Home › Categories › Technology› Mythos Of Model Interpretability

Mythos Of Model Interpretability.pdf

1606.03490.pdf

Preview of Mythos of model interpretability

🔗 Source: arxiv.org

📊 Size: 359 KB

👤 Author: Zachary C. Lipton

⬇️ Downloads: 136

Summary

Supervised machine learning models have remarkable predictive capabilities, but their interpretability is underspecified, with diverse and sometimes non-overlapping motivations and notions of what attributes render models interpretable. The task of interpretation appears ambiguous, with many papers proclaiming interpretability axiomatically without further explanation. The motivations underlying interest in interpretability are diverse, including trust, understanding, and uncovering causal structure in data. Model properties and techniques thought to confer interpretability include transparency to humans and post-hoc explanations, which are competing notions. The feasibility and desirability of different notions of interpretability are discussed, and the assertions that linear models are interpretable and deep neural networks are not are questioned.

Download File

Description

Supervised machine learning models have remarkable predictive capabilities, but their interpretability is underspecified, with diverse and sometimes...

Technical Information

File Format: PDF
File Size: 359 KB
Pages: 9
Language: EN
Author: Zachary C. Lipton
Total Downloads: 136
Last Updated: 2 weeks ago

Document Overview

This PDF document about Mythos of model interpretability provides comprehensive information and guidance. Whether you're a beginner or advanced user, this resource offers valuable insights into Mythos of model interpretability.