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

We introduce the novel concept of anti-transfer learning for speechprocessing with convolutional neural networks. While transfer learning assumesthat the learning process for a target task will benefit from re-usingrepresentations learned for another task, anti-transfer avoids the learning ofrepresentations that have been learned for an orthogonal task, i.e., one thatis not relevant and potentially misleading for the target task, such as speakeridentity for speech recognition or speech content for emotion recognition. Inanti-transfer learning, we penalize similarity between activations of a networkbeing trained and another one previously trained on an orthogonal task, whichyields more suitable representations. This leads to better generalization andprovides a degree of control over correlations that are spurious orundesirable, e.g. to avoid social bias. We have implemented anti-transfer forconvolutional neural networks in different configurations with severalsimilarity metrics and aggregation functions, which we evaluate and analyzewith several speech and audio tasks and settings, using six datasets. We showthat anti-transfer actually leads to the intended invariance to the orthogonaltask and to more appropriate features for the target task at hand.Anti-transfer learning consistently improves classification accuracy in alltest cases. While anti-transfer creates computation and memory cost at trainingtime, there is relatively little computation cost when using pre-trained modelsfor orthogonal tasks. Anti-transfer is widely applicable and particularlyuseful where a specific invariance is desirable or where trained models areavailable and labeled data for orthogonal tasks are difficult to obtain.

Anti-Transfer Learning for Task Invariance in Convolutional Neural Networks for Speech Processing