Titulo Estágio
Using Explainable AI to Disentangle Economic Cycles
Local do Estágio
Laboratório de Informática Industrial e Sistemas do DEI/CISUC (LIIS@DEI-FCTUC)
Enquadramento
The main objective of central banks' mandates is to ensure effective price stability. Achieving this goal is not straightforward, and sometimes trade-offs are necessary, as efforts to contain price surges can hinder economic growth. Extensive research supports the monetary policy decision-making process, yet despite the application of statistical modeling and signal processing tools, our understanding of economic cycles remains incomplete.
With recent advancements in machine learning approaches providing increased performance, their opaque black-box nature presents substantial ethical, legal, and model governance challenges. Addressing these concerns is crucial to complying with regulations, fostering trust and ensure ethical use of AI models.
Given that the financial sector is characterized by risk aversion, it is important to investigate Explainable AI (XAI) techniques to provide insights into the patterns learned by models, allowing stakeholders to understand the factors driving predictions, assess model biases, and comply with regulatory guidelines.
Objetivo
The main objective of this research is the development of XAI methods to disentangle economic cycles and ensure that the models used are not only accurate but also explainable.
Additionally, a key goal is to train an artificial neural network model that successfully identifies the beginning of clusters of volatile price changes or convergences in time horizons where systemic risk is heightened. This would be in line with the principle proposed by Mandelbrot, which suggests that large changes in prices tend to cluster together.
Through these implementations, this system will be capable of explaining the identification of periods of heightened systemic risk, providing policymakers with a clear understanding of the patterns learned by the model.
For the development of this work, existing datasets will be considered, such as those provided by the Open Data World Bank, the International Monetary Fund (IFM) and, in particular, those provided by the ongoing PRR CRAI project, offering a complete range of economic data relevant for this work.
Plano de Trabalhos - Semestre 1
1st Semester:
1. Literature review on economic theories (Kondratieff, Kuznets, Kitchin cycles, etc.);
2. Explore adjacent approaches (Fractal and Chaos theory);
3. Explore current SOTA on ML and signal processing tools applied to pattern recognition;
5. Train a machine learning model with these concepts;
6. Write the documentation.
Plano de Trabalhos - Semestre 2
2nd Semester:
7. Explore current SOTA of XAI in pattern recognition models;
8. Implement XAI methods in the system;
9. Test and validation of the developed models;
10. Prepare a scientific article;
11. Prepare the documentation and final Dissertation.
Condições
The work will be developed within the scope of the PRR CRAI project, and a workplace will be made available at LIIS-CISUC.
Existing datasets will be considered for the development of this work, such as those provided by the Open Data World Bank, the International Monetary Fund (IFM) and, in particular, those provided by the ongoing PRR CRAI project, offering a complete range of economic data relevant for this work.
Observações
The internship may benefit from the award of a research grant for graduates for a period of 6 months, possibly renewable, supported by the ongoing project.
Orientador
Alberto Cardoso; Bruno Direito Leitão
alberto@dei.uc.pt 📩