LatticaBOltzmna

Highlighting the latest validated DEEPLBMNet architecture.

- U- LBMNet: 3D Medical Image Segmentation

U-LBMNet is a three-dimensional physics-inspired encoder–decoder architecture for volumetric medical image segmentation. Inspired by the U-shaped network design and the Lattice Boltzmann Method (LBM), the model integrates trainable collision and streaming operators into a hierarchical multi-scale framework to accurately extract anatomical structures from volumetric medical images. The encoder progressively captures local and global contextual information, while the decoder reconstructs high-resolution segmentation maps through skip connections that preserve fine spatial details.

Unlike conventional 3D convolutional networks, U-LBMNet incorporates lattice-based feature propagation, moment transformations, and physics-inspired information flow to improve feature representation and spatial consistency throughout the segmentation process. This design enables efficient learning of complex three-dimensional anatomical patterns while maintaining computational efficiency and robustness.

The architecture is suitable for a wide range of volumetric imaging modalities, including Computed Tomography (CT), Magnetic Resonance Imaging (MRI), Cone-Beam CT (CBCT), and three-dimensional ultrasound. Typical applications include organ segmentation, tumor delineation, lesion detection, vessel segmentation, radiotherapy planning, surgical navigation, disease monitoring, and computer-assisted diagnosis.

U-LBMNet has been designed as a flexible framework that can be adapted to different clinical segmentation tasks by modifying the lattice configuration, encoder depth, decoder capacity, and attention mechanisms while preserving the core principles of the LBMNet family. Its modular design also facilitates the integration of advanced components such as Multiple-Relaxation-Time (MRT) operators, lattice attention modules, and multi-scale feature fusion strategies.

Curve of learning

Nunc sapien mauris, imperdiet ac pellentesque quis, facilisis non sapien. Maecenas congue vehicula mi, id luctus mi scelerisque nec. Cras viverra libero ut velit ullamcorper volutpat. Maecenas ut dolor eget ante interdum auctor quis sed nunc. Proin faucibus, mauris vitae molestie sodales, erat nisi rhoncus justo, in placerat turpis elit sed eros.
Mauris molestie, justo et feugiat rutrum, arcu metus dapibus quam, sollicitudin tempus tortor dolor et nibh.

Case: BRATS_001.nii.gz

Case: BRATS_004.nii.gz

Image Inpainting

LBMNet for Image Inpainting is a physics-inspired deep learning architecture designed to reconstruct missing or damaged regions in images by leveraging the principles of the Lattice Boltzmann Method. Instead of relying solely on conventional convolutional operations, LBMNet employs lattice-based streaming, collision, and moment computations to propagate structural and contextual information across corrupted areas. This physics-guided framework enables the network to preserve edges, textures, and fine details while generating visually coherent reconstructions. Its lightweight, interpretable, and scalable design makes LBMNet an efficient solution for image restoration in applications such as photo editing, medical imaging, remote sensing, and cultural heritage preservation.

Object Detection

LBMNet for Object Detection is a physics-inspired object detection architecture that leverages the principles of the Lattice Boltzmann Method to accurately localize and classify objects in images. By integrating lattice-based streaming, collision, and moment computations into the feature extraction process, LBMNet efficiently captures both local structures and global contextual information. Its lightweight, interpretable, and scalable design enables robust object detection with high computational efficiency, making it well suited for applications such as autonomous driving, medical imaging, remote sensing, industrial inspection, and intelligent surveillance.

Image Classification

LBMNet for Image Classification is a physics-inspired deep learning architecture that applies the principles of the Lattice Boltzmann Method to image classification. By integrating lattice-based streaming, collision, and moment computations into hierarchical feature learning, the network effectively captures rich spatial and directional representations. Its lightweight, interpretable, and scalable design provides an efficient alternative to conventional convolutional neural networks, delivering accurate image classification across a wide range of computer vision applications.

