During the meta-testing, we randomly sample 2000 snippets from the test set and report the average classification top-one accuracy with a 95% confidence interval. We experiment with our method in five-way, one-shot and five-way, five-shot settings. 4.3. Comparison to Existing Methods ...