Learning Representations using artificial neural networks and complex networks with applications in sensors and biosensors

• Grant number: 21/07289-2
• Support type: Scholarships in Brazil - Post-Doctorate
• Effective date (Start): February 01, 2022
• Effective date (End): January 31, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal researcher: Odemir Martinez Bruno
• Grantee: Lucas Correia Ribas
• Home Institution: Instituto de Física de São Carlos (IFSC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated research grant: 18/22214-6 - Towards a convergence of technologies: from sensing and biosensing to information visualization and machine learning for data analysis in clinical diagnosis, AP.TEM

Abstract

Great efforts have been made by researchers from different areas to develop sensors and biosensors for various purposes such as early cancer diagnosis, detection of contamination in food/water, etc. In this sense, many researchers have focused on the development of new low-cost detection and diagnosis strategies. Particularly, promising results have been achieved using machine learning techniques and computer vision in sensory unit images. Although many works have used computer vision techniques in biological images, so far there are not many studies in the literature that explore sensor and biosensor images. This postdoctoral project aims to develop new representation learning approaches for image characterization using artificial neural networks (ANNs) and complex networks with direct application to biosensor and sensor images. Regarding the research aspects of computational methods for image characterization, we intend to investigate ways to learn representations (ie, feature vectors) with ANNs using as input source attributes from (or the own) image or the modeled complex network of combined mode. In particular, complex networks stand out for their ability to model complex texture patterns that we observed in sensory unit images.Regarding the ANNs, besides the deep neural networks, we will use the random neural networks due to their simplicity, high predictive performance and fast learning algorithm. We also highlight the \textit{vision transformers} architecture that, despite the promising results in recent works of scene/object recognition, have not yet been fully explored in problems of texture and microscopy image analysis and complex networks. The computational methods developed will be used to characterize the images of sensory units to design new strategies for applications such as early diagnosis of cancer and detection of viruses or contaminations. Such data will be generated and provided by the various collaborators in the thematic project financed by FAPESP (process, 2018/22214-6) in which this post-doctorate project is linked. We expect that this research will result in relevant contributions in the computational scope with new techniques for image analysis and related to physical chemistry and applications with new detection and diagnosis strategies. It should be noted that this second contribution is congruent with the objectives of the thematic project.