Face recognition

LBMNet for Face Recognition is a physics-inspired deep learning architecture that applies the principles of the Lattice Boltzmann Method to robust facial representation learning. By integrating lattice-based streaming, collision, and moment computations, the network effectively captures discriminative facial features while preserving both local textures and global structural information. Its lightweight, interpretable, and scalable design provides an efficient solution for accurate face recognition under variations in pose, illumination, expression, and occlusion, making it suitable for biometric authentication, access control, surveillance, and identity verification applications.

Image colorization

LBMNet for Image Colorization is a physics-inspired deep learning architecture that automatically transforms grayscale images into realistic color images using the principles of the Lattice Boltzmann Method. By integrating lattice-based streaming, collision, and moment computations into the feature learning process, the network effectively captures semantic, structural, and contextual information to generate natural and visually consistent colors. Its lightweight, interpretable, and scalable design makes LBMNet an efficient solution for image colorization in applications such as historical photo restoration, medical imaging, digital media enhancement, and image editing.

Medical Imaging Segmentation

LBMNet for Medical Image Segmentation is a physics-inspired deep learning architecture that applies the principles of the Lattice Boltzmann Method to accurately segment anatomical structures and pathological regions in medical images. By integrating lattice-based streaming, collision, and moment computations with hierarchical feature learning, the network effectively captures both fine anatomical details and global contextual information. Its lightweight, interpretable, and scalable design enables precise and efficient segmentation across a wide range of medical imaging modalities, including MRI, CT, ultrasound, and X-ray, supporting applications in computer-aided diagnosis, treatment planning, and clinical decision-making.

Super Resolution

LBMNet for Image Super-Resolution is a physics-inspired deep learning architecture that reconstructs high-resolution images from low-resolution inputs using the principles of the Lattice Boltzmann Method. By integrating lattice-based streaming, collision, and moment computations into the feature extraction process, the network effectively restores fine textures, sharp edges, and structural details while preserving visual consistency. Its lightweight, interpretable, and scalable design provides an efficient solution for image super-resolution across a wide range of applications, including medical imaging, remote sensing, surveillance, and digital photography.

Medical MRI Image Denoising

LBMNet for Medical MRI Image Denoising is a physics-inspired deep learning architecture designed to remove noise from magnetic resonance imaging (MRI) while preserving fine anatomical structures and diagnostic details. By integrating the principles of the Lattice Boltzmann Method through lattice-based streaming, collision, and moment computations, the network effectively suppresses noise while maintaining image sharpness and tissue boundaries. Its lightweight, interpretable, and scalable design provides an efficient solution for high-quality MRI enhancement, supporting improved visualization, diagnosis, and subsequent medical image analysis.

Journal Articles

- A Lattice Boltzmann Method for Image Inpainting Inspired by Fluid Dynamics
Authors: Yassine Douich, et al.Journal: European Journal of Pure and Applied MathematicsYear: 2025DOI:10.29020/nybg.ejpam.v18i3.6192- Eikonal equation solved with a novel Lattice Boltzmann method framework
Authors: Yassine Douich, et al.Journal: Numerical AlgorithmsYear:2026DOI:10.1007/s11075-026-02378-9- Radial Basis Function Neural Networks for Collision Learning and Mask Detection in Image Inpainting
Authors: Yassine Douich, et al.Conference paper:Communications in Computer and Information Science ((CCIS,volume 2817)).DOI: 10.1007/978-3-032-16281-6_10

Manuscripts Under Review

- Neural Collision Learning for Image Inpainting via LBM Transport and CNN Mask Detection
Authors: Yassine Douich, et al.Journal:Signal, Image and Video Processing.Year:2026
- An Artificial Intelligence Architecture Based on Lattice Boltzmann Dynamics for Image Inpainting
Authors: Yassine Douich, et al.Journal:IEEE Transactions on Artificial IntelligenceYear:2026
- From Convolution to Transport: A Lattice Boltzmann Reformulation of Deep Neural Networks
Authors: Yassine Douich, et al.Journal: IEEE AccessYear:2026

