EEG Foundation Model - Ideas and Potential Projects

[FortKnoxWasTaken]

Hi @zeydabadi

I'm Anuj, a recent graduate from University of Mumbai and I'm highly interested in contributing to your project on building open-source EEG models.

I'm proficient with resting-state (RS) EEG-based signal processing and ML (supervised transfer learning) from my past research at UBC, Canada (computational neuroscience) with a solid background in Python, MATLAB and TensorFlow. Coming from a background in Electrical/Electronics Engineering (Minor in Computer Engineering) I have practical research lab experience in statistical analyses where I've worked on an RNN based model for early-stage concussion studies focusing using EEG channels (stat analysis for feature reduction). Additionally, I'm working towards quantifying attention based on EEG in driving experiments.

Like @BitC3t mentioned, EEGNet is a working stable convolutional design majorly for motor-imagery, although (from personal experience) it has a sub-par performance on long-epoch EEG signals (and resting-state data is a major concern). Hence my thoughts on using only convolutional models are a bit reserved. Similarly spectral convolutions could be misleading for resting-state data (most of the research currently going on is moving towards RS because of the ease of gathering this data).

I believe spatial-temporal multi-modal models could perform way better by capturing long temporal inter-channel features (which potentially a lot of researchers miss out on). Combining ConvLSTM paradigms with transformers and transfer learning approaches could yield a solid one-for-all EEG model.

Additionally, I believe it would be new-research and carry out learning on source-based signals. I have experience in the (painful) problem on sensor-source reconstruction and it would be awesome to incorporate causality mechanisms (effective connectivity) to determine signal origin which would answer questions regarding potential change in connectivity.

I'd love to hear your thoughts on this and it would be great if we could connect over a brief meeting to discuss potential avenues further. I believe I could contribute to the lab's project and it would be super amazing to create large open-source models for EEG experiments (and perhaps publish the results).

Best,
Anuj (canuj2312@gmail.com)

[Sreebhargavibalijaa]

Dear Mahmoud Zeydabadinezhad, Babak Mahmoudi,

I am writing to express my strong interest in the "Development of an Open-Source EEG Foundation Model" project listed under the Google Summer of Code initiative. My name is Sree Bhargavi Balija, and I am currently a Master’s student specializing in Machine learning and Data Science at the University of California, San Diego. My academic background, combined with my hands-on experience in machine learning and EEG data analysis, has equipped me with the necessary skills and knowledge to contribute effectively to this exciting project.

The aim of creating an open-source foundation model for EEG data analysis aligns perfectly with my academic interests and career aspirations. I have been deeply involved in research projects that apply machine learning techniques to neuroscience, particularly in analyzing neural signals. This has provided me with a solid understanding of EEG data and its complexities. I am proficient in Python and have practical experience with deep learning frameworks, especially PyTorch, which I believe are critical skills for this project. Recently my paper on Federated LLM's got recently accepted in AAAI, Stanford university.

I am particularly drawn to the challenge of developing algorithms for EEG signal processing and automatic feature extraction. My coursework and research have provided me with a strong foundation in signal processing and neural data analysis. I am eager to apply this knowledge to improve the robustness and versatility of EEG analysis, especially in contexts where data is scarce.

Here are a few specific questions and points I would like to discuss regarding the project:

Could you provide more details on the specific EEG datasets that the project plans to utilize for pre-training the foundation model? Understanding the characteristics of these datasets would be crucial for designing effective pre-training strategies.

I am curious about the expected challenges in implementing deep learning-based algorithms for EEG data, particularly in terms of handling the high dimensionality and variability inherent in EEG signals. How does the project plan to address these challenges?

Finally, I would appreciate more insight into the collaborative aspect of this project. How will the team dynamics be structured, and what opportunities will there be for direct mentorship and guidance?

I am excited about the opportunity to contribute to the development of an open-source EEG foundation model. I believe that this project has the potential to make significant advancements in the field of neuroscience and beyond. I am enthusiastic about the possibility of bringing my background in data science and neuroscience to the team, and I am eager to learn from and collaborate with experts in the field.

Thank you for considering my application. I am looking forward to the possibility of contributing to this groundbreaking project.

Sincerely,

Balija Sree Bhargavi (sbalija@ucsd.edu)
Masters Student, University of California, San Diego

[Precious-EEE]

Hi @Babak Mahmoudi and @zeydabadi
I am Adekunle Precious Oreoluwa (preciousadekunle33@gmail.com), I am a student of Obafemi Awolowo University. I am strongly interested in the "Development of an Open-Source EEG Foundation Model" project. Please can you tell me things I need to study or understand for the project.

[hatif03]

@zeydabadi how can i get my proposal reviewed before submitting (md.hatifosmani15@gmail.com)

[SparshRastogi]

Hello @zeydabadi,

I hope this message finds you well. My name is Sparsh Rastogi, currently pursuing a Bachelor's degree in Computer Science & Engineering at TIET, Patiala I am writing to express my keen interest in contributing to the project 'Development of an Open-Source EEG Foundation Model'

My background primarily revolves around deep learning and artificial intelligence. I have a solid understanding of Python as well as the deep learning frameworks like PyTorch, TensorFlow etc.I have been actively engaged in various research projects on the application of deep learning in the biomedical domain. Through these endeavors, I've gained extensive experience with a range of architectures including variants of transformers, CNNs, LSTMs, and CoAtNet etc. Additionally, my involvement in spectroscopic signal processing has equipped me with valuable skills relevant to the project. I also possess experience working with Large Language Models and fine-tuning them on custom datasets providing me with the skills & knowledge relevant for this project.
I am particularly drawn to this project because it resonates well with my goal, offering me the opportunity to learn and utilize my skills for the betterment of society.

