Camilla Lingj?rde

Towards Robust Machine Learning through Causal Multi-Task Learning

The primary goal of this project is to advance the field of machine learning by developing new supervised learning methods that exhibit enhanced robustness, particularly in their ability to generalise adequately to new domains. This objective is motivated by the need for reliable machine-learning algorithms that perform consistently in unpredictable real-world conditions. Robust algorithms are essential for mitigating errors in critical applications, such as healthcare and environmental monitoring, and ensuring Artificial Intelligence (AI)'s equitable performance across diverse populations. This enhances the fairness and reliability of machine learning in everyday applications.

This project seeks to integrate causal inference principles with supervised machine learning, addressing the limitation that models trained to minimise standard empirical risk often face - degradation in performance when exposed to domain shifts. Unlike association-based relationships, causal relationships are stable across domains, making them invaluable for predictions in varying contexts and scenarios. This work aspires to increase machine learning's robustness, applicability, versability, computational efficiency, and fairness by being inspired by causality and multi-task learning.