Dissecting Lottery Ticket Transformers: Structural and Behavioral Study of Sparse Neural Machine Translation
This work provides insights into the structural and behavioral changes in sparse models for machine translation, which is incremental but useful for understanding pruning effects.
The study investigated how pruning Transformers for neural machine translation affects learned representations, finding that complex semantic information degrades first and higher layers diverge most, while attention mechanisms remain consistent.
Recent work on the lottery ticket hypothesis has produced highly sparse Transformers for NMT while maintaining BLEU. However, it is unclear how such pruning techniques affect a model's learned representations. By probing Transformers with more and more low-magnitude weights pruned away, we find that complex semantic information is first to be degraded. Analysis of internal activations reveals that higher layers diverge most over the course of pruning, gradually becoming less complex than their dense counterparts. Meanwhile, early layers of sparse models begin to perform more encoding. Attention mechanisms remain remarkably consistent as sparsity increases.