English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The selective visual attention mechanism in the human visual system (HVS)restricts the amount of information to reach visual awareness for perceivingnatural scenes, allowing near real-time information processing with limitedcomputational capacity [Koch and Ullman, 1987]. This kind of selectivity actsas an 'Information Bottleneck (IB)', which seeks a trade-off betweeninformation compression and predictive accuracy. However, such informationconstraints are rarely explored in the attention mechanism for deep neuralnetworks (DNNs). In this paper, we propose an IB-inspired spatial attentionmodule for DNN structures built for visual recognition. The module takes asinput an intermediate representation of the input image, and outputs avariational 2D attention map that minimizes the mutual information (MI) betweenthe attention-modulated representation and the input, while maximizing the MIbetween the attention-modulated representation and the task label. To furtherrestrict the information bypassed by the attention map, we quantize thecontinuous attention scores to a set of learnable anchor values duringtraining. Extensive experiments show that the proposed IB-inspired spatialattention mechanism can yield attention maps that neatly highlight the regionsof interest while suppressing backgrounds, and bootstrap standard DNNstructures for visual recognition tasks (e.g., image classification,fine-grained recognition, cross-domain classification). The attention maps areinterpretable for the decision making of the DNNs as verified in theexperiments. Our code is available at https://github.com/ashleylqx/AIB.git.

Information Bottleneck Approach to Spatial Attention Learning