Bayesian selective scan for robust long-horizon action recognition via reliability-aware discretization

Chunjie Wang · 2026

Read Bayesian selective scan for robust long-horizon action recognition via reliability-aware discretization by Chunjie Wang (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/pmc-13547921.txt.

Description

Chunjie Wang, a solo author, presents a video model that stays steady when parts of a clip are corrupted. Long action clips in live streams can suffer brief blur or attacks that poison a model's memory, causing errors long after the bad frames end. The 2026 paper builds on a fast model family that scans frames in order.

The new model judges each frame's trust level from two kinds of uncertainty, then slows its memory updates when frames look unreliable. On two standard action sets, clean scores matched or slightly beat a plain version of the same model. Under bursty blur and crafted attacks, accuracy held up better, and errors faded far faster once the corruption passed.

Catalog record

CREATORChunjie Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC13547921

Browse the full library · About the Tower of Babel