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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

The prior knowledge (a.k.a. priors) integrated into the design of a machinelearning system strongly influences its generalization abilities. In thespecific context of deep learning, some of these priors are poorly understoodas they implicitly emerge from the successful heuristics and tentativeapproximations of biological brains involved in deep learning design. Throughthe lens of supervised image classification problems, this thesis investigatesthe implicit integration of a natural clustering prior composed of threestatements: (i) natural images exhibit a rich clustered structure, (ii) imageclasses are composed of multiple clusters and (iii) each cluster containsexamples from a single class. The decomposition of classes into multipleclusters implies that supervised deep learning systems could benefit fromunsupervised clustering to define appropriate decision boundaries. Hence, thisthesis attempts to identify implicit clustering abilities, mechanisms andhyperparameters in deep learning systems and evaluate their relevance forexplaining the generalization abilities of these systems. We do so through anextensive empirical study of the training dynamics as well as the neuron- andlayer-level representations of deep neural networks. The resulting collectionof experiments provides preliminary evidence for the relevance of the naturalclustering prior for understanding deep learning.

Towards understanding deep learning with the natural clustering prior