Research Areas

- Physics-Inspired Deep Learning
- Lattice Boltzmann Neural Networks (LBMNet)
- Computer Vision
- Medical Image Analysis
- Image Restoration and Inpainting
- Image Segmentation
- Image Super-Resolution
- Image Colorization
- Object Detection
- Face Recognition
- Pattern Recognition
- Scientific Machine Learning (SciML)
- Partial Differential Equations (PDEs)
- Fluid Dynamics-Inspired AI
- Medical Imaging
- AI for Healthcare
- Explainable and Efficient AI

Our Vision

We believe that future AI systems should not only achieve high performance but also incorporate scientific knowledge into their design. LBMNet aims to establish a new family of neural architectures inspired by the dynamics of the Lattice Boltzmann Method, offering improved efficiency, robustness, and physical interpretability for a wide range of computer vision and scientific computing tasks.

Current Research

Our ongoing research explores novel LBMNet architectures for:
- Image Inpainting

- Medical Image Segmentation

- MRI Denoising

- Super-Resolution

- Object Detection

- Image Classification

- Face Recognition

- Image Colorization

- Scientific Simulation

- PDE Learning and Solvers

Several research projects are currently under development and will be published in international journals and conferences as they become available.

Open Research

LBMNet is an active research initiative. We welcome collaborations with universities, research laboratories, industry partners, and students interested in physics-informed AI, computer vision, medical imaging, and scientific machine learning. Together, we aim to advance the future of intelligent systems through innovative, mathematically grounded neural architectures.

Getting Started

Learn the fundamentals of LBMNet, its philosophy, and the core concepts behind physics-inspired deep learning based on the Lattice Boltzmann Method.

Installation

Follow the installation guide to set up the required environment, dependencies, and software needed to run LBMNet models.

Architecture Overview

Discover the design principles of LBMNet and understand how Lattice Boltzmann dynamics are integrated into modern neural network architectures.

Tutorials

Step-by-step tutorials demonstrate how to train, evaluate, and apply LBMNet to different computer vision and scientific computing tasks.

Model Zoo

Browse available LBMNet architectures, including models for image restoration, segmentation, object detection, classification, super-resolution, and other AI applications.

Datasets

Access information about the datasets commonly used for training and evaluating LBMNet models, along with recommended preprocessing procedures.

API Reference

Explore the complete API documentation, including modules, classes, functions, parameters, and usage examples for developers.

Examples

A collection of practical examples illustrates how to use LBMNet in real-world projects, from research experiments to production-ready applications.

Performance Benchmarks

Compare the performance of LBMNet models across different datasets and tasks using standard evaluation metrics.

Frequently Asked Questions

Find answers to common questions regarding installation, training, inference, troubleshooting, and best practices.

Contributing

We encourage contributions from the research and open-source communities. Whether you are reporting bugs, improving documentation, or developing new LBMNet modules, your contributions are welcome.

Support

If you encounter any issues or have questions about LBMNet, please visit the Help & Support page or contact the development team.

Documentation Status

The LBMNet documentation is continuously evolving. New tutorials, examples, APIs, research updates, and implementation guides will be added regularly as the framework grows and new architectures become available.

Founder of LBMNet

Ph.D. Candidate in Computer Science

Vision

To establish LBMNet as a global reference for physics-inspired artificial intelligence by developing innovative deep learning architectures grounded in the mathematical principles of the Lattice Boltzmann Method.

Mission

- Advance the frontiers of Physics-Inspired Artificial Intelligence.
- Develop novel Lattice Boltzmann-based neural architectures.
- Bridge numerical physics and modern deep learning.
- Promote open scientific research and reproducible AI.
- Support researchers, students, and engineers through open-source software, publications, and educational resources.
- Foster international collaborations in Artificial Intelligence and Scientific Computing.

