Author name cluster

Bo Yu

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

2 author rows

IROS Conference 2025 Conference Paper

3DWSNet: A Novel 3D Wavelet Spiking Neural Network for Event-based Action Recognition

Junkang Fang
Yonghao Dang
Wending Zhao
Bo Yu
Zehao Wang
Jianqin Yin

In robotics applications, event cameras provide low-latency and high-dynamic-range sensing by asynchronously detecting brightness changes, making them well-suited for capturing fast motions and subtle cues in dynamic environments. However, most existing Spiking Neural Network (SNN)-based methods enhance spatial information by stacking multiple frames of events, while neglecting the explicit modeling of high-and low-frequency components in the event stream. To address this limitation, we proposes a 3D Wavelet Spiking Neural Network (3DWSNet), which integrates a 3D wavelet transform with a cascaded Wavelet Spiking Convolution (WSC) module as its core. Specifically, the 3D wavelet transform decomposes input data into eight frequency sub-bands across spatial and temporal dimensions, enabling the model to preserve fine-grained high-frequency details while enriching low-frequency motion representations. The cascaded WSC architecture further improves the extraction of multi-scale spatio-temporal features by integrating information from feature maps at different resolutions. Extensive experiments show that our 3DWSNet significantly outperforms SOTA SNN performances on the CIFAR-10, CIFAR-100, DVS128 Gesture, and CIFAR10-DVS datasets. The source code will be publicly released soon.

Details

EAAI Journal 2025 Journal Article

A position-aware sets based weakly supervised framework for whole-slide subtype classification

Jiuman Song
Bo Yu
Xiaomin Liu
Lele Cong
Zilong Zhou
Xianling Cong
Hongyan Sun
Shuchao Pang

Identifying cancer subtypes is essential for personalized treatment and accurate prognosis due to the varying sensitivities of subtypes to therapies. However, in cancer subtype classification tasks, normal slides are usually scarce or absent as negative samples, while subtype whole-slide images (WSIs) often contain extensive unannotated normal tissue regions. These regions introduce significant noise during feature fusion and subtype classification, leading to degraded performance of existing weakly supervised methods In this paper, we propose the Position-aware Sets based Weakly Supervised learning framework (PSWS), designed for cancer subtype classification using WSIs, with a two-stage structure to enhance model efficiency. Specifically, it first presents a novel patch organization approach, distinct from the bag concept of traditional Multiple Instance Learning (MIL), called position-aware sets, as basic units for learning. Then, PSWS automatically selects subtype-specific features based on enhanced histological features and mutual-patch relations, mitigating the negative impact of unannotated negative regions. In the experiments, the superior performance of PSWS over representative MILs is validated through subtype classification tasks on both public datasets and our internally constructed dataset. Furthermore, class probabilities of position-aware sets and attention region visualizations demonstrate its post-hoc interpretability, assisting pathologists in locating suspicious areas.

Details DOI

NeurIPS Conference 2025 Conference Paper

EfficientNav: Towards On-Device Object-Goal Navigation with Navigation Map Caching and Retrieval

Zebin Yang
Sunjian Zheng
Tong Xie
Tianshi Xu
Bo Yu
Fan Wang
Jie Tang
Shaoshan Liu

Object-goal navigation (ObjNav) tasks an agent with navigating to the location of a specific object in an unseen environment. Embodied agents equipped with large language models (LLMs) and online constructed navigation maps can perform ObjNav in a zero-shot manner. However, existing agents heavily rely on giant LLMs on the cloud, e. g. , GPT-4, while directly switching to small LLMs, e. g. , LLaMA3. 2-11b, suffer from significant success rate drops due to limited model capacity for understanding complex navigation maps, which prevents deploying ObjNav on local devices. At the same time, the long prompt introduced by the navigation map description will cause high planning latency on local devices. In this paper, we propose EfficientNav to enable on-device efficient LLM-based zero-shot ObjNav. To help the smaller LLMs better understand the environment, we propose semantics-aware memory retrieval to prune redundant information in navigation maps. To reduce planning latency, we propose discrete memory caching and attention-based memory clustering to efficiently save and re-use the KV cache. Extensive experimental results demonstrate that EfficientNav achieves 11. 1\% improvement in success rate on HM3D benchmark over GPT-4-based baselines, and demonstrates 6. 7$\times$ real-time latency reduction and 4. 7$\times$ end-to-end latency reduction over GPT-4 planner. Our code is available on https: //github. com/PKU-SEC-Lab/EfficientNav.

PDF Details

EAAI Journal 2025 Journal Article

Myocardial ischemic classification using a knowledge-guided polar transformer in two-dimensional echocardiography

