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

One of the challenges in learning to perform abstract reasoning is thatproblems are often posed as monolithic tasks, with no intermediate subgoals. InRaven Progressive Matrices (RPM), the task is to choose one of the availableanswers given a context, where both contexts and answers are composite imagesfeaturing multiple objects in various spatial arrangements. As this high-levelgoal is the only guidance available, learning is challenging and mostcontemporary solvers tend to be opaque. In this study, we propose a deeplearning architecture based on the transformer blueprint which, rather thandirectly making the above choice, predicts the visual properties of individualobjects and their arrangements. The multidimensional predictions obtained inthis way are then directly juxtaposed to choose the answer. We consider a fewways in which the model parses the visual input into tokens and several regimesof masking parts of the input in self-supervised training. In experimentalassessment, the models not only outperform state-of-the-art methods but alsoprovide interesting insights and partial explanations about the inference. Thedesign of the method also makes it immune to biases that are known to exist insome RPM benchmarks.

Learning Abstract Visual Reasoning via Task Decomposition: A Case Study in Raven Progressive Matrices