Research Areas

- Physics-Inspired Artificial Intelligence
- Scientific Machine Learning
- Lattice Boltzmann Deep Learning
- Computer Vision
- Medical Image Analysis
- Image Restoration and Inpainting
- Image Segmentation
- Biomedical Signal Processing
- Time-Series Forecasting
- Scientific Computing
- Numerical Methods for Partial Differential Equations

A Personal Commitment

LBMNet represents more than a research project. It reflects a long-term commitment to advancing scientific innovation by combining mathematics, physics, and artificial intelligence. The goal is not only to develop high-performance AI models but also to contribute to a deeper scientific understanding of intelligent systems inspired by the laws of nature.

CV

For additional information regarding my academic background, research experience, publications, technical skills, and professional achievements, please refer to my complete Curriculum Vitae. The CV provides a comprehensive overview of my education, research activities, scientific contributions, and ongoing projects within the LBMNet framework.

CV

Stay Updated with the Latest Developments

Welcome to the LBMNet News page. Here you will find the latest updates on our research, new architectures, scientific publications, collaborations, software releases, and upcoming events. Follow our journey as we continue advancing physics-inspired artificial intelligence based on the Lattice Boltzmann Method.

LBMNet Website Officially Launched

he official LBMNet website is now online, providing access to our research vision, architectures, applications, and future scientific contributions.

New LBMNet Architectures

Several new LBMNet architectures have been developed for applications including image classification, semantic segmentation, image inpainting, super-resolution, medical image analysis, cybersecurity, natural language processing, and object detection.

Research Papers in Progress

Multiple LBMNet research papers are currently under review or in preparation for publication in international scientific journals. Additional results and resources will be released after publication.

Expanding Research Applications

LBMNet is being explored across multiple domains, including computer vision, medical imaging, cybersecurity, scientific computing, remote sensing, and artificial intelligence for healthcare.

Join Our Community

We welcome researchers, students, and collaborators interested in physics-inspired AI. Feel free to contact us for collaborations, discussions, or research opportunities

Coming Soon

Scientific publications Open-source implementations
Experimental benchmarks
Tutorials and demonstrations
Research collaborations
New LBMNet architectures

Stay Connected

More exciting updates are coming soon as LBMNet continues to evolve.

Advancing Physics-Inspired Artificial Intelligence

LBMNet is an independent research initiative dedicated to developing the next generation of physics-inspired deep learning architectures based on the Lattice Boltzmann Method (LBM). Our mission is to bridge computational physics and artificial intelligence to create innovative, interpretable, and efficient solutions for real-world challenges.

Why Donate?

Your contribution directly supports:

Development of new LBMNet architectures
GPU cloud computing and computational resources
Scientific publications and open research
AI research in medical imaging, computer vision, NLP, cybersecurity, robotics, and scientific computing
Building an open and collaborative research ecosystem

Every donation, regardless of its size, helps move the project forward.

Make a Contribution

Choose any amount you'd like to contribute. $5 — Support the Project $10 — AI Research $20 — GPU Computing 50 — LBMNet Development $100 — Research Sponsor Custom Amount

Secure payment via PayPal.

Transparency

All donations are used exclusively to support the development of LBMNet, including computational resources, research activities, software development, scientific publications, and educational content.

Thank You

Every contribution helps us push the boundaries of physics-inspired artificial intelligence.

Thank you for supporting LBMNet and helping shape the future of AI research.

Together, we are building the next generation of intelligent systems inspired by physics.

LBMNet for classification

Deep LBM-Based Neural Network is a physics-inspired deep learning architecture that replaces conventional convolutional operations with Lattice Boltzmann dynamics. By integrating streaming, collision, and moment computation into a hierarchical framework, the network efficiently learns multi-scale feature representations while maintaining high computational efficiency and interpretability for image classification tasks.

