Jia Cheng Hu

ORCID: 0009-0008-1611-966X
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Multimodal Machine Learning Applications
  • Advanced Image and Video Retrieval Techniques
  • Human Pose and Action Recognition
  • Domain Adaptation and Few-Shot Learning
  • Image and Signal Denoising Methods
  • Manufacturing Process and Optimization
  • Image and Video Quality Assessment
  • Topic Modeling
  • Chaos-based Image/Signal Encryption
  • Risk and Safety Analysis
  • Photoacoustic and Ultrasonic Imaging
  • Video Surveillance and Tracking Methods
  • Autonomous Vehicle Technology and Safety
  • Cell Image Analysis Techniques
  • Flow Measurement and Analysis
  • Visual Attention and Saliency Detection
  • Industrial Vision Systems and Defect Detection
  • MRI in cancer diagnosis
  • Advanced Image Fusion Techniques
  • Music Technology and Sound Studies
  • Surface Roughness and Optical Measurements
  • Computer Graphics and Visualization Techniques
  • Energy, Environment, and Transportation Policies
  • Fluid Dynamics Simulations and Interactions
  • Vehicle emissions and performance

University of Modena and Reggio Emilia
2019-2024

State Grid Corporation of China (China)
2024

Liverpool Hope University
2012-2015

China Jiliang University
2011-2015

Xidian University
2009

Zhejiang University of Technology
2006

Autoregressive Sequence-To-Sequence (Seq2Seq) models are the foundation of many Deep Learning achievements in major research fields such as Vision and Natural Language Processing. However, their limitations motivated researchers to explore different architectures methodologies toward bidirectional solutions. In this work, we introduce Bidirectional Awareness Induction (BAI), a flexible training method that enhances information retained subset network results, which call pivot, through loss...

10.24135/iconip25 article EN cc-by-nc-sa 2025-03-17

We introduce a method called the Expansion mechanism that processes input unconstrained by number of elements in sequence. By doing so, model can learn more effectively compared to traditional attention-based approaches. To support this claim, we design novel architecture ExpansionNet v2 achieved strong results on MS COCO 2014 Image Captioning challenge and State Art its respective category, with score 143.7 CIDErD offline test split, 140.8 online evaluation server 72.9 AllCIDEr nocaps...

10.48550/arxiv.2208.06551 preprint EN cc-by-nc-nd arXiv (Cornell University) 2022-01-01

We introduce a method called the Expansion mechanism that processes input unconstrained by number of elements in sequence. By doing so, model can learn more effectively compared to traditional attention-based approaches. To support this claim, we design novel architecture ExpansionNet v2 achieved strong results on MS COCO 2014 Image Captioning challenge and State Art its respective category, with score 143.7 CIDErD offline test split, 140.8 online evaluation server 72.9 AllCIDEr nocaps...

10.1109/bigdata59044.2023.10386812 article EN 2021 IEEE International Conference on Big Data (Big Data) 2023-12-15

Abstract Modern teaching has made significant progress, with many advanced equipment and technologies being introduced into the process. Experimental of engineering design courses is important. Due to limited resources, students need effective guidance during laboratory time. We will introduce artificial intelligence solutions education. use technology for classroom behavior analysis improve practice courses' effectiveness. In an instructional milieu, image acquisition tools such as cameras...

10.1007/s44196-024-00572-y article EN cc-by International Journal of Computational Intelligence Systems 2024-06-27

Most recent state of the art architectures rely on combinations and variations three approaches: convolutional, recurrent self-attentive methods. Our work attempts in laying basis for a new research direction sequence modeling based upon idea modifying length. In order to do that, we propose method called "Expansion Mechanism" which transforms either dynamically or statically input into one featuring different Furthermore, introduce novel architecture that exploits such achieves competitive...

10.48550/arxiv.2207.03327 preprint EN cc-by-nc-nd arXiv (Cornell University) 2022-01-01

Summary In unmanned aerial vehicles navigation, path planning is aimed at obtaining the optimal safety between start and destination locations. The efficiency optimality criterion depend on environment method adopted. this paper, a general fast framework proposed for navigation. Standard A* search performed online roadmap, which consists of segments that are pre‐computed offline with aid multi‐resolution grid terminate somewhere along boundary adjacent cells. Fast marching (FMM) was employed...

10.1002/cpe.3291 article EN Concurrency and Computation Practice and Experience 2014-05-29

Abstract Digital Breast Tomosynthesis (DBT) is a modern 3D Computed Tomography X-ray technique for the early detection of breast tumors, which receiving growing interest in medical and scientific community. Since DBT performs incomplete sampling data, image reconstruction approaches based on iterative methods are preferable to classical analytic techniques, such as Filtered Back Projection algorithm, providing fewer artifacts. In this work, we consider Model-Based Iterative Reconstruction...

10.1038/s41598-019-56920-y article EN cc-by Scientific Reports 2020-01-08

The emergence of digital music platforms has fundamentally transformed the way we engage with and organize music. As playlist creation gained widespread popularity, there is an increasing desire among aficionados industry experts to comprehend factors that drive success. This paper presents a machine learning-based approach designed predict success playlists. By analyzing various musical characteristics songs, our model achieves impressive accuracy 89.6% in predicting Notably, it exhibits...

