Saccade Attention Networks: Using Transfer Learning of Attention to Reduce Network Sizes
This work addresses the computational bottleneck of transformer networks for sequence processing, offering a method to reduce model size and computation while maintaining accuracy.
The authors tackle the quadratic complexity of attention in transformers by introducing Saccade Attention Networks, which learn sparse attention from a pre-trained model to reduce input sequence length. They achieve nearly 80% reduction in computation with similar performance.
One of the limitations of transformer networks is the sequence length due to the quadratic nature of the attention matrix. Classical self attention uses the entire sequence length, however, the actual attention being used is sparse. Humans use a form of sparse attention when analyzing an image or scene called saccades. Focusing on key features greatly reduces computation time. By using a network (Saccade Attention Network) to learn where to attend from a large pre-trained model, we can use it to pre-process images and greatly reduce network size by reducing the input sequence length to just the key features being attended to. Our results indicate that you can reduce calculations by close to 80% and produce similar results.