work is motivated by the intriguing property that deep networks are surprisingly good at linearizing features, such that certain directions in the deep feature space correspond to meaningful semantic transformations, e. g. , adding sunglasses or changing backgrounds. As a consequence, translating training samples along many semantic directions in the feature space can effectively augment the dataset to improve generalization. To implement this idea effectively and efficiently, we first perform an online estimate of the covariance matrix of deep features for each class, which captures the intra-class semantic variations. Then random vectors are drawn from a zero-mean normal distribution with the estimated covariance to augment the training data in that class. Importantly, instead of augmenting the samples explicitly, we can directly minimize an upper bound of the expected cross-entropy (CE) loss on the augmented training set, leading to a highly efficient algorithm. In fact, we show that the proposed ISDA amounts to minimizing a novel robust CE loss, which adds negligible extra computational cost to a normal training procedure. Although being simple, ISDA consistently improves the generalization performance of popular deep models (ResNets and DenseNets) on a variety of datasets, e. g. , CIFAR-10, CIFAR-100 and ImageNet. Code for reproducing our results are available at https: //github. com/blackfeather-wang/ISDA-for-Deep-Networks.