- Abstract: Existing attention mechanisms, are mostly item-based in that a model is trained to attend to individual items in a collection (the memory) where each item has a predefined, fixed granularity, e.g., a character or a word. Intuitively, an area in the memory consisting of multiple items can be worth attending to as a whole. We propose area attention: a way to attend to an area of the memory, where each area contains a group of items that are either spatially adjacent when the memory has a 2-dimensional structure, such as images, or temporally adjacent for 1-dimensional memory, such as natural language sentences. Importantly, the size of an area, i.e., the number of items in an area or the level of aggregation, is dynamically determined via learning, which can vary depending on the learned coherence of the adjacent items. By giving the model the option to attend to an area of items, instead of only individual items, a model can attend to information with varying granularity. Area attention can work along multi-head attention for attending to multiple areas in the memory. We evaluate area attention on two tasks: neural machine translation (both character and token-level) and image captioning, and improve upon strong (state-of-the-art) baselines in all the cases. These improvements are obtainable with a basic form of area attention that is parameter free. In addition to proposing the novel concept of area attention, we contribute an efficient way for computing it by leveraging the technique of summed area tables.
- Keywords: Deep Learning, attentional mechanisms, neural machine translation, image captioning
- TL;DR: The paper presents a novel approach for attentional mechanisms that can benefit a range of tasks such as machine translation and image captioning.