CAREForMe: Contextual Multi-Armed Bandit Recommendation Framework for Mental Health

Presentation slides of our paper "CAREForMe: Contextual Multi-Armed Bandit Recommendation Framework for Mental Health" at MOBILESoft 2024.
Presentation Video: https://youtu.be/oJZRwtq8As4

Transcript

1. CAREForMe: Contextual Multi- Armed Bandit Recommendation Framework for Mental Health New Idea Track
2. 1 in 4 people are affected by mental illness worldwide 56% adults with mental illness receive no treatment $280 billion spent on mental health in the U.S. in 2020 2 Mental Health Crisis
3. AI Chatbot Meditation Apps Health Tracking Apps …… 3 mHealth Solutions Low Cost 24/7
4. AI Chatbot Meditation Apps Health Tracking Apps …… 4 mHealth Solutions Low Cost 24/7
5. Easier said than done! 5 Ideal Reality I know what to do. I’ll be happy ever after! Wait what now? Breathe? Meditate? Gratitude? NOOOOOO I’m a LOSER!
6. CAREForMe Cares Just-in-Time No burden on the user to initiate help Understand user context & situation Just-in-time support Personalization No same general advice No slow start to understand user Continuous learning from similar users Modularity No ad-hoc pipeline Separation of concerns Customizable & flexible framework 6
7. CAREForMe CMAB Recommendation Framework 7
8. CAREForMe CMAB Recommendation Framework 8
9. CAREForMe CMAB Recommendation Framework 9
10. CAREForMe CMAB Recommendation Framework 10 3 ChatBots https://github.com/ANRGUSC/mab-reco
11. What is Contextual Multi- Armed Bandit (CMAB)? 11
12. What is Contextual Multi- Armed Bandit (CMAB)? 12 No need to know which machine works better!
13. Why Contextual Multi- Armed Bandit (CMAB)? AI Technique with continuous learning (NOT LLM😂) Small, tailored models à low cost, accessibility 13
14. Why Contextual Multi- Armed Bandit (CMAB) AI Technique with continuous learning (NOT LLM😂) Small, tailored models à low cost, accessibility 14
15. Why Contextual Multi- Armed Bandit (CMAB) AI Technique with continuous learning (NOT LLM😂) Small, tailored models à low cost, accessibility Adaptive when learning from new data (no retrain) Handle “cold start” problem Context-aware & personalized 15
16. Why Contextual Multi- Armed Bandit (CMAB) AI Technique with continuous learning (NOT LLM😂) Small, tailored models à low cost, accessibility Adaptive when learning from new data (no retrain) Handle “cold start” problem Context-aware & personalized 16
17. Easier said than done! 17 Ideal Reality I know what to do. I’ll be happy ever after! Wait what now? Breathe? Meditate? Gratitude? NOOOOOO I’m a LOSER!
18. Easier to do & I CAN DO IT! 18 Home Work Great job meditating! This will help you maintain your cool 😎 Hi there, finding it hard to relax? It’s normal! Why not put on some relaxing music?
19. Easier to do & I CAN DO IT! 19 Home Work Great job meditating! This will help you maintain your cool 😎 Hi there, finding it hard to relax? It’s normal! Why not put on some calming music? Or take a walk?
20. CAREForMe’s Vision Reference pipeline for new comers Foster interdisciplinary collaboration Living repository for JITAI recommendation systems beyond mental health 20
21. CAREForMe: Contextual Multi- Armed Bandit Recommendation Framework for Mental Health New Idea Track
22. Thanks! CAREForMe: https://tinyurl.com/careforme2024 Email: yzhao@isi.edu LinkedIn: www.linkedin.com/in/yixue-zhao/ Twitter/X: @yixue_zhao 22