NeurIPS 2020
### Optimal Learning from Verified Training Data

### Meta Review

This paper focuses on solving least squares regression problem in the setting where some of the inputs are strategically corrupted. The authors show that certain special setting of the general problem, there exists an algorithm that achieves a provably optimal solution.
Overall, the results are interesting and target an improved problems. The main concerns are regarding the significance of the results (as the setting targeted here seems to be fairly narrow) and providing a proper context wrt previous work (this hopefully can be remedied in the revision).