Scholarlink

<p>To be robust, AI systems need to quickly detect and effectively react to novelties in their environments. Novelty is characterized by a sudden change in the environment where this is little time to collect new data and retrain. This is particularly true for smart home systems, where novelties abound and accurate handling is required for reliable health monitoring and home automation that can react quickly and effectively to the novelties. In this paper, we introduce a Bayesian nonparametric method, called OTACON, for novelty handling. OTACON finds surprising situations using change point detection and adapts accordingly. To evaluate this proposed approach, we design a smart home novelty generator that embeds novelties of varying type and difficulty into CASAS real-world smart home datasets. We observe that OTACON outperforms a state-of-the-art method for eight types of novel scenarios, in some cases even outperforming its own pre-novelty baseline. The results provide evidence that this method can boost AI systems in their ability to handle a variety of unexpected situations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detecting and reacting to smart home novelties

  • 期刊论文

  • 发表日期：

    2025年9月6日

查看全文
  • Lawrence B. Holder,
  • Baxter Eaves,
  • Patrick Shafto,
  • Christopher Pereyda,
  • Brian Thomas,
  • Diane J. Cook

摘要

To be robust, AI systems need to quickly detect and effectively react to novelties in their environments. Novelty is characterized by a sudden change in the environment where this is little time to collect new data and retrain. This is particularly true for smart home systems, where novelties abound and accurate handling is required for reliable health monitoring and home automation that can react quickly and effectively to the novelties. In this paper, we introduce a Bayesian nonparametric method, called OTACON, for novelty handling. OTACON finds surprising situations using change point detection and adapts accordingly. To evaluate this proposed approach, we design a smart home novelty generator that embeds novelties of varying type and difficulty into CASAS real-world smart home datasets. We observe that OTACON outperforms a state-of-the-art method for eight types of novel scenarios, in some cases even outperforming its own pre-novelty baseline. The results provide evidence that this method can boost AI systems in their ability to handle a variety of unexpected situations.