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

Modern machine learning systems are increasingly characterized by extensivepersonal data collection, despite the diminishing returns and increasingsocietal costs of such practices. Yet, data minimisation is one of the coredata protection principles enshrined in the European Union's General DataProtection Regulation ('GDPR') and requires that only personal data that isadequate, relevant and limited to what is necessary is processed. However, theprinciple has seen limited adoption due to the lack of technicalinterpretation.  In this work, we build on literature in machine learning and law to proposeFIDO, a Framework for Inhibiting Data Overcollection. FIDO learns to limit datacollection based on an interpretation of data minimization tied to systemperformance. Concretely, FIDO provides a data collection stopping criterion byiteratively updating an estimate of the performance curve, or the relationshipbetween dataset size and performance, as data is acquired. FIDO estimates theperformance curve via a piecewise power law technique that models distinctphases of an algorithm's performance throughout data collection separately.Empirical experiments show that the framework produces accurate performancecurves and data collection stopping criteria across datasets and featureacquisition algorithms. We further demonstrate that many other families ofcurves systematically overestimate the return on additional data. Results andanalysis from our investigation offer deeper insights into the relevantconsiderations when designing a data minimization framework, including theimpacts of active feature acquisition on individual users and the feasabilityof user-specific data minimization. We conclude with practical recommendationsfor the implementation of data minimization.

Learning to Limit Data Collection via Scaling Laws: A Computational Interpretation for the Legal Principle of Data Minimization