Federated learning allows multiple parties to collaboratively develop a deep learning model, without sharing private data. Models can be generated from the most up-to-date data while taking unique and not publicly available data into account. However, the distributed nature of federated learning causes problems too, and clients are not guaranteed to hold independently identically distributed (iid) data, causing performance degradation. This work analyzes existing methods of generating such skewed datasets and finds that the Earth Movers Distance (EMD) can be used to compare them. A novel scheme called phase-shift is introduced, which allows clients to communicate more frequently, without increasing communication, hereby reducing drift caused by non-iid data. Finally, we propose a data-driven approach that can reduce the data skew by supplementing local datasets with augmented data. A novel method of balancing unaltered and augmented data is introduced, taking the skew of the dataset into account. Empirical analysis shows that phase-shift can reduce the instantaneous communication load on the system by 37.5% without suffering a performance loss or reducing convergence rate. Evaluation of data augmentation on a heavily skewed cifar10 dataset shows that accuracy is improved by 10%. Finally, phase-shift and data augmentation are combined, resulting in a 13% accuracy improvement, surpassing algorithms such as FedNova and FedProx when dealing with label-heterogeneity.

This report describes the design and implementation of a fall detection algorithm for a fall detection system using a pressure based floor sensor. The goal of the system is to detect falls and alarm the relevant personnel when an elderly person has fallen. The fall detection algorithm has a strong connection with the interface subsystem, which uses the algorithm as a function. The interface subsystem supplies matrices containing the raw sensor values of the pressure floor. The algorithm has been divided into multiple sub-algorithms. First, pre-processing: data linearization was applied on the raw sensor values and the sensor matrix was processed such that an image formed that looked like the real world scenario. Second, image processing techniques were applied to detect contours. Contours were being tracked through time, and being grouped. The characteristics of the contours and groups were used to classify falls. Tests have been done to validate the behaviour of the algorithm, from which an average false negative ratio of 30% was achieved in a time window of 30 seconds. The created prototype proves that image processing is a viable tool for detecting falls with the use of a pressure-based floor sensor. Overall, this results in a strong alternative for fall detection that could be used to improve the time an elderly person can live at home safely without the need to move to a nursing home