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

The work concerns the problem of reducing a pre-trained deep neuronal networkto a smaller network, with just few layers, whilst retaining the network'sfunctionality on a given task  The proposed approach is motivated by the observation that the aim to deliverthe highest accuracy possible in the broadest range of operational conditions,which many deep neural networks models strive to achieve, may not necessarilybe always needed, desired, or even achievable due to the lack of data ortechnical constraints. In relation to the face recognition problem, weformulated an example of such a usecase, the `backyard dog' problem. The`backyard dog', implemented by a lean network, should correctly identifymembers from a limited group of individuals, a `family', and should distinguishbetween them. At the same time, the network must produce an alarm to an imageof an individual who is not in a member of the family. To produce such anetwork, we propose a shallowing algorithm. The algorithm takes an existingdeep learning model on its input and outputs a shallowed version of it. Thealgorithm is non-iterative and is based on the Advanced Supervised PrincipalComponent Analysis. Performance of the algorithm is assessed in exhaustivenumerical experiments. In the above usecase, the `backyard dog' problem, themethod is capable of drastically reducing the depth of deep learning neuralnetworks, albeit at the cost of mild performance deterioration.  We developed a simple non-iterative method for shallowing down pre-traineddeep networks. The method is generic in the sense that it applies to a broadclass of feed-forward networks, and is based on the Advanced SupervisePrincipal Component Analysis. The method enables generation of families ofsmaller-size shallower specialized networks tuned for specific operationalconditions and tasks from a single larger and more universal legacy network.

How deep should be the depth of convolutional neural networks: a backyard dog case study