10.1109/ccnc51664.2024.10454829 article EN 2024-01-06

Autoregressive Sequence-To-Sequence models are the foundation of many Deep Learning achievements in major research fields such as Vision and Natural Language Processing. Despite that, they still present significant limitations. For instance, when errors occur early steps prediction, whole output is severely affected. Such reliance on previously predicted tokens inherent computational unfriendliness sequential algorithms, motivated researchers to explore different architectures methods search...

10.48550/arxiv.2408.13959 preprint EN arXiv (Cornell University) 2024-08-25

Image Captioning is an important Language and Vision task that finds application in a variety of contexts, ranging from healthcare to autonomous vehicles. As many real-world applications rely on devices with limited resources, much effort the field was put into development lighter faster models. However, current optimizations focus Transformer architecture contrast existence more efficient methods. In this work, we introduce SwiFTeR, almost entirely based Fourier Transform Retention, tackle...

10.48550/arxiv.2408.13963 preprint EN arXiv (Cornell University) 2024-08-25

The deployment of Pre-trained Language Models in memory-limited devices is hindered by their massive number parameters, which motivated the interest developing smaller architectures. Established works model compression literature showcased that small models often present a noticeable performance degradation and need to be paired with transfer learning methods, such as Knowledge Distillation. In this work, we propose parameter-sharing method consists sharing parameters between embeddings...

10.1609/aaai.v38i16.29781 article EN Proceedings of the AAAI Conference on Artificial Intelligence 2024-03-24

10.1109/ccdc62350.2024.10588146 article EN 2022 34th Chinese Control and Decision Conference (CCDC) 2024-05-25

Reinforcement Learning (RL) offers a promising solution to enable evolutionary automated driving. However, the conventional RL method is always concerned with risk performance. The updated policy may not obtain performance enhancement, even leading deterioration. To address this challenge, research proposes High Confidence Policy Improvement Learning-based (HCPI-RL) planner. It intended achieve monotonic evolution of A novel update paradigm designed newly learned consistently surpass that...

10.48550/arxiv.2412.10822 preprint EN arXiv (Cornell University) 2024-12-14

Error diffusion dithering is a technique that used to represent grey-scale image in format usable by printer. At every step, an algorithm converts the value of pixel new within allowed ones, generating conversion error. To achieve effect continuous-tone illusion, error distributed neighboring pixels. Among existent algorithms, most commonly Floyd-Steinberg. However, this suffers two issues: artifacts and slowness. Regarding artifacts, those are textures can appear after elaboration, making...

10.1109/iciip47207.2019.8985831 article EN 2019-11-01

Dithering or error diffusion is a technique used to obtain binary image, suitable for printing, from grayscale one.At each step, the algorithm computes an allowed value of pixel one, applying threshold and, therefore, causing conversion error.To optical illusion continuous tone, obtained distributed adjacent pixels.In literature there are many algorithms this type, cite some Jarvis, Judice and Ninke (JJN), Stucki, Atkinson, Burkes, Sierra but most known Floyd-Steinberg.We compared various...

10.25046/aj050679 article EN Advances in Science Technology and Engineering Systems Journal 2020-11-01

Diaphragm gas meter is a specialized flow that measures the volume of fuel such as natural and coal gas. Valve bonnet valve seat are key parts diaphragm meter, which compose main factors its metering error. The widely detecting ways direct observational method pneumatic pressure method. Direct only realizes qualitative detection with lacking science. Pneumatic can realize quantitative detection, but low accuracy bad reliability. In order to evaluate tightness accurately, surface texture was...

10.4028/www.scientific.net/kem.609-610.1170 article EN Key engineering materials 2014-04-01

White light interference technique for topography measurement effectively avoids phase ambiguity in phase-shifting interferometry. The spatial frequency domain algorithm based on scanning white has the advantage of insensitivity to noise and higher calculation accuracy compared with other methods. sensor is constructed nano positioning measuring machine (NMM), calibrated step height standard 100±3nm measured. adopted data processing, repetitive test result 97.9nm deviation 0.48nm are...

10.4028/www.scientific.net/amm.738-739.904 article EN Applied Mechanics and Materials 2015-03-01

This special issue of Concurrency and Computation: Practice Experience provides a forum for presenting advances current research development in all aspects Parallel Distributed Computing Communications.Because the tremendous broad spectrum technologies topics including wireless networking, cloud computing sensor systems, distributed communications has evolved into an active important area development.The past decade witnessed proliferation powerful parallel systems practice high performance...

10.1002/cpe.3547 article EN Concurrency and Computation Practice and Experience 2015-05-21

SPC technology was applied to the process of DVR production a company in Hangzhou Zhejiang province China. The product defects were analyzed statistically first. main effects which affects performance found by T-type matrix chart analysis.

10.1109/icieem.2011.6035326 article EN 2011-09-01

The Image Captioning research field is currently compromised by the lack of transparency and awareness over End-of-Sequence token (<Eos>) in Self-Critical Sequence Training. If <Eos> omitted, a model can boost its performance up to +4.1 CIDEr-D using trivial sentence fragments. While this phenomenon poses an obstacle fair evaluation comparison established works, people involved new projects are given arduous choice between lower scores unsatisfactory descriptions due competitive nature...

10.48550/arxiv.2305.12254 preprint EN cc-by-nc-nd arXiv (Cornell University) 2023-01-01
Coming Soon ...