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/

Zendy Plus

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

Zendy Tools

Zendy Open

Since their inception, learning techniques under the Reservoir Computingparadigm have shown a great modeling capability for recurrent systems withoutthe computing overheads required for other approaches. Among them, differentflavors of echo state networks have attracted many stares through time, mainlydue to the simplicity and computational efficiency of their learning algorithm.However, these advantages do not compensate for the fact that echo statenetworks remain as black-box models whose decisions cannot be easily explainedto the general audience. This work addresses this issue by conducting anexplainability study of Echo State Networks when applied to learning tasks withtime series, image and video data. Specifically, the study proposes threedifferent techniques capable of eliciting understandable information about theknowledge grasped by these recurrent models, namely, potential memory, temporalpatterns and pixel absence effect. Potential memory addresses questions relatedto the effect of the reservoir size in the capability of the model to storetemporal information, whereas temporal patterns unveils the recurrentrelationships captured by the model over time. Finally, pixel absence effectattempts at evaluating the effect of the absence of a given pixel when the echostate network model is used for image and video classification. We showcase thebenefits of our proposed suite of techniques over three different domains ofapplicability: time series modeling, image and, for the first time in therelated literature, video classification. Our results reveal that the proposedtechniques not only allow for a informed understanding of the way these modelswork, but also serve as diagnostic tools capable of detecting issues inheritedfrom data (e.g. presence of hidden bias).

On the Post-hoc Explainability of Deep Echo State Networks for Time Series Forecasting, Image and Video Classification