Perks of Being Lazy: Boosting Retrieval Performance

Fast anytime retrieval with confidence in large-scale temporal case bases

“…a CB of electronic health records of patients. Our algorithm is based on Lazy kNN [7] which is an effective exact kNN algorithm in CBR literature for such domains. However, for some applications, the notable speed-up provided by this algorithm may not suffice and the execution time for exact kNN may still be intolerable.…”

“…As a more recent work, Lazy kNN [7] which our algorithm is based on is a novel NNS which significantly reduces the number of evaluated cases in kNN search for large-scale CBs composed of temporally related cases. Lazy kNN leverages the triangle inequality in metric spaces by using it as a cutoff in NNS and evaluates only the ''true kNN candidate'' cases for a given query.…”

“…In this section, we first deal with the design of an anytime kNN algorithm that exhibits the desired performance profile outlined in Section 2.3. For the core of the AA, we propose to extend an existing algorithm, Lazy kNN [7]. Specifically, although Lazy kNN was designed to provide exact kNNs as a whole list after running to completion, we show how it can be converted to an incremental kNN search without any redundant computation in Section 3.1.…”

“…algorithm, Anytime Lazy kNN (ALK ). We base our algorithm on a fast exact kNN algorithm Lazy kNN [7], and we extend it to a fullyfledged anytime algorithm. ALK finds exact kNNs when allowed to run to completion and otherwise, if interrupted, it returns best-so-far kNNs together with a confidence value attached to each neighbor.…”