The paper proposes a gradient-based method for tackling multi-task learning problem, in which "conflicting" gradients are detected and altered so that such situations would not harm the common objective. The reviewers came to a consensus that the proposed technology is sound and the paper has its value to the NeurIPS community. However, the reviewers also gave some constructive feedback, e.g., moving the comparisons with important baselines to the main body of the paper, instead of placing them in the appendix. We hope the authors could follow these suggestions when preparing their final version of the paper,