Using Active Learning to Allow Activity Recognition on a Large Scale

Alemdar, Hande; Kasteren, T.L.M. van; Ersoy, Cem

doi:10.1007/978-3-642-25167-2_12

Cited by 27 publications

(29 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…But for longer activities such as sleeping or leave home we can afford a more coarse-grained discretisation. A choice of 60 s per time slot has shown to be fine enough with an acceptable error rate as shown by [28,29]. For this particular dataset, timeslices of 60 s incur in a performance error below 2 % [26], which proves to be a good trade-off between loss of information and efficiency.…”

Section: Data Pre-processingmentioning

confidence: 92%

Modelling and Detection of User Activity Patterns for Energy Saving in Buildings

Ortega

Han

Bowring

2016

Emerging Trends and Advanced Technologies for Computational Intelligence

View full text Add to dashboard Cite

Recently, it has been noted that user behaviour can have a large impact on the final energy consumption in buildings. Through the combination of mathematical modelling and data from wireless ambient sensors, we can model human behaviour patterns and use the information to regulate building management systems (BMS) in order to achieve the best trade-off between user comfort and energy efficiency. Furthermore, streaming sensor data can be used to perform real-time classification. In this work, we have modelled user activity patterns using both offline and online learning approaches based on non-linear multi-class Support Vector Machines. We have conducted a comparison study with other machine learning approaches (i.e. Linear SVM, Hidden-Markov and K-nearest models). Experimental results show that our proposed approach outperforms the other methods for the scenarios evaluated in terms of accuracy and processing speed.

show abstract

Section: Data Pre-processingmentioning

confidence: 92%

Modelling and Detection of User Activity Patterns for Energy Saving in Buildings

Ortega

Han

Bowring

2016

Emerging Trends and Advanced Technologies for Computational Intelligence

View full text Add to dashboard Cite

show abstract

“…Active learning has been studied in many previous works as a means for improving the AR performance with a minimal supervision [12,7,1]. It leverages the unlabeled data to identify uncertain/informative instances for labeling.…”

Section: Related Workmentioning

confidence: 99%

“…This problem is further referred to as supervised activity discovery. In the AR domain, the problem of minimizing supervision was studied in the context of active learning [12,7,1]. To improve the recognition results with minimal supervision, an active learning system asks a user to label only the most uncertain instances.…”

Section: Introductionmentioning

confidence: 99%

SuperAD

Nguyen

Tague

Zhang

et al. 2015

Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2015 AC

View full text Add to dashboard Cite

Activity recognition (AR) has become an essential component of many applications present in our everyday lives such as life-logging, fitness tracking, health and wellbeing monitoring. To build an AR system, one needs to first identify a set of activities of interest and collect labeled training data for these activities. However, activities of interest are not often known in advance. For example, a system designed to monitor a user's life style for potential diabetes risk needs to recognize all physical activities a user performs in her daily life. Given the large number of possible human activities, many of them cannot be foreseen during the model training time. In this work, we study the problem of discovering these unknown activities after the system is deployed by asking users to provide additional labels. Our goal is to discover all the unknown activities (i.e., obtain at least one label per class) while minimizing the amount of labels a user needs to provide. We propose SuperAD (Supervised Activity Discovery) approach, which combines active learning, semi-supervised learning and generative modeling to discover new unknown activities. We show that the proposed approach is especially effective when discovering activities with imbalance class distribution.

show abstract

“…An information gain function is used to select the most informative points in the pool, and requests are issued to the user in order to obtain the corresponding label. This produces an updated pool, which is again evaluated to obtain a new label, iteratively [1]. Unfortunately, the user's memory cannot be trusted for reliable annotation if the target activity took place in the distant past or too many other activities took place after the target activity.…”

Section: Technical Perspectivementioning

confidence: 99%

On strategies for budget-based online annotation in human activity recognition

Miu

Plötz

Missier

et al. 2014

Proceedings of the 2014 ACM International Joint Conference on Pervasive and Ubiquitous Computing: Adjunct Publication

View full text Add to dashboard Cite

Bootstrapping activity recognition systems in ubiquitous and mobile computing scenarios often comes with the challenge of obtaining reliable ground truth annotations. A promising approach to overcome these difficulties involves obtaining online activity annotations directly from users. However, such direct engagement has its limitations as users typically show only limited tolerance for unwanted interruptions such as prompts for annotations. In this paper we explore the effectiveness of approaches to online, user-based annotation of activity data. Our central assumption is the existence of a fixed, limited budget of annotations a user is willing to provide. We evaluate different strategies on how to spend such a budget most effectively. Using the Opportunity benchmark we simulate online annotation scenarios for a variety of budget configurations and we show that effective online annotation can still be achieved using reduced annotation effort.

show abstract

Using Active Learning to Allow Activity Recognition on a Large Scale

Cited by 27 publications

References 18 publications

Modelling and Detection of User Activity Patterns for Energy Saving in Buildings

Modelling and Detection of User Activity Patterns for Energy Saving in Buildings

SuperAD

On strategies for budget-based online annotation in human activity recognition

Contact Info

Product

Resources

About