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

Deep learning techniques have demonstrated significant capacity in modelingsome of the most challenging real world problems of high complexity. Despitethe popularity of deep models, we still strive to better understand theunderlying mechanism that drives their success. Motivated by observations thatneurons in trained deep nets predict attributes indirectly related to thetraining tasks, we recognize that a deep network learns representations moregeneral than the task at hand to disentangle impacts of multiple confoundingfactors governing the data, in order to isolate the effects of the concerningfactors and optimize a given objective. Consequently, we propose a generalframework to augment training of deep models with information on auxiliaryexplanatory data variables, in an effort to boost this disentanglement andtrain deep networks that comprehend the data interactions and distributionsmore accurately, and thus improve their generalizability. We incorporateinformation on prominent auxiliary explanatory factors of the data populationinto existing architectures as secondary objective/loss blocks that take inputsfrom hidden layers during training. Once trained, these secondary circuits canbe removed to leave a model with the same architecture as the original, butmore generalizable and discerning thanks to its comprehension of datainteractions. Since pose is one of the most dominant confounding factors forobject recognition, we apply this principle to instantiate a pose-aware deepconvolutional neural network and demonstrate that auxiliary pose informationindeed improves the classification accuracy in our experiments on SAR targetclassification tasks.

Enlightening Deep Neural Networks with Knowledge of Confounding Factors