Machine-learning techniques have been considered in many application domains, including lnternet of Things (loT) systems. The adoption of machine learning in loT systems creates several new opportunities, e.g., detection of health abnormalities using wearable devices. However, enabling machine learning in the loT domain also involves several challenges inherent to these systems, e.g., privacy. My project will focus on the key privacy challenges in the adoption of machine-learning techniques in the loT domain as well as novel privacy-preserving machine-learning techniques to tackle these challenges. In my opinion, there are three interesting research directions to be explored within my project: 1. Adjusting Federated Learning so that it can be applied to more scenarios, e.g. non-IID data distributions, unsupervised learning 2. Making Federated Learning more privacy-preserving, e.g. through homeomorphic encryption, secure multi-party computation 3. Improving the compliance of Machine Learning pipelines with GDPR regulations, e.g. through machine unlearning