Exploring semi-supervised and active learning for activity recognition

Sensor-based adaptive activity recognition with dynamically available sensors

“…However, it does not necessarily mean the contradiction between our experiments and previous work [23]. In our cases, the recognition performance is limited by the discriminative power of the features rather than the amount of training data, as we build the initial model with sufficient training data, especially for the later two datasets which include activity data from multiple users.…”

“…While in the semi-supervised area, unlabelled examples classified with high confidence are added to the training dataset to retrain and refine the model. Examples are self-training, co-training [23] and label propagation [22]. The problem of aforementioned methods is that only high-confidence examples are considered, due to the fact that they can minimize the entropy [6].…”

“…The problem of aforementioned methods is that only high-confidence examples are considered, due to the fact that they can minimize the entropy [6]. However, high-confidence examples are less informative and make less contribution to the convergence of the model [23], especially for discriminative classifiers which perform classification based on the boundaries(i.e. hyperplanes in SVM), high-confident examples are normally far wary from the boundaries and are unhelpful for the boundaries adjustment.…”

Sensor Fusion for Semantic Place Labeling

“…Other studies on semantic place labeling so far [Reddy et al, 2010, Consolvo et al, 2008, Arase et al, 2010, Bouten et al, 1997, Perrin et al, 2000, Junker et al, 2004, Preece et al, 2009, Berchtold et al, 2010, Ravi et al, 2005, Bao and Intille, 2004, Chang et al, 2007, Farringdon et al, 1999, Kern et al, 2003, Mantyjarvi et al, 2001, Stikic et al, 2008, Zinnen et al, 2009, Lester et al, 2005, Siewiorek et al, 2003 are mostly based on unlabeled data or on a small number of sensor and state data. The field of physical activity recognition based on accelerometer sensor data is well researched [Consolvo et al, 2008, Arase et al, 2010, Berchtold et al, 2010, Bao and Intille, 2004, Farringdon et al, 1999, Kern et al, 2003].…”

A Human Activity Recognition Framework for Healthcare Applications: Ontology, Labelling Strategies, and Best Practice

“…In this approach, classifiers are initially trained on a small set of data. When the classifier is applied to unlabelled data, the user is asked to label only data that, for example, is classified with low confidence or those with disagreement between classes (Stikic et al, 2008). Hoque and Stankovic (Hoque and Stankovic, 2012) employed a clustering technique to group smart home environmental data into activities and the user labelled the clusters.…”