Kyle D. Feuz

Transfer learning for activity recognition: a survey

Many intelligent systems that focus on the needs of a human require information about the activities being performed by the human. At the core of this capability is activity recognition, which is a challenging and well-researched problem. Activity recognition algorithms require substantial amounts of labeled training data yet need to perform well under very diverse circumstances. As a result, researchers have been designing methods to identify and utilize subtle connections between activity recognition datasets, or to perform transfer-based activity recognition. In this paper, we survey the literature to highlight recent advances in transfer learning for activity recognition. We characterize existing approaches to transfer-based activity recognition by sensor modality, by differences between source and target environments, by data availability, and by type of information that is transferred. Finally, we present some grand challenges for the community to consider as this field is further developed.

Transfer Learning across Feature-Rich Heterogeneous Feature Spaces via Feature-Space Remapping (FSR)

Transfer learning aims to improve performance on a target task by utilizing previous knowledge learned from source tasks. In this paper we introduce a novel heterogeneous transfer learning technique, Feature-Space Remapping (FSR), which transfers knowledge between domains with different feature spaces. This is accomplished without requiring typical feature-feature, feature instance, or instance-instance co-occurrence data. Instead we relate features in different feature-spaces through the construction of metafeatures. We show how these techniques can utilize multiple source datasets to construct an ensemble learner which further improves performance. We apply FSR to an activity recognition problem and a document classification problem. The ensemble technique is able to outperform all other baselines and even performs better than a classifier trained using a large amount of labeled data in the target domain. These problems are especially difficult because, in addition to having different feature-spaces, the marginal probability distributions and the class labels are also different. This work extends the state of the art in transfer learning by considering large transfer across dramatically different spaces.

Automated Detection of Activity Transitions for Prompting

Individuals with cognitive impairment can benefit from intervention strategies like recording important information in a memory notebook. However, training individuals to use the notebook on a regular basis requires a constant delivery of reminders. In this study, we design and evaluate machine-learning-based methods for providing automated reminders using a digital memory notebook interface. Specifically, we identify transition periods between activities as times to issue prompts. We consider the problem of detecting activity transitions using supervised and unsupervised machine-learning techniques and find that both techniques show promising results for detecting transition periods. We test the techniques in a scripted setting with 15 individuals. Motion sensors data are recorded and annotated as participants perform a fixed set of activities. We also test the techniques in an unscripted setting with eight individuals. Motion sensor data are recorded as participants go about their normal daily routine. In both the scripted and unscripted settings, a true positive rate of greater than 80% can be achieved while maintaining a false positive rate of less than 15%. On average, this leads to transitions being detected within 1 min of a true transition for the scripted data and within 2 min of a true transition on the unscripted data.

Heterogeneous transfer learning for activity recognition using heuristic search techniques

Purpose – The purpose of this paper is to study heterogeneous transfer learning for activity recognition using heuristic search techniques. Many pervasive computing applications require information about the activities currently being performed, but activity recognition algorithms typically require substantial amounts of labeled training data for each setting. One solution to this problem is to leverage transfer learning techniques to reuse available labeled data in new situations. Design/methodology/approach – This paper introduces three novel heterogeneous transfer learning techniques that reverse the typical transfer model and map the target feature space to the source feature space and apply them to activity recognition in a smart apartment. This paper evaluates the techniques on data from 18 different smart apartments located in an assisted-care facility and compares the results against several baselines. Findings – The three transfer learning techniques are all able to outperform the baseline compar...

Prompting technologies: A comparison of time-based and context-aware transition-based prompting

BACKGROUND While advancements in technology have encouraged the development of novel prompting systems to support cognitive interventions, little research has evaluated the best time to deliver prompts, which may impact the effectiveness of these interventions. OBJECTIVE This study examined whether transition-based context prompting (prompting an individual during task transitions) is more effective than traditional fixed time-based prompting. METHODS Participants were 42 healthy adults who completed 12 different everyday activities, each lasting 1-7 minutes, in an experimental smart home testbed and received prompts to record the completed activities from an electronic memory notebook. Half of the participants were delivered prompts during activity transitions, while the other half received prompts every 5 minutes. Participants also completed Likert-scale ratings regarding their perceptions of the prompting system. RESULTS Results revealed that participants in the transition-based context prompting condition responded to the first prompt more frequently and rated the system as more convenient, natural, and appropriate compared to participants in the time-based condition. CONCLUSIONS Our findings suggest that prompting during activity transitions produces higher adherence to the first prompt and more positive perceptions of the prompting system. This is an important finding given the benefits of prompting technology and the possibility of improving cognitive interventions by using context-aware transition prompting.

Collegial Activity Learning Between Heterogeneous Sensors

Activity recognition algorithms have matured and become more ubiquitous in recent years. However, these algorithms are typically customized for a particular sensor platform. In this paper, we introduce PECO, a Personalized activity ECOsystem, that transfers learned activity information seamlessly between sensor platforms in real time so that any available sensor can continue to track activities without requiring its own extensive labeled training data. We introduce a multi-view transfer learning algorithm that facilitates this information handoff between sensor platforms and provide theoretical performance bounds for the algorithm. In addition, we empirically evaluate PECO using datasets that utilize heterogeneous sensor platforms to perform activity recognition. These results indicate that not only can activity recognition algorithms transfer important information to new sensor platforms, but any number of platforms can work together as colleagues to boost performance.