Reviews: Noise-Tolerant Life-Long Matrix Completion via Adaptive Sampling

The model is such that the learner are allowed to adaptively choose to unifomrly sample d features, or do a full measurement. Two noise models are studied, namely the bounded deterministic noise model, and the sparse random noise model. The authors provide recovery guarantees for both cases (robust recovery for the first case and exact recovery for the second case). The result matches or improved the state-of-art. However, notice that results in literature are for passive learning, where the authors assumes active learning/adaptive sampling, so the comparison is just illustrative.