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

With the ever increasing data deluge and the success of deep neural networks,the research of distributed deep learning has become pronounced. Two commonapproaches to achieve this distributed learning is synchronous and asynchronousweight update. In this manuscript, we have explored very simplistic synchronousweight update mechanisms. It has been seen that with an increasing number ofworker nodes, the performance degrades drastically. This effect has beenstudied in the context of extreme imbalanced classification (e.g. outlierdetection). In practical cases, the assumed conditions of i.i.d. may not befulfilled. There may also arise global class imbalance situations like that ofoutlier detection where the local servers receive severely imbalanced data andmay not get any samples from the minority class. In that case, the DNNs in thelocal servers will get completely biased towards the majority class that theyreceive. This would highly impact the learning at the parameter server (whichpractically does not see any data). It has been observed that in a parallelsetting if one uses the existing federated weight update mechanisms at theparameter server, the performance degrades drastically with the increasingnumber of worker nodes. This is mainly because, with the increasing number ofnodes, there is a high chance that one worker node gets a very small portion ofthe data, either not enough to train the model without overfitting or having ahighly imbalanced class distribution. The chapter, hence, proposes a workaroundto this problem by introducing the concept of adaptive cost-sensitive momentumaveraging. It is seen that for the proposed system, there was no to minimaldegradation in performance while most of the other methods hit their bottomperformance before that.

Improving the Robustness of Federated Learning for Severely Imbalanced Datasets