Online supervised incremental learning of link quality estimates in wireless networks

An adaptable fuzzy-based model for predicting link quality in robot networks.

“…To mitigate this issue, other authors have suggested that nodes should exchange labeled samples across multiple links [24,30], but the problem is that the statistical relationship between the inputs and target are link-specific and should not be aggregated. In summary, there are several opportunities to advance the state of the art in LQ estimation.…”

“…The use of regression techniques in the form of supervised learning have also been explored in a series of works by the same research center [28][29][30]91]. In [28,29], a distributed protocol was designed to exploit the mobility of nodes for gathering diverse training samples, and afterwards, use the training samples in an offline supervised learning algorithm.…”

“…To address the adaptability issues in [28,29], Di Caro et al developed an online learning framework that incrementally retrains its regression model [30]. According to Di…”

“…(iv) Reduce sampling and labeling expenses -Some schemes ignore the costs associated with collecting labeled training examples as evident in the large training sets used in [28,29] and the continuous online labeling and tuning used in [24,30]. However, there are energy and network costs associated with obtaining labels due to them residing at the opposite end of the link.…”

“…(v) Improved accuracy -Some authors advocate for the mixing of training sets gathered across disparate links to expedite model startup [30,93].…”