Using prediction to conserve energy in recognition on mobile devices

Abstract

Abstract-As devices are expected to be aware of their environment, the challenge becomes how to accommodate these abilities with the power constraints which plague modern mobile devices. We present a framework for an embedded approach to context recognition which reduces power consumption. This is accomplished by identifying class-sensor dependencies, and using prediction methods to identify likely future classes, thereby identifying sensors which can be temporarily turned off. Different methods for prediction, as well as integration with several classifiers is analyzed and the methods are evaluated in terms of computational load and loss in quality of context. The results indicate that the amount of energy which can be saved is dependent on two variables (the acceptable loss in quality of recognition, and the number of most likely classes which should be accounted for), and two scenario-dependent properties (predictability of the context sequences and size of the context-sensor dependency sets).

Global Peer-to-Peer Classification in Mobile Ad-Hoc Networks: A Requirements Analysis

“…On the other hand, the network is extremely stable as failed nodes do not adversely affect the classification of the rest of the network, as long as the classifier used can accommodate the variable feature vector length (see [4]). Also, new nodes which are added to the network must only receive the parameters for the classifier and be added to the global list of data publishers and subscribers in order to become functioning members of the new system.…”

3 rd workshop on recent advances in behavior prediction and pro-active pervasive computing

“…For instance, Gordon et al reduce energy consumption in sensor nodes by powering only those components that are likely needed in the near future [26] and Schrempf et al Investigate the prediction of user intention in order to pro-actively plan tasks of a robot the human is interacting with [27].…”

Adaptive sampling in context-aware systems: a machine learning approach

“…Alternatively, unsupervised learning recognises emerging patterns in data without manually marked training data. In this category, approaches such as neural networks [17], nearest-neighbour algorithms and hidden Markov models have been explored [9]. The advantages of unsupervised learning are clear, in that the process is completely autonomic, but the trade-off is that the learnt contexts identified may not clearly map to userinterpretable situations.…”

“…Recognition of contexts is, however, associated with energy costs entailed by the use of a number of heterogeneous sensors, and by costs associated with the transmission and processing of the data [9,17]. There are various approaches to saving energy in these situations, such as the use of ultra-low power sensors.…”

“…In a context-aware sensor system, selective activation of sensors according to the contexts to be sensed significantly reduces the energy costs of transmitting data [11]. This approach can be extended to make use of prediction in adjusting sensor configurations to be best suited to sensing the most likely next contexts [9]. Research has largely focussed on the communication energy costs of the data, rather than the sampling costs, although these costs are considered in [2].…”