Machine learning, dense sensing and decision theoretic planning for large-scale assistance systems

点击翻译按钮获取中文摘要

Assistance for elderly persons is becoming an important and pressing need with serious effects on society because of the changing demographic towards an aging population (the so-called "Silver Tsunami''). The burden of care is shifting from the professional arena (e.g., hospitals) into the home and community. Technology to support people in their need to live independently is currently available in the form of personal alarms and environmental aids. Looking to the future, we can imagine intelligent, pervasive computing technologies using sensors and effectors that help with more difficult cognitive problems in planning, sequencing and attention. For example, a elderly woman with dementia may require help in recalling important steps in her everyday life. A set of sensors, such as cameras and switches, embedded in her home could extract information about her behaviours using computational models of temporal sequences. This information could guide the provision of pre-recorded audio prompts, to which she is known to be responsive. The system could adapt to her as she changes over time, or as her environment changes (e.g., if she buys new furniture). There are three artificial intelligence (AI) problems that remain to be addressed in order to build such effective assistance systems. The first problem is that of sensing of the environment, specifically in relation to planning in a human assistance task. The second problem is that of planning and decison making in large-scale domains with high dimensional sensor data. The third problem is that of learning effective assistance mechanisms by eliciting prior knowledge and by integrating knowledge and empirical evidence into sensing and decision making. This research program will provide solutions to these three foundational problems with the eventual aim of offering an exciting and viable way to enable persons to live independently for longer, thereby maintaining quality of life.