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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

There has been a lot of interest in understanding what information iscaptured by hidden representations of language models (LMs). Typically,interpretation methods i) do not guarantee that the model actually uses theencoded information, and ii) do not discover small subsets of neuronsresponsible for a considered phenomenon. Inspired by causal mediation analysis,we propose a method that discovers within a neural LM a small subset of neuronsresponsible for a particular linguistic phenomenon, i.e., subsets causing achange in the corresponding token emission probabilities. We use adifferentiable relaxation to approximately search through the combinatorialspace. An $L_0$ regularization term ensures that the search converges todiscrete and sparse solutions. We apply our method to analyze subject-verbnumber agreement and gender bias detection in LSTMs. We observe that it is fastand finds better solutions than the alternative (REINFORCE). Our experimentsconfirm that each of these phenomenons is mediated through a small subset ofneurons that do not play any other discernible role.

Sparse Interventions in Language Models with Differentiable Masking