Quantum LBMNet for clasification

Quantum-LBMNet is a novel quantum-inspired extension of LBMNet that processes images using complex-valued (real and imaginary) feature representations and lattice Boltzmann dynamics. By combining physics-inspired computation with quantum-inspired encoding, the architecture efficiently captures directional and spatial information before generating compact feature representations for accurate image classification. Its lightweight design offers an excellent balance between computational efficiency, scalability, and predictive performance.

LBMNet for Natural Language Processing (NLP)

LBMNet for Natural Language Processing (NLP) is a physics-inspired neural architecture that extends the principles of the Lattice Boltzmann Method to language understanding and generation. By combining multi-scale LBM encoding, moment-based feature extraction, and attention mechanisms, the network effectively captures local and global linguistic dependencies. Its hierarchical design enables efficient representation learning for a wide range of NLP tasks, including text classification, machine translation, sentiment analysis, question answering, and next-word prediction.

Attention U-LBMNet for image inpainting

Attention U-LBMNet is a U-shaped Lattice Boltzmann neural network designed for image inpainting. The architecture combines LBM-based encoder and decoder blocks with attention mechanisms to effectively reconstruct missing image regions while preserving structural details and textures. Skip connections and attention gates enable efficient feature fusion across multiple scales, resulting in accurate and visually coherent image restoration. Its physics-inspired design provides an interpretable, scalable, and efficient framework for image reconstruction and related image restoration tasks.

U-LBMNet

U-LBMNet is a U-shaped neural network architecture that integrates the principles of the Lattice Boltzmann Method into an encoder–decoder framework for dense image prediction tasks. By replacing conventional feature extraction with LBM streaming, collision, and moment operations, the network efficiently learns hierarchical multi-scale representations while preserving fine spatial details through skip connections. Its physics-inspired design provides an interpretable, scalable, and computationally efficient solution for semantic segmentation, medical image analysis, and other pixel-level computer vision applications.

MRT-LBMSegNet-Att

MRT-LBMSegNet-Att is a physics-inspired semantic segmentation architecture that combines Multi-Relaxation-Time (MRT) Lattice Boltzmann blocks with attention mechanisms in an encoder–decoder framework. By integrating MRT collision dynamics, lattice-based feature propagation, and multi-scale attention, the network effectively captures both local details and global contextual information. Skip connections preserve fine spatial features, enabling accurate and robust segmentation while maintaining high computational efficiency and interpretability across a wide range of medical and computer vision applications.

Fast Quantum-LBM-SR

Fast Quantum-LBM-SR is a physics-inspired image super-resolution architecture that integrates quantum-inspired Lattice Boltzmann dynamics with deep learning to reconstruct high-resolution images from low-resolution inputs. By combining streaming, MRT collision, moment computation, and efficient reconstruction within a lightweight framework, the network effectively restores fine textures and structural details while maintaining low computational complexity. Its scalable and interpretable design provides an efficient alternative to conventional super-resolution models for both natural and medical imaging applications.

LBMNet_cybersecurity

LBMNet-D2Q9 is a physics-inspired intrusion detection architecture that applies the principles of the Lattice Boltzmann Method to cybersecurity. By combining D2Q9 lattice encoding, streaming, Multi-Relaxation-Time (MRT) collision, and moment-based feature extraction, the network effectively learns complex traffic patterns for accurate attack detection and classification. Its lightweight and interpretable design provides a scalable and efficient solution for modern network intrusion detection systems, delivering robust performance across diverse cyberattack scenarios.

Graph-MRT-LBMNet

Graph-MRT-LBMNet is a physics-inspired graph neural network that combines graph representation learning with Multi-Relaxation-Time (MRT) Lattice Boltzmann dynamics for cybersecurity applications. By integrating graph-based information propagation, MRT collision, and moment-based feature extraction, the architecture effectively captures complex relationships between network connections for accurate intrusion detection. Its scalable and interpretable design provides a robust and efficient framework for detecting cyber threats in modern network environments.