Possible Applications of Machine Learning Techniques in Neural Data Classification

Bhagwat, Rohit Uttam (2017) Possible Applications of Machine Learning Techniques in Neural Data Classification. Masters thesis, Indian Institute of Science Education and Research Kolkata.

PDF (MS dissertation of Rohit Uttam Bhagwat (12MS055)) Rohit_Bhagwat__(12MS055)_MS_Thesis.pdf - Submitted Version Restricted to Repository staff only Download (1MB)

Abstract

Machine learning or ML has emerged as a new and effective tool for making predictions. The focus of advanced ML-techniques is to gain a useful insight from complex data. With the advent of new technology and optimised algorithms, it is now possible to build complex models using these techniques and test them on a particular research problem involving prediction or a classification task. Deep learning is a particular ML-technique which has been quite successful in tasks related to image classification, handwritten digit recognition and others. To handle sequential data in prediction and classification tasks a special type of Recurrent Neural Networks (RNNs), LSTMs have shown good results. This thesis work explores these ML-techniques and their possible applications in the area of cognitive neuroscience by designing a classifier for object recognition stimuli. The EEG data records brain waves and inherently has spatiotemporal nature. Moreover, MLmethods based on statistical properties (statistical ML) often depends on the good representation of raw data. In general, finding these suitable representations is difficult, especially for high dimensional data such as EEG. Hence, the exploration of ML-techniques that are well known for learning spatial as well as temporal features viz. Deep learning, LSTMs etc. is necessary. In summary, this work focuses on, A] exploration of ML & various ML techniques (Chapter 2-4) and B] possible applications of these techniques (Chapter 5-6), in the neural data classification.

Actions (login required)

View Item