GAPS

GAPS

Signal Processing Aplications Group (Grupo de Aplicaciones del Procesado de Señal)

Join us

Machine Learning methods for health

We have open positions for PhD candidates, as well as TFM/TFG students. If you are interested in any of these lines, do not hesitate to contact us!

Deep Generative models

We use Deep Generative Models, such as Variational Autoencoders, Large Language Models, and diffusion methods, in order to generate synthetic patients, predict the patient’s mortality, or the predicted effect of a treatment.

Federated learning

Real patient’s health records are privately stored at different hospitals. Federated learning aims at training a model using data scattered among different hospitals, without compromising the privacy, and enhancing the performance compared to training using the patients of each hospital in an isolated way.

Few shot learning

Although Machine Learning methods usually focus on Big Data settings, where many patients are available, the reality is that real data is often scarce and nooisy. We aim to design Machine Learning methods focused on the few-shot problem: learning as good as possible when only a small number of patients is available.

Optimal control

In this line, we aim to design optimal therapies for patients, taking into account the patient evolution, using techniques from optimal control theory, such as Deep Reinforcement Learning.