Lightning Talk: Bridging the Gap: Engineering Compliant...
summary
A PyTorch-based Clinical Auditing System backpropagates gradients from 4D fMRI outputs into EEG bands to expose what drives predictions, aiming to satisfy the EU AI Act for high‑risk medical AI. Using IntegratedGradients (Captum) on a NeuroBOLT EEG→fMRI model, the audit finds a surprising 60% reported accuracy but attention maps that focus on frontal electrodes instead of expected occipital ones, suggesting the model is “cheating.” A simple Streamlit dashboard and GitHub repo demonstrate the audit pipeline and clinician-facing visualizations.
AI-written summary · about
related talks
35mTranslating Brain Electrical Signals into Actions through Models...
09mLightning Talk: A Novel Domain Generalization Technique for Medical Imaging Using...
11mConnected brains
26mBrainDrain Using Machine Learning and Brain Waves to Detect Errors in Human
27mBuild Your Own ML Data Feedback Loop
12mLightning Talk: Leveraging PyTorch 2.0 for Bias Reduction in AI
31mLearn to control your brain: Brain Computer Interfacing with Python
08mModel Interpretability with Captum