Abstract: Can we perform Neural Architecture Search (NAS) with a smaller subset of target dataset and still fair better in terms of performance with significant reduction in search cost? In this work, ...