Advances in Nonlinear PDE-based Models for Artificial Intelligence and Computer Vision
|
Organizer(s): |
Name:
|
Affiliation:
|
Country:
|
Tudor Barbu
|
Institute of Computer Science, Romanian Academy – Iasi Branch
|
Romania
|
|
|
|
|
|
|
|
|
|
Introduction:
| | The nonlinear partial differential equations (PDE), that have long been used to formulate various dynamical phenomena, have been applied successfully in many important sub-domains of Artificial Intelligence (AI) in the last four decades. They include image and video processing and analysis and computer vision fields, such as image/video filtering, inpainting, segmentation, compression, decomposition, registration and motion estimation. The PDEs can be also used successfully to create scale-space representations that are used for various multi-scale image analysis tasks.
An important AI sub-domain, Deep Learning, represents an application area of the PDEs, too. The neural partial differential equations (NPDE) could describe AI systems and the architectures of some deep models, like recurrent neural networks (RNN), may be interpreted as nonlinear PDEs. Some evolution partial differential equation could be learned from certain datasets using deep neural networks, that also predict their dynamical behavior. The Convolutional Neural Networks (CNN) are increasingly used for solving nonlinear diffusion-based models and are also used in connection to PDEs to solve many computer vision tasks.
This special session aims to disseminate advanced and original research in these PDE-based AI and CV areas, bring together researchers working in these fields and promote exchange of valuable ideas between them.
|
|
|
|
| |
