Sparsifying Transformer Models with Trainable Representation Pooling

37091

Knowledge Triples

Source Entity	Relationship	Target Entity
Sparsify Attention Method	appliesTo	Transformer Model Method
Sparsify Attention Method	improves	InferenceEfficiency Metric
Sparsify Attention Method	improves	Long Document Summarization Task
Sparsify Attention Method	reduces	Training Time Metric
Sparsify Attention Method	uses	Top K Operator Method
Top K Operator Method	enhances	Computational Efficiency Metric