Ziwei Pang
Yi Du
Yanhui Guo
Shuang Chen
Bo Yu
Siqi Guo
Guo-Qing Du

Myocardial ischemia, characterized by inadequate blood supply to the heart muscles, is critical to cardiovascular diseases. Timely and accurate identification of ischemic segments is essential for prompt intervention and patient care. This study developed a Transformer-based model to identify myocardial ischemia in left ventricle short-axis (LVSA) two-dimensional echocardiography (2DE) images where a novel Knowledge-Guided Polar Transformer (KGPT) was proposed that integrated the unique characteristics of 2DE images with the prior clinical knowledge. 305 patients (aged 57. 6 ± 8. 8 years) were selected and underwent transthoracic echocardiography within 1–3 days prior to invasive coronary angiography (ICA). With ICA and quantitative flow ratio as the gold standard of myocardial ischemia, the KGPT model was trained to classify the LVSA 2DE images as ischemia or non-ischemia by capturing spatial features in a radial orientation. Its performance was evaluated with five-fold cross-validation and receiver operating characteristic curve (ROC) analysis. It achieved an area under ROC (AUC) of 0. 8326 ± 0. 0906, with an accuracy of 79. 50 ± 5. 40 %, precision of 79. 07 ± 6. 70 %, recall of 80. 79 ± 7. 87 %, and F1 score of 78. 43 ± 6. 56 %. In comparison, the original Swin-Transformer model produced an AUC of 0. 7011 ± 0. 0334, accuracy of 70. 20 ± 1. 04 %, precision of 68. 58 ± 3. 12 %, recall of 63. 21 ± 3. 60 %, and F1 score of 63. 13 ± 3. 78 %. The differences were statistically significant (P < 0. 05). The KGPT also demonstrated significantly superior performance to radiologists. It effectively classifies ischemic regions in 2DE images, presenting a promising tool for diagnosing myocardial ischemia. The integration of clinical knowledge with Transformer enhances the accuracy and reliability of ischemia classification, potentially revolutionizing the diagnosis and monitoring of myocardial ischemic diseases.

Details DOI

NeurIPS Conference 2025 Conference Paper

Towards Large-Scale In-Context Reinforcement Learning by Meta-Training in Randomized Worlds

Fan Wang
Pengtao Shao
Yiming Zhang
Bo Yu
Shaoshan Liu
Ning Ding
Yang Cao
Yu Kang

In-Context Reinforcement Learning (ICRL) enables agents to learn automatically and on-the-fly from their interactive experiences. However, a major challenge in scaling up ICRL is the lack of scalable task collections. To address this, we propose the procedurally generated tabular Markov Decision Processes, named AnyMDP. Through a carefully designed randomization process, AnyMDP is capable of generating high-quality tasks on a large scale while maintaining relatively low structural biases. To facilitate efficient meta-training at scale, we further introduce decoupled policy distillation and induce prior information in the ICRL framework. Our results demonstrate that, with a sufficiently large scale of AnyMDP tasks, the proposed model can generalize to tasks that were not considered in the training set through versatile in-context learning paradigms. The scalable task set provided by AnyMDP also enables a more thorough empirical investigation of the relationship between data distribution and ICRL performance. We further show that the generalization of ICRL potentially comes at the cost of increased task diversity and longer adaptation periods. This finding carries critical implications for scaling robust ICRL capabilities, highlighting the necessity of diverse and extensive task design, and prioritizing asymptotic performance over few-shot adaptation.

PDF Details

AAAI Conference 2025 Conference Paper

VA-AR: Learning Velocity-Aware Action Representations with Mixture of Window Attention

Jiangning Wei
Lixiong Qin
Bo Yu
Tianjian Zou
Chuhan Yan
Dandan Xiao
Yang Yu
Lan Yang

Action recognition is a crucial task in artificial intelligence, with significant implications across various domains. We initially perform a comprehensive analysis of seven prominent action recognition methods across five widely-used datasets. This analysis reveals a critical, yet previously overlooked, observation: as the velocity of actions increases, the performance of these methods variably declines, undermining their robustness. This decline in performance poses significant challenges for their application in real-world scenarios. Building on these findings, we introduce the Velocity-Aware Action Recognition (VA-AR) framework to obtain robust action representations across different velocities. Our principal insight is that rapid actions (e.g., the giant circle backward in uneven bars or a smash in badminton) occur within short time intervals, necessitating smaller temporal attention windows to accurately capture intricate changes. Conversely, slower actions (e.g., drinking water or wiping face) require larger windows to effectively encompass the broader context. VA-AR employs a Mixture of Window Attention (MoWA) strategy, dynamically adjusting its attention window size based on the action's velocity. This adjustment enables VA-AR to obtain a velocity-aware representation, thereby enhancing the accuracy of action recognition. Extensive experiments confirm that VA-AR achieves state-of-the-art performance on the same five datasets, demonstrating VA-AR's effectiveness across a broad spectrum of action recognition scenarios.

PDF Details DOI

EAAI Journal 2024 Journal Article

Using meta-learning to establish a highly transferable driving speed prediction model from the visual road environment

Bo Yu
Xiangyu Feng
You Kong
Yuren Chen
Zeyang Cheng
Shan Bao

Driving speed prediction is of great significance for the realization of autonomous driving and intelligent transportation systems. The effectiveness of the existing driving speed prediction models often relies on a large number of training data, and their performance will drop sharply with the change in the usage environment. To solve these problems, this study adopted a Meta-learning algorithm to propose a highly transferable driving speed prediction model based on the visual road environment. The data used in this study were the visual road environment pictures and corresponding driving speed from three different types of scenarios (i. e. , rural roads, urban roads, and highways) in a naturalistic driving experiment and a public KITTI dataset. The analysis of a Random forest model showed that the influence degree of visual road environment elements on driving speed was quite distinct in different scenarios. Then, a Meta-learning-based driving speed prediction model was presented to address two issues, including the small sample size and transferability. To examine the performance of this new model, a driving speed prediction model based on a convolutional neural network (CNN) was also used for comparison. The results showed that the prediction accuracy of Meta-learning was significantly improved compared to CNN in the face of the small sample size problem and when transferred to new scenarios. The findings in this study can contribute to the optimization of the driving speed prediction model and the improvement of the visual road environment design.

Details DOI