Name: GABRIEL INÁCIO BARBOZA
Publication date: 09/07/2026
Examining board:
Name |
Role |
|---|---|
| ISAAC PINHEIRO DOS SANTOS | Presidente |
| LUCIA CATABRIGA | Coorientador |
| VINICIUS DE CARVALHO RISPOLI | Examinador Externo |
| WELLINGTON BETENCURTE DA SILVA | Examinador Externo |
Summary: Although epidemiological data were already being widely collected before the COVID-
19 pandemic, the outbreak that began in 2020 significantly boosted the availability,
dissemination, and use of data related to the spread of infectious diseases. Understanding
these processes is essential to support decision-making and the formulation of public
policies aimed at containing and mitigating their impacts. Compartmental epidemiological
models, based on ordinary differential equations, are widely used to describe the dynamics
of infectious diseases due to their simplicity, interpretability, and the availability of efficient
numerical methods for their solution. With the advancement of machine learning and
data science, techniques capable of extracting information and identifying mathematical
structures from large volumes of data have emerged. In this context, Physics-Informed
Neural Networks (PINNs) stand out for incorporating the governing equations directly into
the loss function, acting as universal approximators regularized by physical knowledge. In
addition to the direct solution of differential equations, these networks can also be applied
to solving inverse problems. In this work, PINNs are used for parameter inference in
compartmental epidemiological models. Initially, experiments are conducted with synthetic
data to evaluate the method’s ability to solve inverse problems and its robustness to the
presence of noise. Subsequently, the method is applied to real influenza data, aiming to
estimate the temporal evolution of epidemiological parameters, such as the infectiousness
transmission rate.
