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

In-context learning, i.e., learning from in-context samples, is an impressiveability of Transformer. However, the mechanism driving the in-context learningis not yet fully understood. In this study, we aim to investigate from anunderexplored perspective of representation learning. The representation ismore complex for in-context learning senario, where the representation can beimpacted by both model weights and in-context samples. We refer the above twoconceptually aspects of representation as in-weight component and in-contextcomponent, respectively. To study how the two components affect in-contextlearning capabilities, we construct a novel synthetic task, making it possibleto device two probes, in-weights probe and in-context probe, to evaluate thetwo components, respectively. We demonstrate that the goodness of in-contextcomponent is highly related to the in-context learning performance, whichindicates the entanglement between in-context learning and representationlearning. Furthermore, we find that a good in-weights component can actuallybenefit the learning of the in-context component, indicating that in-weightslearning should be the foundation of in-context learning. To further understandthe the in-context learning mechanism and importance of the in-weightscomponent, we proof by construction that a simple Transformer, which usespattern matching and copy-past mechanism to perform in-context learning, canmatch the in-context learning performance with more complex, best tunedTransformer under the perfect in-weights component assumption. In short, thosediscoveries from representation learning perspective shed light on newapproaches to improve the in-context capacity.

How does representation impact in-context learning: A exploration on a synthetic task