A meta-learning approach to the regularized learning—Case study: Blood glucose prediction

Abstract

In this paper we present a new scheme of a kernel-based regularization learning algorithm, in which the kernel and the regularization parameter are adaptively chosen on the base of previous experience with similar learning tasks. The construction of such a scheme is motivated by the problem of prediction of the blood glucose levels of diabetic patients. We describe how the proposed scheme can be used for this problem and report the results of the tests with real clinical data as well as comparing them with existing literature.

Filtered Legendre expansion method for numerical differentiation at the boundary point with application to blood glucose predictions

“…Since ∆(f )(1) = 2f ′ (1) for any f (7), (59) implies (26). In order to prove (27), we use (42) and (57) to deduce that …”

“…The estimate (28) follows easily by applying (27) with y = f − f δ and using the resulting estimate together with (26) and triangle inequality.…”

“…There are several prediction techniques, and a variety of the glucose predictors has been recently proposed, see, for example, [26] and references therein. In this section we discuss the predictors based on the numerical differentiation [13].…”

Subspace-based linear multi-step predictors in type 1 diabetes mellitus

“…Their models utilized only recent glucose history from a CGM device, achieving 3-5% error for 30-min-ahead prediction. In [10] a kernel-based regularization learning algorithm, in which the kernel and the regularization parameter are adaptively chosen on the basis of previous similar learning tasks, using past glucose concentration information, was presented. A short-coming of the methods listed above is the lack of exploitation of the dynamic interplay between previously injected insulin, meal intake and eventually exercise to the purpose of improving glucose prediction.…”

Short-term prediction of blood glucose concentration using interval probabilistic models

“…predictions are developed in [2], neural networks for predictions up to 180 minutes are considered in [3], subspace-based techniques are employed in [4], autoregressive models which could be transferred from patient to patient, without need for individual tuning are developed in [5], adaptive autoregressive models which also take the inputs meal carbohydrates and insulin into account are presented in [6] and a kernel-based meta-learning approach is proposed in [7].…”