id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_n3xwttbopngatcfdmiu7tbqtxu	Alasdair Tran	Combining active learning suggestions	2018	34	.pdf	application/pdf	13974	1807	69	Keywords Active learning, Bandit, Rank aggregation, Benchmark, Multiclass classification propose to combine active learning suggestions with bandit and rank aggregation In active learning, we use the class probability estimates from a trained classifier to learning, where we select an object from a pool of unlabeled examples at each time step, estimate the true classifier performance using only the training set. Algorithm 2 Pool-based active learning with bandit theory. learning heuristics ℛ and the test set ℒS, some bandit algorithms also need to know n, the maximum active learning heuristics ℛ, and bandit algorithm b with two functions SELECT and UPDATE. for each i and select heuristic r� that has the highest sampled value of the mean reward: Figure 2 Active learning pipeline with rank aggregation methods. Input: unlabeled set U, labeled training set ℒT, classifier h, set of active learning suggestions R, ranking learning heuristics, five bandit algorithms, and three aggregation methods.	./cache/work_n3xwttbopngatcfdmiu7tbqtxu.pdf	./txt/work_n3xwttbopngatcfdmiu7tbqtxu.txt