Learning Half-Spaces from Perturbed Contrastive Examples

Aryan Alavi Razavi Ravari, Farnam Mansouri, Yuxin Chen, Valentio Iverson, Adish Singla, Sandra Zilles · 2026

Read Learning Half-Spaces from Perturbed Contrastive Examples by Aryan Alavi Razavi Ravari, Farnam Mansouri, Yuxin Chen, Valentio Iverson, Adish Singla, Sandra Zilles (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02080.txt.

Description

Aryan Alavi Razavi Ravari and colleagues study learning under a two-step contrastive oracle, where each queried example pairs with an opposite-label example. While earlier work assumes an idealized setting with the contrastive example at minimum distance, they introduce a non-decreasing noise function f perturbing the ideal. Perturbation is controlled by f of d, the distance to the boundary, yielding higher-quality examples near it. They study fixed and stochastic magnitudes. For one-dimensional thresholds and half-spaces under uniform distribution, they characterize active and passive complexity.

Catalog record

CREATORAryan Alavi Razavi Ravari, Farnam Mansouri, Yuxin Chen, Valentio Iverson, Adish Singla, Sandra Zilles
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE88646
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.02080

Browse the full library · About the Tower of Babel