Abstract:
In the present era, human-machine collaboration is increasing each day with more applications of ergonomics and human factors in industrial and socio-technical environments. This has amplified the need for human factors while designing collaborative applications. Among these macro-human factors; human's mental workload (MWL), stress level, and mental cognitive states are vital to consider while planning system safety and risk assessment. Similarly, Brain-Computer Interface (BCI) provides a means of contact between the human brain and external devices by recognizing the person‘s intent using brain generated signals and translating them into external commands and is critical for patients suffering from severe motor disabilities. Cognitive brain signals acquired with functional near-infrared spectroscopy (fNIRS) has come out to be a potential non-invasive neuroimaging solution to monitor brain states for said purposes. Conventionally, the Machine Learning (ML) algorithms are used for the classification of brain states from acquired neuroimaging signals. The difficult part in the conventional ML classification algorithms is feature extraction, feature selection, and dimensionality reduction the neuroimaging data. A novel deep learning (DL) framework is proposed, which utilizes a convolutional neural network (CNN) and recurrent neural network (RNN) variant namely Long Short-Term Memory (LSTM) that solved the feature engineering challenges. However, bypassing the challenges of feature engineering through DL techniques comes at the cost of long training time, the computational complexity of the system, and the requirement for an enormous amount of data for training. The computational complexity of ML and DL algorithms is measured and the appropriate algorithms are suggested in different use cases. The symmetric homogenous instance-based transfer learning method is applied to CNN to solve the complex training time, big data requirement, and calibration time of BCI systems.
