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

Neural networks can be trained to solve regression problems by usinggradient-based methods to minimize the square loss. However, practitionersoften prefer to reformulate regression as a classification problem, observingthat training on the cross entropy loss results in better performance. Byfocusing on two-layer ReLU networks, which can be fully characterized bymeasures over their feature space, we explore how the implicit bias induced bygradient-based optimization could partly explain the above phenomenon. Weprovide theoretical evidence that the regression formulation yields a measurewhose support can differ greatly from that for classification, in the case ofone-dimensional data. Our proposed optimal supports correspond directly to thefeatures learned by the input layer of the network. The different nature ofthese supports sheds light on possible optimization difficulties the squareloss could encounter during training, and we present empirical resultsillustrating this phenomenon.

Regression as Classification: Influence of Task Formulation on Neural Network Features