Losing Heads in the Lottery: Pruning Transformer Attention in Neural Machine...

This paper presents a simple pruning method which could have been described in 3 sentences.

  1. Train the transformer for several steps.
  2. Prune some heads based on their confidence on devset.
  3. Reinitialize pruned heads and repeat 1-3 until converge.

Their results look good but their L_0 baseline must be under-tuned. I have experimented with L_0 a lot and it takes lots of efforts to tune. But once tuned well, L_0 can yield comparable results (within 0.1 loss) with at least 50\% heads pruned.


  • What will happen if we do not use the late resetting checkpoint? By not using it, we draw another lottery and have a chance to win too.

Overall Recommendation

  • 5: Transformative: This paper is likely to change our field. It should be considered for a best paper award.
  • 4.5: Exciting: It changed my thinking on this topic. I would fight for it to be accepted.
  • 4: Strong: I learned a lot from it. I would like to see it accepted.
  • 3.5: Leaning positive: It can be accepted more or less in its current form. However, the work it describes is not particularly exciting and/or inspiring, so it will not be a big loss if people don’t see it in this conference.
  • 3: Ambivalent: It has merits (e.g., it reports state-of-the-art results, the idea is nice), but there are key weaknesses (e.g., I didn’t learn much from it, evaluation is not convincing, it describes incremental work). I believe it can significantly benefit from another round of revision, but I won’t object to accepting it if my co-reviewers are willing to champion it.
  • 2.5: Leaning negative: I am leaning towards rejection, but I can be persuaded if my co-reviewers think otherwise.
  • 2: Mediocre: I would rather not see it in the conference.
  • 1.5: Weak: I am pretty confident that it should be rejected.
  • 1: Poor: I would fight to have it rejected.