Saved in:
Bibliographic Details
Main Authors: Sedaghat, Nima, Chatchadanoraset, Tanawan, Chandler, Colin Orion, Mahabal, Ashish, Eslami, Maryam
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2412.00077
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916500754923520
author Sedaghat, Nima
Chatchadanoraset, Tanawan
Chandler, Colin Orion
Mahabal, Ashish
Eslami, Maryam
author_facet Sedaghat, Nima
Chatchadanoraset, Tanawan
Chandler, Colin Orion
Mahabal, Ashish
Eslami, Maryam
contents Motivated by the scarcity of proper labels in an astrophysical application, we have developed a novel technique, called Selfish Evolution, which allows for the detection and correction of corrupted labels in a weakly supervised fashion. Unlike methods based on early stopping, we let the model train on the noisy dataset. Only then do we intervene and allow the model to overfit to individual samples. The ``evolution'' of the model during this process reveals patterns with enough information about the noisiness of the label, as well as its correct version. We train a secondary network on these spatiotemporal ``evolution cubes'' to correct potentially corrupted labels. We incorporate the technique in a closed-loop fashion, allowing for automatic convergence towards a mostly clean dataset, without presumptions about the state of the network in which we intervene. We evaluate on the main task of the Supernova-hunting dataset but also demonstrate efficiency on the more standard MNIST dataset.
format Preprint
id arxiv_https___arxiv_org_abs_2412_00077
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Selfish Evolution: Making Discoveries in Extreme Label Noise with the Help of Overfitting Dynamics
Sedaghat, Nima
Chatchadanoraset, Tanawan
Chandler, Colin Orion
Mahabal, Ashish
Eslami, Maryam
Computer Vision and Pattern Recognition
Instrumentation and Methods for Astrophysics
Artificial Intelligence
Machine Learning
Motivated by the scarcity of proper labels in an astrophysical application, we have developed a novel technique, called Selfish Evolution, which allows for the detection and correction of corrupted labels in a weakly supervised fashion. Unlike methods based on early stopping, we let the model train on the noisy dataset. Only then do we intervene and allow the model to overfit to individual samples. The ``evolution'' of the model during this process reveals patterns with enough information about the noisiness of the label, as well as its correct version. We train a secondary network on these spatiotemporal ``evolution cubes'' to correct potentially corrupted labels. We incorporate the technique in a closed-loop fashion, allowing for automatic convergence towards a mostly clean dataset, without presumptions about the state of the network in which we intervene. We evaluate on the main task of the Supernova-hunting dataset but also demonstrate efficiency on the more standard MNIST dataset.
title Selfish Evolution: Making Discoveries in Extreme Label Noise with the Help of Overfitting Dynamics
topic Computer Vision and Pattern Recognition
Instrumentation and Methods for Astrophysics
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2412.00077