Vinicyus Brasil

Pesquisa

• Reinfocement learning

In my Master’s thesis I studied reinfocement learning, where I developed a new algorithm called QSVR. The algorithm uses a Support Vector Regression algorithm call Online-SVR to estimate the Q-function of a environment. The results show it can perform better than the Deep Q-Network and Q-learning in a environment with small state space and a discrete action space. The code of the project is in this repository.

• Other machine learning algorithms

I also worked on using neural networks along GARCH models in a hybrid approach to improve the forecasting of time series related to soybean price volatility. Another article I worked on uses the framework of Bayesian networks along with time series to predict diseases spread along the Brazilian territory.