Skip to main content

Enhancing DNN Explainability in BCI

This project tackled the challenge posed by the opaque nature of Deep Neural Network (DNN) models, especially crucial in safety-driven fields like Brain-Computer Interface (BCI) applications. The primary aim was to unleash the full potential of DNN models, specifically in analyzing EEG signals.

In this initial study, I highlighted the importance of not just relying on quantitative performance but also considering qualitative evaluation. The investigation delved into a key paper in the field, providing a comprehensive analysis as a starting point. I then compared the model against common alternatives, tweaking various hyper-parameters to validate its performance.

To demystify the "black-box," I utilized the Layer-wise Relevance Propagation (LRP) method, revealing insightful details about each model's reliability. LRP proved invaluable in explaining the inner workings of the DNN, making it a useful tool in interpreting complex neural network decisions and enhancing transparency.


Stay curious with me.

Comments

Post a Comment

Popular posts from this blog

Ensuring Responsible Integration of LLMs in Data Science Workflows

It is crucial to exercise caution when incorporating LLMs into data science workflows, particularly with previously unseen data or in unfamiliar domains. Data science at its core involves a comprehensive understanding of data within its unique context. LLMs, while capable of generating functional code and providing insightful suggestions, do not inherently comprehend the underlying implications of the data they process. This disconnect can introduce biases in methodology and potentially lead to a misinterpretation of the problem. To mitigate these risks, it is advisable to limit the use of LLMs to labor-intensive, basic data wrangling tasks. Even then, sensitive data should not be directly fed into the LLM. Instead, use dummy examples or limited portions of the data to guide the LLM without compromising data integrity. This approach depends heavily on understanding the intricate relationships between features and adhering to data protection requirements. However, be aware that strippin...
Post a Comment
Read more

Public Speaking and Communication Training Application

EmoSpeak Coach, one of my latest creations, is a groundbreaking public speaking and communication training application born out of a personal necessity. As someone deeply invested in effective communication, I created this tool to provide real-time insights into emotional cues and speech patterns during public speaking, aiming to empower individuals on their journey to becoming confident communicators. The inspiration behind EmoSpeak Coach lies in my own experiences, realizing the need for a comprehensive tool that combines face and hands detection, speech-to-text capabilities, and text context evaluation. The goal was to offer users a holistic understanding of their communication style, beyond just emotional expression and speech delivery. The project kicked off with an in-depth analysis of relevant studies, setting the foundation for EmoSpeak Coach. Through meticulous comparisons with alternative models and experimentation with various features, the application evolved to provide use...
Post a Comment
Read more

Dynamic Learning Tool for Medical Education

Embarking on a transformative journey in medical education, I present a cutting-edge self-learning and self-evaluating platform. Originally built for the aspiring minds of medical students at Sapienza University, this innovation takes the form of a tree-structured questionnaire, introducing a dynamic approach to learning. How It Works Designed to adapt to individual progress, our platform customizes the question flow based on previously selected answers. This dynamic feature not only provides instant feedback but serves as a motivational guide for learners. Users engage with a series of thought-provoking questions, scenarios, and answer choices, witnessing results and explanations unfold in real-time. Active Recall Technique At the heart of this revolutionary platform is an underlying algorithm that employs the powerful Active Recall technique. This technique enhances memory retention by prompting users to actively recall information, contributing to a more robust and enduring understa...
Post a Comment
Read more