Model Monitoring in the Absence of Labeled Data via Feature Attributions Distributions

Carlos Mougan Navarro · 2025

Read Model Monitoring in the Absence of Labeled Data via Feature Attributions Distributions by Carlos Mougan Navarro (2025) 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-2501-10774.txt.

Introduction

Carlos Mougan's thesis tackles a hard gap in AI care: watching a deployed model when no labeled answers exist. Once a model is live, its true accuracy is often unknown for months. Yet drift and quiet failure can start at once. Mougan looks for warning signs inside the model's own explanations.

The thesis splits monitoring into two strands. One checks that the model still behaves in line with human values. The other checks that it still meets its accuracy goals. Both strands track how feature attributions, the model's stated reasons, are spread across cases. Shifts in that spread can betray change before harm lands. The author draws guarantees from the theory of these explanations. The work offers a shared method for early, label-free oversight of AI systems.

Catalog record

CREATORCarlos Mougan Navarro
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE273 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-10774

Browse the full library · About the Tower of Babel