Roberto Marcondes Cesar Junior.

The convergence of Artificial Intelligence (AI) with cloud-based 5G networks creates unprecedented research opportunities to develop innovative applications in intelligent systems. However, integrating cutting-edge AI techniques into 5G networks brings several significant challenges, particularly in data collection and management, scalability of model training and deployment, interoperability, resource constraints at the edge, latency and real-time processing, security and privacy concerns, bandwidth management, data transfer costs, and energy efficiency and sustainability. To address these challenges, the Center for Innovation in 5G Networks and Artificial Intelligence aims to advance research at the intersection of 5G and AI. The Center will explore technical solutions and establish a comprehensive environment for experimentation and testing, fostering the development of transformative AI applications that leverage the potential of 5G networks. This proposal strongly emphasizes research, technology development, and experimentation to define requirements, evaluate trade-offs, and support prototyping under real-world conditions. To facilitate this, a 5G Smart Campus Living Lab will be established at the USP Cidade Universitária campus, where approximately 70,000 people share a 4.1 km area daily. Additionally, the Industry 4.0 laboratories at Fábrica do Futuro (USP) and the Instituto de Pesquisas Tecnológicas (IPT) will be transformed into 5G demonstrators. The proposal takes a multidisciplinary approach, bringing together domain experts in urban, industrial, and agricultural fields with specialists in networks and AI. The project is organized into four Lines of Research (LR): one cross-domain, supporting frameworks and solutions for deploying AI in wireless networks, and three domain-specific, targeting Smart Cities, Industry 4.0, and urban agriculture. The Technology Transfer Office will establish an ecosystem of partners to collaborate with the Center, leveraging the Campus Living Lab and Demonstration facilities. This collaborative approach aims to expand the project's impact, enhancing quality of life and boosting productivity across various sectors.

Computer Vision methods are used to extract information from images and videos, but their contextual elements are not always sufficient to extract correct and accurate information. In these cases, content from other sources and types of data such as audio and text, or other information external to the data, such as a priori knowledge, can be used to complement and enrich the context of the information of interest. Additionally, the application context can impose various restrictions such as hardware limitations, the need to guarantee privacy, among others. Therefore, modern Computer Vision methods need to be able to automatically integrate the contextual elements of the information of interest and also those related to the application in question. The objective of this project is the development of computer vision models and methods that are capable of generating context-rich representations. The project will be organized around three main integrated research lines: (i) Optimum use of unsupervised data; (ii) Alignment of multi-modal domains; (iii) Properties of representations. Of special interest are computer vision applications involving edge devices (edge computing) and mobile devices (such as smartphones and mini-computers). To develop, test and validate the methods, we intend to build an experimental setup consisting of multiple cameras and sensors that will allow the construction of supervised datasets to be explored by the group.

This project focuses on a unified strategy for knowledge and emerging dynamics discovery in Computational Science using intermediate representations. The intended applications are in areas characterized by large volumes of data in which knowledge discovery implies the transition from raw data bases for intermediate representations (usually feature vectors and graphs), thus allowing for the subsequent use of different analytical methods. In this context, integration and transformation methods to be used in the generation of intermediate data should also ensure the quality and reliability of data generated for the intermediate representation. The results of the analysis phase may influence both experiments and the integration methods for generating new data by feedback mechanisms. This project has two general goals: 1) to develop methodologies to solve Computational Science problems based on a common approach of intermediate mathematical-computational representations; 2) to apply the developed methodologies to different scientific problems, thus creating specific solutions to each problem. This methodological strategy will be used to address specific problems in areas which our group has been working in recent years: intermediate representations in computer vision and urban informatics; study of biological networks dynamics to characterize the mechanisms of the health-disease transition; development of computational tools for processing of MRI images high field and their integration with biological data; development of new techniques for characterization and visualization of intermediate representations in complex dynamic networks, with applications in Systems Biology. (AU)

