Online active learning for human activity recognition from sensory data streams

A modular, deep learning-based holistic intent sensing system tested with Parkinson's disease patients and controls

People living with mobility-limiting conditions such as Parkinson's disease can struggle to physically complete intended tasks. Intent-sensing technology can measure and even predict these intended tasks, such that assistive technology could help a user to safely complete them. In prior research, algorithmic systems have been proposed, developed and tested for measuring user intent through a Probabilistic Sensor Network, allowing multiple sensors to be dynamically combined in a modular fashion. A time-segmented deep-learning system has also been presented to predict intent continuously. This study combines these principles, and so proposes, develops and tests a novel algorithm for multi-modal intent sensing, combining measurements from IMU sensors with those from a microphone and interpreting the outputs using time-segmented deep learning. It is tested on a new data set consisting of a mix of non-disabled control volunteers and participants with Parkinson's disease, and used to classify three activities of daily living as quickly and accurately as possible. Results showed intent could be determined with an accuracy of 97.4% within 0.5 s of inception of the idea to act, which subsequently improved monotonically to a maximum of 99.9918% over the course of the activity. This evidence supports the conclusion that intent sensing is viable as a potential input for assistive medical devices.

kNN Prototyping Schemes for Embedded Human Activity Recognition with Online Learning

The kNN machine learning method is widely used as a classifier in Human Activity Recognition (HAR) systems. Although the kNN algorithm works similarly both online and in offline mode, the use of all training instances is much more critical online than offline due to time and memory restrictions in the online mode. Some methods propose decreasing the high computational costs of kNN by focusing, e.g., on approximate kNN solutions such as the ones relying on Locality-Sensitive Hashing (LSH). However, embedded kNN implementations also need to address the target device’s memory constraints, especially as the use of online classification needs to cope with those constraints to be practical. This paper discusses online approaches to reduce the number of training instances stored in the kNN search space. To address practical implementations of HAR systems using kNN, this paper presents simple, energy/computationally efficient, and real-time feasible schemes to maintain at runtime a maximum number of training instances stored by kNN. The proposed schemes include policies for substituting the training instances, maintaining the search space to a maximum size. Experiments in the context of HAR datasets show the efficiency of our best schemes.