Untitled

Abstract

Abstract. This paper describes a method for selecting training examples for a partial memory learning system. The method selects extreme examples that lie at the boundaries of concept descriptions and uses these examples with new training examples to induce new concept descriptions. Forgetting mechanisms also may be active to remove examples from partial memory that are irrelevant or outdated for the learning task. Using an implementation of the method, we conducted a lesion study and a direct comparison to examine the effects of partial memory learning on predictive accuracy and on the number of training examples maintained during learning. These experiments involved the STAGGER Concepts, a synthetic problem, and two real-world problems: a blasting cap detection problem and a computer intrusion detection problem. Experimental results suggest that the partial memory learner notably reduced memory requirements at the slight expense of predictive accuracy, and tracked concept drift as well as other learners designed for this task.

Adaptive inference for Bayesian network soft‐sensor in the presence of process and sensor drift

“…Moreover, the past value of the newly introduced hidden state, v s,kÀ1 , can be seen as an arrival cost-value, carrying all the past information of the drift to the current time instance, so the drifting state can be estimated. Thus, the analytical expressions for the hidden state value of the slow rate target node x s,k vary depending on the availability of laboratory data, that is, if 4 ¼ 0 or 4 ¼ 1: Thus, using first-order optimality conditions on Equation (10) results in a set of simultaneous linear equations, where estimates of quality variable (x s ) at k th instance is obtained through the following analytical solutions given in Equations ( 11) and (12). In Equations ( 11) and ( 12), N Ch refers to the number of child nodes of the target variable node (x s ) and σ 2 ch x s ð Þ refers to the variance between the target node x s ð Þ and its child node.…”

“…[7,8] To deal with drifts, first, a drift has to be detected through symptoms, such as degrading model performance. [9] Once the drift is detected, it is handled using either instance selection (moving window techniques), [10] instance weighting (recursive adaptation techniques [11] ), or ensemble methods. [12] Examples of the existing moving window-based adaptive approaches include blockwise and sample-wise moving window techniques.…”