Interpreting intentionally flawed models with linear probes

Graziani, Mara (University of Applied Sciences and Arts Western Switzerland (HES-SO Valais-Wallis)) ; Müller, Henning (University of Applied Sciences and Arts Western Switzerland (HES-SO Valais-Wallis); University of Geneva, Switzerland) ; Andrearczyk, Vincent (University of Applied Sciences and Arts Western Switzerland (HES-SO Valais-Wallis))

The representational differences between generalizing networks and intentionally flawed models can be insightful on the dynamics of network training. Do memorizing networks, e.g. networks that learn random label correspondences, focus on specific patterns in the data to memorize the labels? Are the features learned by a generalizing network affected by randomization of the model parameters? In high-risk applications such as medical, legal or financial domains, highlighting the representational differences that help generalization may be even more important than the model performance itself. In this paper, we probe the activations of intermediate layers with linear classification and regression. Results show that the bias towards simple solutions of generalizing networks is maintained even when statistical irregularities are intentionally introduced.


Conference Type:
full paper
Faculty:
Economie et Services
School:
HEG-VS
Institute:
Institut Informatique de gestion
Subject(s):
Informatique
Publisher:
Seoul, Korea, 27 October 2019
Date:
2019-10
Seoul, Korea
27 October 2019
Pagination:
5 p.
Published in:
Proceedings of the first workshop on Statistical Deep Learning for Computer Vision (ICCV 2019)
Appears in Collection:



 Record created 2019-10-25, last modified 2020-03-25

Fulltext:
Download fulltext
PDF

Rate this document:

Rate this document:
1
2
3
 
(Not yet reviewed)