Machine Learning and Sensing for Remote Biomedicine
- 1. Machine Learning and Sensing for Remote Biomedicine Tomás Lozano-Pérez MIT CSAIL
- 2. Remotely assisted living Growth of elderly population a major challenge for 21 st century Exploit cost-effective communication, computation and robotics to provide support for elderly On-site robotic assistance Remote medical monitoring
- 3. Pervasive and Transparent Existing computer systems are inadequate for our goal Computers must live in user’s world not force users to live in computer’s world
- 4. In general… Reduce need for careful prior design Adapt to wide range of environments Understand the user
- 5. Adaptive Computing Adapting to users People and Machines Adapting to computational environment Computers and Networks Adapting to physical environment Sensors and Effectors
- 6. Adapting to Users Physical environment Where are they? Goals Find glasses Activities Cooking dinner Knowledge state Do they know Jane called? Mental state Are they depressed? Cultural and social milieu Currency, food, family, etc.
- 7. Adapting to Physical Environment Sensors Audio, video, haptics, X-ray/MRI, etc Effectors Displays, speakers, robots, etc
- 8. Adapting to Computational Environment Bandwidth and latency Sensor/effector variation Computing resources available New software/hardware
- 9. A Scenario: SARS management Hospital Contact Tracing Hospital Admission Treatment Recovery Disease Worsening Patient Discharged Calling Ambulance Community Infection Control Collecting Disease Information Supporting Clinical and Policy Research Public Education Courtesy: Leong Tze Yun
- 10. Other applications Remote consultation Tele-surgery Distance education Meeting support
- 11. Fundamental Technological Needs Signal to Symbol Object/Activity recognition Recognizing patterns Recognize deviations from normal behavior Goal oriented systems Achieve goals not follow rote instructions
- 12. Getting there… Perception Vision, speech, medical imaging Machine Learning Fitting complex models, classification Optimal data acquisition and decision making Modular, adaptive, distributed systems Adaptive software design methodology Reconfigurable hardware Algorithms Real time algorithms Correctness