Unified probabilistic deep continual learning through generative replay and open set recognition
- Modern deep neural networks are well known to be brittle in the face of unknown data instances and recognition of the latter remains a challenge. Although it is inevitable for continual-learning systems to encounter such unseen concepts, the corresponding literature appears to nonetheless focus primarily on alleviating catastrophic interference with learned representations. In this work, we introduce a probabilistic approach that connects these perspectives based on variational inference in a single deep autoencoder model. Specifically, we propose to bound the approximate posterior by fitting regions of high density on the basis of correctly classified data points. These bounds are shown to serve a dual purpose: unseen unknown out-of-distribution data can be distinguished from already trained known tasks towards robust application. Simultaneously, to retain already acquired knowledge, a generative replay process can be narrowed to strictly in-distribution samples, in order to significantly alleviate catastrophic interference.
Author: | Martin MundtORCiDGND, Iuliia PliushchGND, Sagnik MajumderORCiD, Yongwon HongORCiD, Visvanathan RameshORCiD |
---|---|
URN: | urn:nbn:de:hebis:30:3-828237 |
DOI: | https://doi.org/10.3390/jimaging8040093 |
ISSN: | 2313-433X |
Parent Title (English): | Journal of imaging |
Publisher: | MDPI |
Place of publication: | Basel |
Document Type: | Article |
Language: | English |
Date of Publication (online): | 2022/03/31 |
Date of first Publication: | 2022/03/31 |
Publishing Institution: | Universitätsbibliothek Johann Christian Senckenberg |
Release Date: | 2024/06/25 |
Tag: | catastrophic forgetting; continual deep learning; deep generative models; open-set recognition; variational inference |
Volume: | 8 |
Issue: | 4, art. 93 |
Article Number: | 93 |
Page Number: | 34 |
First Page: | 1 |
Last Page: | 34 |
Note: | Funding: EU H2020 Project AEROBI ; 687384 |
Note: | Funding: EU H2020 Project RESIST ; 769066 |
Note: | Funding: BMBF project AISEL ; 01IS19062 |
Note: | Additional financial support from Goethe University was instrumental in concluding the research. |
HeBIS-PPN: | 520872789 |
Institutes: | Informatik und Mathematik |
Dewey Decimal Classification: | 0 Informatik, Informationswissenschaft, allgemeine Werke / 00 Informatik, Wissen, Systeme / 004 Datenverarbeitung; Informatik |
Sammlungen: | Universitätspublikationen |
Licence (German): | ![]() |