Augmenting Mental Health Care in the Digital Age: Machine Learning as a Therapist Assistant
summary
Describes a deployed crisis-risk alert built on anonymized Talkspace messages that flags potential self-harm and related crises and surfaces alerts routed to therapists rather than messaging clients directly. Models start simple (logistic regression, decision trees) before moving to PyTorch and attention methods, use interpretable outputs via ELI5, and train on 20–30k proxy examples plus transfer from Reddit and language-model pretraining on millions of messages; final validation uses 800 clinician-labeled cases from six annotators and runs through a HIPAA-compliant AWS pipeline.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:14Barriers to Therapy
- 2:06Our Solution
- 5:35HIPAA Compliance
- 7:00Data Security
- 9:45Qualitative Data
- 10:40HumanCentered Design
- 12:14Interpretability
- 13:16Open Source Tools
- 14:16Predefined Word Categories
- 15:42Augmentation Datasets
- 16:28Workflow
- 17:19MultiTask Learning
- 17:54Language Modeling
- 18:18Communication
- 19:13Conclusion
related talks
32mAnalyzing social media for suicide risk using natural language processing
40mBringing mental health data to doctors
24mEngaging new and diverse audiences with ML
03mDBTB INT Sai Moturu r
20mThe Decade Ahead: Two Scientists’ Visions for the Future of Technology
23mLeveraging Generative AI and Transformers in Healthcare
22mLeveraging AI/ML for Transformative Software Solutions in Healthcare
01h 03mUsing Machine Learning and AI to serve people and communities