Top-down and bottom-up attention for joint pattern classification and reconstruction

Ricardo A Veiga, Bonifacio Silvano Zanutto · 2026

Read Top-down and bottom-up attention for joint pattern classification and reconstruction by Ricardo A Veiga, Bonifacio Silvano Zanutto (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0351985.txt.

Description

Ricardo Veiga and Bonifacio Zanutto introduce a new pattern-recognition method in this 2026 paper. The problem is untangling two patterns printed on top of each other. Ordinary classifiers see only the confusing mixture. Their framework, called CROP, alternates between two kinds of attention. Bottom-up steps guess which pattern is present, and top-down steps rebuild it.

Each round, the method names the most likely pattern and reconstructs its clean form. A mask then removes that pattern from the mixture. The leftover becomes clearer, so the next round can name the second pattern. Training needs only clean examples, never labelled mixtures. On overlapping handwritten digits, the procedure separated and labelled both numbers successfully. The design mimics how human attention picks one object out of a cluttered scene.

Catalog record

CREATORRicardo A Veiga, Bonifacio Silvano Zanutto
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE49951
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0351985

Browse the full library · About the Tower of Babel