CONCLAD: COntinuous Novel CLAss Detector

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Rios, Amanda, Ndiour, Ibrahima, Datta, Parual, Tickoo, Omesh, Ahuja, Nilesh
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866929629805150208
author Rios, Amanda
Ndiour, Ibrahima
Datta, Parual
Tickoo, Omesh
Ahuja, Nilesh
author_facet Rios, Amanda
Ndiour, Ibrahima
Datta, Parual
Tickoo, Omesh
Ahuja, Nilesh
contents In the field of continual learning, relying on so-called oracles for novelty detection is commonplace albeit unrealistic. This paper introduces CONCLAD ("COntinuous Novel CLAss Detector"), a comprehensive solution to the under-explored problem of continual novel class detection in post-deployment data. At each new task, our approach employs an iterative uncertainty estimation algorithm to differentiate between known and novel class(es) samples, and to further discriminate between the different novel classes themselves. Samples predicted to be from a novel class with high-confidence are automatically pseudo-labeled and used to update our model. Simultaneously, a tiny supervision budget is used to iteratively query ambiguous novel class predictions, which are also used during update. Evaluation across multiple datasets, ablations and experimental settings demonstrate our method's effectiveness at separating novel and old class samples continuously. We will release our code upon acceptance.
format Preprint
id arxiv_https___arxiv_org_abs_2412_10473
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle CONCLAD: COntinuous Novel CLAss Detector
Rios, Amanda
Ndiour, Ibrahima
Datta, Parual
Tickoo, Omesh
Ahuja, Nilesh
Machine Learning
Artificial Intelligence
In the field of continual learning, relying on so-called oracles for novelty detection is commonplace albeit unrealistic. This paper introduces CONCLAD ("COntinuous Novel CLAss Detector"), a comprehensive solution to the under-explored problem of continual novel class detection in post-deployment data. At each new task, our approach employs an iterative uncertainty estimation algorithm to differentiate between known and novel class(es) samples, and to further discriminate between the different novel classes themselves. Samples predicted to be from a novel class with high-confidence are automatically pseudo-labeled and used to update our model. Simultaneously, a tiny supervision budget is used to iteratively query ambiguous novel class predictions, which are also used during update. Evaluation across multiple datasets, ablations and experimental settings demonstrate our method's effectiveness at separating novel and old class samples continuously. We will release our code upon acceptance.
title CONCLAD: COntinuous Novel CLAss Detector
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2412.10473