A growing proportion of the aged in population provokes shortage of caregivers and restructuring of living spaces. One of the most promising solutions is to provide with a smart home environment that ensures independence of users. In this paper, we first call attention to the fact that a learning capability of human behavior patterns can play a central role in adequate functioning of such systems. Specifically, we give an overview of important related studies to illustrate how a variety of learning functions can be successfully incorporated into the smart home environment. We then present our approaches towards the issues of life-long learning and non-supervised learning, which are considered essential aspects of a smart home system. The two learning schemes are shown to be satisfactory in facilitating independent living over different time scales and with less human intervention. Finally, we mention about a prospective model of a future smart home.