Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neurocognitive Modeling for Text Generation: Deep Learning Architecture for EEG Data

Khushiyant

TL;DR

This work addresses the challenge of EEG-based text generation by introducing a data- and compute-efficient framework that couples an RNN-based EEG feature extractor with a classifier-LLM pipeline built around the Gemma 2B model. By extracting compact EEG representations and leveraging pre-trained language modeling capabilities, the approach enables EEG-conditioned text generation with substantially less training data than traditional end-to-end methods, while maintaining competitive performance and robustness to resource constraints. The study demonstrates notable results on the ImageNet of the Brain dataset, including 2-class and 3-class classification performance and perplexity-based text generation metrics, and highlights practical potential for assistive BCIs. It also discusses limitations in long-range text coherence and emphasizes future directions such as multi-modal inputs, personalized calibration, and real-time deployment to enhance real-world applicability.

Abstract

Text generating capabilities have undergone a substantial transformation with the introduction of large language models (LLMs). Electroencephalography (EEG)-based text production is still difficult, though, because it requires a lot of data and processing power. This paper introduces a new method that combines the use of the Gemma 2B LLM with a classifier-LLM architecture to incorporate a Recurrent Neural Network (RNN) encoder. Our approach drastically lowers the amount of data and compute power needed while achieving performance close to that of cutting-edge methods. Notably, compared to current methodologies, our methodology delivers an overall performance improvement of 10%. The suggested architecture demonstrates the possibility of effective transfer learning for EEG-based text production, remaining strong and functional even in the face of data limits. This work highlights the potential of integrating LLMs with EEG decoding to improve assistive technologies and improve independence and communication for those with severe motor limitations. Our method pushes the limits of present capabilities and opens new paths for research and application in brain-computer interfaces by efficiently using the strengths of pre-trained language models. This makes EEG-based text production more accessible and efficient.

Neurocognitive Modeling for Text Generation: Deep Learning Architecture for EEG Data

TL;DR

This work addresses the challenge of EEG-based text generation by introducing a data- and compute-efficient framework that couples an RNN-based EEG feature extractor with a classifier-LLM pipeline built around the Gemma 2B model. By extracting compact EEG representations and leveraging pre-trained language modeling capabilities, the approach enables EEG-conditioned text generation with substantially less training data than traditional end-to-end methods, while maintaining competitive performance and robustness to resource constraints. The study demonstrates notable results on the ImageNet of the Brain dataset, including 2-class and 3-class classification performance and perplexity-based text generation metrics, and highlights practical potential for assistive BCIs. It also discusses limitations in long-range text coherence and emphasizes future directions such as multi-modal inputs, personalized calibration, and real-time deployment to enhance real-world applicability.

Abstract

Text generating capabilities have undergone a substantial transformation with the introduction of large language models (LLMs). Electroencephalography (EEG)-based text production is still difficult, though, because it requires a lot of data and processing power. This paper introduces a new method that combines the use of the Gemma 2B LLM with a classifier-LLM architecture to incorporate a Recurrent Neural Network (RNN) encoder. Our approach drastically lowers the amount of data and compute power needed while achieving performance close to that of cutting-edge methods. Notably, compared to current methodologies, our methodology delivers an overall performance improvement of 10%. The suggested architecture demonstrates the possibility of effective transfer learning for EEG-based text production, remaining strong and functional even in the face of data limits. This work highlights the potential of integrating LLMs with EEG decoding to improve assistive technologies and improve independence and communication for those with severe motor limitations. Our method pushes the limits of present capabilities and opens new paths for research and application in brain-computer interfaces by efficiently using the strengths of pre-trained language models. This makes EEG-based text production more accessible and efficient.

Paper Structure

This paper contains 34 sections, 20 equations, 6 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Recent Work
  3. BCI Systems for Text Generation
  4. Neural-Symbolic Approaches
  5. Forward Process Classifier-LLM Network
  6. RNN Encoder Model
  7. Classifier Module
  8. Gemma 2B
  9. Methodology
  10. Hardware Setup for Training and Preprocessing
  11. Dataset
  12. Preprocessing
  13. Results and Analysis
  14. Text Generation Benchmarking and Evaluation
  15. Perplexity
  16. ...and 19 more sections

Figures (6)

  • Figure 1: BCI System Architecture for text generation using event-related potentials.
  • Figure 2: Overview of the Neural-Symbolic Processor framework integrating neural and symbolic reasoning.
  • Figure 3: EEG Feature extraction network architecture for text generation showing parallel convolutional blocks, LSTM layers, and dense classifier.
  • Figure 4: Electrode locations of International 10-20 system for EEG recording showing AF3, AF4, T7, T8, and Pz channels.
  • Figure 5: Example of brain signals evoked by visual stimuli of Goose ImageNet Class, showing raw and preprocessed EEG data across five channels.
  • ...and 1 more figures