I have explored the existing literature, some of my findings are listed are as follows:

1.) EEGConvTransformer : An architecture that leverages the best of both convolutions & multi-ahead attention for single trial EEG classification, extracting the features in both temporal as well as spatial domain simultaneously, the inter-regional dependencies are handled by self-attention heads followed by a convolutional layer to capture the temporal features

2.) Classification of EEG signals using Transformer based deep learning and ensemble models : An ensemble model based approach utilising slightly different transformer architectures for time domain analysis and frequency domain analysis and then taking their ensemble model to retreive the overall results

3.) EEGFormer : An architecture focusing upon SSVEP analysis using a single dimenisonal CNN to extract temporal, spatial, and convolutional features followed by an encoder decoder architecture for mapping the relation between them using self-attention mechanism

I also have a few other ideas related to fine-tuning LLM on signal data and using it for zero-shot classification which I would be discussing in detail in my proposal. It would also be great if you could provide more information, if there are any specific datasets that we need to focus upon in this project. Would love to hear your feedback! Thank you for considering my application

With regards,
Sparsh Rastogi(sparshrastogi03@gmail.com)

[achalac]

Hi @zeydabadi,

I am writing to express my profound interest in contributing under your esteemed guidance and supervision.

My name is Achala Chathuranga Aponso, and I am a PhD candidate at Edith Cowan University, Australia, deepening my exploration of the human mind at the intersection of psychology and artificial intelligence. My academic and professional journey, underscored by a Master of Science in Artificial Intelligence and a Bachelor of Science in Software Engineering, has been a testament to my commitment to advancing our understanding of cognitive processes through innovative technological solutions.

Throughout my career, I have been recognised for my contributions to the field of artificial intelligence and its application in psychology, as evidenced by:

Leading Research and Publications: My work has led to pioneering research outcomes, including a novel framework for automated computer-aided diagnosis of medical images and critical evaluations of machine learning approaches to identify early stages of dementia. These contributions are documented in esteemed journals and conference proceedings, underscoring my role as a key contributor to advancing our understanding and capabilities in EEG data analysis and machine learning.

Award-Winning Excellence: My dedication to scientific research has been honoured with the President’s Award for Scientific Research Publication, acknowledging my outstanding contribution to the field. This accolade, along with my work being cited in patents by major corporations like NISSAN MOTOR CO., LTD., highlights the impactful and innovative nature of my research endeavours.

Scholarly Recognition: I have secured the ECU Higher Degree by Research Scholarship, a testament to my academic excellence and the potential for significant research contributions. This scholarship, covering full tuition and living expenses, enables me to pursue my PhD with an undivided focus on developing machine learning models for psychological research.

Competitive Achievements: My skills have also been recognised in competitive environments, evidenced by my achievements in GovHack 2023 Challenges, where I was a Runner-Up in both New South Wales and Western Australia Challenges, showcasing my ability to apply my expertise in real-world scenarios effectively.

My technical proficiency encompasses a broad spectrum of programming languages and frameworks crucial for EEG data analysis, including but not limited to Python, MATLAB, and TensorFlow. This proficiency is bolstered by my in-depth knowledge and hands-on experience with neural networks, signal processing, and machine learning algorithms, making me uniquely equipped to tackle the challenges and opportunities presented by the development of an open-source EEG foundation model.

With a solid foundation in both the theoretical and practical aspects of artificial intelligence (Machine Learning . Deep Learning ) and psychology, combined with a proven track record of achievements in research and innovation, I am enthusiastic about contributing to the "Development of an Open-Source EEG Foundation Model" project under GSoC 2024 with Emory BMI. I am keen to bring my expertise, insights, and proven problem-solving abilities to this project, aiming to contribute significantly to its success and the broader field of biomedical informatics.
My expertise in EEG analysis is deeply rooted in an advanced understanding of machine learning techniques and their application to neurological data. Through my academic research and professional projects, I have honed my ability to process, analyze, and interpret EEG signals using a variety of machine learning models, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and deep learning algorithms tailored for temporal and spatial data analysis. My proficiency in Python, MATLAB, and TensorFlow and my rigorous background in artificial intelligence from my MSc and ongoing PhD research have enabled me to develop innovative EEG-based models that enhance diagnostic accuracy, cognitive state assessment, and neurological disorder identification. My work includes designing and implementing algorithms for automatic feature extraction, signal enhancement, and developing predictive models that navigate the complexities of EEG data, demonstrating a significant impact in psychology and biomedical informatics. This blend of technical skills and research experience underscores my unique capability to contribute to EEG machine learning projects, pushing the boundaries of what's possible in the analysis and understanding of brain activity.

I am eager to contribute to groundbreaking advancements in open-source biomedical informatics alongside the Emory BMI team.

Warm regards,

Google Scholar: https://scholar.google.com.au/citations?user=7QwYScEAAAAJ&hl=en&oi=ao
LinkedIn: https://www.linkedin.com/in/achalac/