Irina Rish, Researcher, IBM Watson, NYC 2017
summary
Applies machine learning to neuroimaging with concrete results: fMRI functional-network features classify schizophrenia vs controls up to 93% accuracy; ensemble classifiers show methylphenidate (Ritalin) makes cocaine addicts’ resting-state networks harder to distinguish from controls, implying partial normalization; EEG is turned into brain movies (electrodes → 2D RGB frames) and recurrent convnets plus LSTM and temporal convolutions boost memory-load classification by about nine percent and code is on GitHub; proposes neurogenesis-inspired dictionaries with birth/death group sparsity that improve adaptation on nonstationary data.
AI-written summary · about
chapters (15)
- 0:00Intro
- 2:11AI 2 Brain: Mental State Recognition to Improve Mental Function
- 4:46ML in Neuroimaging: Some Examples
- 9:32Example 1: Cocaine Addiction fMRI Study
- 10:27MPH: A Stimulant for a stimulant?
- 12:15Resting-state Functional MRI
- 14:18Classification Results: MPH Normalizes CUD's Networks
- 16:56Example 2: Working Memory Load Classification from EEG
- 19:37Brain Movie' Classification with Recurrent ConvNets
- 21:22Baseline: Non-temporal Approach with ConvNets
- 21:56Adding Time is Better: Recurrent ConvNets
- 23:16Interpretability via Deconvolution
- 25:17Brain 2 Al: Adult Neurogenesis
- 30:33Better Adaptation in Non-Stationary Environment
- 32:18Some Lessons
related talks
32mIrina Rish, Research Staff, IBM T.J. Watson Research Center
21mIvy Zhu, Research Scientist, Intel @ MLconf SEA
09mNeuroAssist: Enhancing Cognition Computer Synergy with Adaptive AI & Enhance Neural Decoding
48mNeural Networks and Brains
26mBrainDrain Using Machine Learning and Brain Waves to Detect Errors in Human
42mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
26mTed Willke, Senior Principal Engineer, Intel Labs
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics