Enhancing Real-Time Language Processing via Advanced PET Signal Analysis and Deep Learning
Year: 2025
Type: conference-paper
Abstract: Efficient comprehension of the neural substrates of language processing that characterize the discipline of cognitive science is important. However, it should be acknowledged there are some limitations with the current analytical techniques employed for examining signals stemming from PET based labelling of language processing in real-time; often producing outcomes that may be inaccurate and/or costly in terms of computational resources. The propose a hybrid deep learning model containing LSTM networks and Bidirectional Encoder Representations from Transformers (BERT). BERT has the advantage of pulling contextual embeddings from texts while LSTM retains sequential dependencies; a hybrid model is therefore ideal for process extraction from PET signals. With a composite loss function, BERT embeddings fine-tuned with domain-correct data, dropout regularization optimization, the model demonstrated enhanced stability and robustness. A PET signal dataset, annotated for process was used to train and evaluate the model. Results from the experiment writ-ten using Python demonstrated a BERT-LSTM hybrid model achieved an accuracy of 22% growing accuracy above the customary analyst led approaches used for examining PET signals. The improvement suggests candidate model has the potential to assist with increasing the accuracy of PET signal analyses during language comprehending tasks. The contribution will add to our understanding of the neural processes of language and offers exciting new conceptualizations to contemporaneously analyze language.
Authors: Deepa Nagalavi, Shankar Balla, Pushpalatha. M, Nashwan Adnan Othman, Malik Bader Alazzam, A. Balakumar
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W4414405860