Author name cluster

Chenxi Huang

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.

6 papers

1 author row

EAAI Journal 2026 Journal Article

Hybrid semantic segmentation with broad context and attention encoded network for urban street scenario

Khawaja Iftekhar Rashid
Abid Hussain
Chenhui Yang
Chenxi Huang

Real-time semantic segmentation helps intelligent transportation systems analyze dynamic urban environments with precision. So far, even advanced models struggle to capture fine details in live scenarios because they cannot adapt quickly to changing conditions and require heavy computational resources. In this study, we present a hybrid model for semantic segmentation incorporating generative semantic segmentation as a key notch. We present the Dynamic Context-Aware network during the latent prior learning process to capture contextual details. A maskige image is used to generate the segmentation mask for posterior distribution. The conditioning network bridges the gap between the mask's posterior distribution and the latent prior distribution of the training data. Finally, we present the attention-encoded decoder unit to further refine the feature map before the final segmentation map. Rigorous evaluations on established benchmarks confirm that our Hybrid Semantic Segmentation (HSS) method has achieved competitive performance compared to previous state-of-the-art alternatives in the semantic segmentation context. Additionally, our Hybrid Semantic Segmentation system sets a new benchmark in the more challenging real-world urban scenario.

Details DOI

EAAI Journal 2026 Journal Article

Spatial Frequency Guidance Network for real-time lung computed tomography image segmentation

Mohammed A.M. Elhassan
JiaZhou Xiao
Bo Liu
Xu Li
Chenxi Huang
Jinbao Xie
Minglian Qiu

With the increasing demand for real-time lung tumor segmentation in clinical settings, there is a pressing need for efficient and accurate segmentation models. However, current state-of-the art methods are often burdened by high computational complexity and struggle to simultaneously capture fine-grained local details and global contextual cues essential for precise tumor delineation. To address these challenges, we present the Spatial Frequency Guidance Network (SFGNet), a lightweight encoder–decoder architecture tailored for real-time lung tumor segmentation. The proposed framework incorporates three key components: (1) a Residual Downsampling Block that reduces spatial resolution while preserving high-frequency structural cues; (2) a Spatial Frequency Guidance Module that integrates spatial and frequency-domain information to enhance multi-scale contextual representation; and (3) a Multi-Scale Feature Interaction Module that adaptively fuses features across resolutions to improve pixel-level classification accuracy. Experiments on two public lung computed tomography datasets demonstrate that the proposed architecture achieves a favorable balance between accuracy and efficiency. For example, the base variant attains best average across all metrics 97. 32% at real-time inference speed, while also producing best performance on a benchmark dataset of coronavirus disease (COVID-19) computed tomography scans. The source code, will be released publicly at: https: //github. com/mohamedac29/SFGNet.

Details DOI

EAAI Journal 2025 Journal Article

Dynamic context-aware high-resolution network for semi-supervised semantic segmentation

Khawaja Iftekhar Rashid
Chenhui Yang
Chenxi Huang

Real-time semi-supervised semantic segmentation provides a detailed understanding of dynamic urban situations for Intelligent transportation systems. However, state-of-the-art models following different segmentation tasks fail to incorporate High-Resolution Networks (HRNet) for real-time situations due to the inability to adjust in dynamic environments and high computational complexity. The study aims to examine the efficacy of student-teacher networks in semi-supervised semantic segmentation, mainly focusing on intelligent transportation. We present a Dynamic Context-aware High-Resolution Network (DC-HRNet) to enhance semantic segmentation in a dynamic urban environment incorporating a Student-Teacher knowledge distillation mechanism. We train teacher networks utilizing high-resolution networks to enhance robustness and precision in autonomous driving. Additionally, we employ Multi-path Blocks (MPBs) with HRNet to train our student network to address the challenges of the scarcity of labeled datasets in real-world scenarios. Utilization of MPBs for downsampling helps avoid low-resolution loss and generate pseudo labels by leveraging feature maps. We integrate a Dynamic Context-Aware Segmentation Network (DCSNet) to reduce the computational cost further and increase segmentation accuracy. Integrating DCSNet across various resolutions of HRNet with diverse dilation rates has enhanced contextual data utilization. This study contributes to intelligent transportation systems by improving segmentation accuracy, model generalization, and overall resilience, expanding the research scope in this area. Extensive investigations on the publicly available datasets show that our model can handle high-resolution images in real-time with state-of-the-art performance.

