| dc.description.abstract | Electroencephalogram (EEG) analysis has garnered attention in the research domain
due to its ability to detect various neural activities starting from brain seizures to a
person’s concentration level. To make it more beneficial for the users, it is important
to miniaturize the currently available clinical-grade large EEG monitors to wearables
which can provide decisions at the edge. Traditionally, for performing such analysis, the
raw EEG signals are collected at the edge which is then transferred to the cloud where
the data is interpreted and forwarded accordingly. However, this method of transferring
the user data for analysis poses a risk of security and privacy breach as well as consumes
a considerable bandwidth and time which makes it inefficient in terms of scalability.
In this vein, we investigated on transferring the Artificial Intelligence (AI)-logic of the
analysis to the sensors, so that a localized decision can be made on the edge, without
transferring the data, thus saving precious bandwidth and restoring privacy of the users.
However, the main challenge in achieving such a scenario is the devices’ inability to
perform complex computations due to its resource constraints. Hence, we have explored
various AI-based techniques throughout this thesis to find out a lightweight model
which will be able to give a decent performance while consuming lower resources. We
have validated our candidate models in various use-cases throughout the chapters to
compute the performance of the AI models. It is believed that, this type of analysis can
encourage the sensor foundries to integrate AI-logic with wearable sensors, to conduct
localized EEG analysis on the sensor level, which will be more practical, cheaper, and
scalable. | en_US |
