Supervised learning has been used in an uncountable number of applications. Face recognition, disaster prediction, climate change detection, are some of applications where they showed incomparable results. However, they heavily depend on the availability of labelled data and generally struggle in results generalisation when labeled data are scarce.
As a result, new methods have been recently purposed, including semi-learning and unsupervised learning. Among them, self-supervised learning showed powerful results in learning meaningful structure, representations, and features from unlabeled data using unsupervised pretext tasks. Derived results can be used for speeding and improving classical supervised classification.
One straightforward application is the digital healthcare. In digital healthcare, the pervasiveness of wearable devices represent a very interesting opportunity for remote patients life-style monitoring, falls or dangerous occurrences prediction, and early disease detection. Unfortunately, the lack of labeled data can distort and even prevent results reliability.
- Implement self-supervised algorithm on time series data, using different pretext tasks and use the result for a machine learning based supervised classification.
- When possible, visualise the original data and the data transformed for the pretext tasks (i.e. if you apply some data rotations, visualise the differences between the original data and the rotated data)
- Compare results obtained in 1. with one or more end-to-end supervised machine learning classification (optional).
- Compare results with more datasets (optional).
- Machine learning
- Deep learning
- Classification
- Unsupervised learning
- Supervised learning
- Python
Self-supervised learning from sensors: https://arxiv.org/pdf/2009.13233v1.pdf [1]
Supervised learning techniques with wearable sensor data (in human Activity Recognition): https://www.researchgate.net/publication/352949846_Trends_in_human_acti…]
Dataset 1 (acceleration and angular velocity of human activity data from smartphone): https://archive.ics.uci.edu/ml/datasets/human+activity+recognition+usin… [3]
Dataset 2 (accelerometer and gyroscope sensors of human activity data from smartphone): https://github.com/mmalekzadeh/motion-sense [4]
UCI Repository: https://archive.ics.uci.edu/ml/datasets.php?format=&task=cla&att=&area=… [5]