Modern science is interdisciplinary and data-intensive. For instance, in the 1000 Genomes Project (www.1000genomes.org), the comparative study of 629 individuals has already generated 7.3 TB of data. Analogous situations exist in fields such as astronomy, agriculture, social sciences, etc. Ten years ago, the problem was how to obtain data. Today, the bottleneck is the need for new computational strategies and tools so that scientists can manage these massive volumes of heterogeneous, distributed, data, so that they can generate new knowledge from the processing, analysis and visualization of the data. This launched the basis of the so-called eScience: the combination of advanced research in computer science and mathematical modeling to allow and accelerate research in other knowledge domains. National programs in eScience have been created in the US, GB, Australia and other countries, that recognized the importance of this theme for the advancement of science. The main goal of this project is the design and construction of a collaborative network for research in eScience, in a partnership that involves computer science, mathematical modeling and specific domains in the exact, life, agricultural sciences and social sciences.

A ciência moderna é crescentemente interdisciplinar e intensiva em dados. Na área de ciências da vida, por exemplo, com o surgimento de plataformas de alto desempenho para análise de imagens e estudos genômicos, o gargalo não está mais na aquisição de dados, mas sim no seu armazenamento, processamento, análise e visualização. Este cenário levou ao surgimento de um novo campo de pesquisa - eScience - que combina pesquisa avançada em computação e em modelagem matemática para permitir e acelerar pesquisa em outros domínios do conhecimento, desde as ciências exatas até as humanidades e artes. A eScience envolve a chamada "computação centrada em dados" (data-intensive computing), com a busca de soluções para gerenciamento de grandes volumes de dados produzidos por (e para) experimentos científicos, para que a descoberta científica não venha a ser detida pelo "dilúvio de dados". Este projeto visa a criação de uma rede colaborativa de eScience para acelerar pesquisa avançada em ciências da vida (biologia, medicina, oceanografia) e ciências agrárias. Está estruturado em tomo de cinco linhas de pesquisa - biologia de sistemas, planejamento de safras, computação visual, modelagem matemática e bancos de dados. Dentro dessas linhas, serão tratadas questões em aberto associadas às principais componentes de um ambiente de pesquisa em eScience: armazenamento, processamento, análise e visualização de grandes volumes de dados científicos. Os pesquisadores principais têm histórico de cooperação e coordenação de projetos nessas linhas. Questões de interoperalidade permeiam todo o projeto.

Projeto em colaboração com a Professora Isabelle Bloch (ENST-Paris): O raciocínio espacial em imagens é uma área de pesquisa ainda muito pouco desenvolvida, 
necessitando da criação de modelos de representação das entidades espaciais, de suas
relações e de sua integração em modos de raciocínio. A importância das relações espaciais para o
reconhecimento estrutural de padrões já foi mostrada em diversas situações. Entretanto, tais
relações possuem diferentes significados dependendo da forma dos objetos envolvidos, podendo ainda
variar com o tempo em seqüências de vídeo. Este projeto prevê o desenvolvimento de modelos matemáticos
para tais relações, bem como de algoritmos de reconhecimento que os explorem no contexto de análise
de imagens. Está previsto o desenvolvimento de modelos matemáticos baseados em conjuntos nebulosos que permitam
a exploração de ferramentas de fusão de informações e de raciocínio. Tais modelos serão integrados
a representações por grafos para o raciocínio e o reconhecimento de padrões. Aplicações previstas 
incluem o processamento de imagens médicas do cérebro e o reconhecimento de faces em seqüências
de vídeo.

Reconhecimento de faces e expressões faciais em seqüências de vídeo usando informação 3D

Reconhecimento estrutural de padrões.

Análise de configuração espacial de objetos e rastreamento em seqüências de vídeo.

Aplicações em rede GIGA.

Análise de formas e reconhecimento de padrões.

Reconhecimento estrutural de padrões através de casamento de grafos relacionais com atributos.

Métodos de reconhecimento de padrões em bioinformática.

Análise de formas representadas por landmarks.

Desenvolvimento de métodos para análise multiescala de formas.