Irina Rish, Research Staff, IBM T.J. Watson Research Center
summary
Uses elastic net and group-sparse graphical models to get interpretable brain biomarkers, with elastic net reaching about 0.8 correlation to self-reported pain and group sparsity producing much cleaner cocaine-vs-control networks than vanilla graphical lasso. Functional-network features classify schizophrenia up to 93% accuracy. Cheap EEG headsets (NeuroSky, Emotiv) yield roughly 74–75% within-person discrimination of emotional versus educational videos after heavy preprocessing. Speech/text features like sentence-to-sentence coherence (cosine of averaged word embeddings) predict future schizophrenia onset in a small longitudinal cohort (≈5 of 30).
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:46Measuring Brain Activity with Functional MRI
- 3:04Example: Playing Videogames
- 4:24Another Example: Pain Perception
- 5:57What are we looking for in fMRI data?
- 6:59Interpretable predictive models
- 8:31Mental State Prediction via Linear Regression
- 9:46LASSO and Beyond
- 10:49Adding Proper Structure (Prior)
- 12:28Learning Full-Brain Sparse Markov Networks
- 17:00Summary: "Mind Reading" from Functional MRI
- 17:41Motivation: Detect Mental States, Improve Mental Function
- 18:12Wearable Devices
- 19:37Measuring Driver's Stress Level
- 20:24Meditation at Work (for Science's Sake)
- 23:22Text Analytics for Computational Psychiatry
- 26:18Ongoing Work: Speech Coherence
related talks
34mIrina Rish, Researcher, IBM Watson, NYC 2017
28mCan we measure depression
14mEEG Topologies with Neo4j: Interpolation-Driven Graph Learning for Chronic Pain Detection
11mConnected brains
13mEnhanced EEG Data Classification with 3D SNN and On-Line Learning ESN
01h 05mUsing EEG and Azure Machine Learning to Perform Lie Detection
01h 01mKeynote: Using EEG and Machine Learning to Perform Lie Detection
31mJennifer Marsman, Principal Developer Evangelist, Microsoft