Details DOI

EAAI Journal 2025 Journal Article

Real-time medical image segmentation via attentive atrous pyramid network

Mohammed A.M. Elhassan
Sifang Chen
Chenxi Huang

Semantic segmentation networks often prioritize achieving high accuracy, which typically results in increased computational complexity and slower inference speeds. As a result, balancing accuracy with efficiency has become a critical challenge, particularly in the domain of medical imaging. To alleviate this problem, we present a novel dilated depthwise block and design lightweight encoder. In addition, our approach addresses key limitations of existing methods through three main innovations: Firstly, An Attentive Atrous Feature Pyramid Module (A2PM) that effectively captures long-range dependencies and multi-scale contextual information without excessive computational overhead. Secondly, a Spatial-Channel Attention Block (SCAB) that combines spatial and channel attention mechanisms to enhance feature representation and selectivity. Thirdly, a Residual Global Context module (RGCM) that efficiently merges low-level and high-level features, mitigating the semantic gap often present in encoder–decoder architectures. A2PNet overcomes common challenges such as feature degradation from repeated pooling operations and sub-optimal skip connections. Our comprehensive evaluation on Breast UltraSound Image and three retinal vessel datasets demonstrates that A2PNet consistently outperforms existing methods across various metrics. Notably, A2PNet-L achieves an F1 score of 84. 22%, 77. 6% and 83. 56% on the CHASEDB1, STARE and DRIVE datasets, respectively. The code will be available on https: //github. com/mohamedac29/A2PNet.

Details DOI

NeurIPS Conference 2024 Conference Paper

kGym: A Platform and Dataset to Benchmark Large Language Models on Linux Kernel Crash Resolution

Alex Mathai
Chenxi Huang
Petros Maniatis
Aleksandr Nogikh
Franjo Ivančić
Junfeng Yang
Baishakhi Ray

Large Language Models (LLMs) are consistently improving at increasingly realistic software engineering (SE) tasks. In real-world software stacks, significant SE effort is spent developing foundational system software like the Linux kernel. Unlike application-level software, a systems codebase like Linux is multilingual (low-level C/Assembly/Bash/Rust); gigantic (>20 million lines); critical (impacting billions of devices worldwide), and highly concurrent (involving complex multi-threading). To evaluate if machine learning (ML) models are useful while developing such large-scale systems-level software, we introduce kGym (a platform) and kBench (a dataset). The kGym platform provides a SE environment for large-scale experiments on the Linux kernel, including compiling and running kernels in parallel across several virtual machines, detecting operations and crashes, inspecting logs, and querying and patching the code base. We use kGym to facilitate evaluation on kBench, a crash resolution benchmark drawn from real-world Linux kernel bugs. An example bug in kBench contains crashing stack traces, a bug-reproducer file, a developer-written fix, and other associated data. To understand current performance, we conduct baseline experiments by prompting LLMs to resolve Linux kernel crashes. Our initial evaluations reveal that the best performing LLM achieves 0. 72\% and 5. 38\% in the unassisted and assisted (i. e. , buggy files disclosed to the model) settings, respectively. These results highlight the need for further research to enhance model performance in SE tasks. Improving performance on kBench requires models to master new learning skills, including understanding the cause of crashes and repairing faults, writing memory-safe and hardware-aware code, and understanding concurrency. As a result, this work opens up multiple avenues of research at the intersection of machine learning and systems software.

PDF Details DOI

AAAI Conference 2023 Conference Paper

LagNet: Deep Lagrangian Mechanics for Plug-and-Play Molecular Representation Learning

Chunyan Li
Junfeng Yao
Jinsong Su
Zhaoyang Liu
Xiangxiang Zeng
Chenxi Huang

Molecular representation learning is a fundamental problem in the field of drug discovery and molecular science. Whereas incorporating molecular 3D information in the representations of molecule seems beneficial, which is related to computational chemistry with the basic task of predicting stable 3D structures (conformations) of molecules. Existing machine learning methods either rely on 1D and 2D molecular properties or simulate molecular force field to use additional 3D structure information via Hamiltonian network. The former has the disadvantage of ignoring important 3D structure features, while the latter has the disadvantage that existing Hamiltonian neural network must satisfy the “canonial” constraint, which is difficult to be obeyed in many cases. In this paper, we propose a novel plug-and-play architecture LagNet by simulating molecular force field only with parameterized position coordinates, which implements Lagrangian mechanics to learn molecular representation by preserving 3D conformation without obeying any additional restrictions. LagNet is designed to generate known conformations and generalize for unknown ones from molecular SMILES. Implicit positions in LagNet are learned iteratively using discrete-time Lagrangian equations. Experimental results show that LagNet can well learn 3D molecular structure features, and outperforms previous state-of-the-art baselines related molecular representation by a significant margin.

PDF Details DOI