Publikation
Continual Learning Should Move Beyond Incremental Classification
Rupert Mitchell; Antonio Alliegro; Raffaello Camoriano; Dustin Carrión-Ojeda; Antonio Carta; Georgia Chalvatzaki; Nikhil Churamani; Carlo D'Eramo; Samin Hamidi; Robin Hesse; Fabian Hinder; Roshni Kamath; Vincenzo Lomonaco; Subarnaduti Paul; Francesca Pistilli; Tinne Tuytelaars; Gido M. van de Ven; Kristian Kersting; Simone Schaub-Meyer; Martin Mundt
In: Computing Research Repository eprint Journal (CoRR), Vol. abs/2502.11927, Pages 1-11, Computing Research Repository, 2025.
Zusammenfassung
Continual learning (CL) is the sub-field of ma-
chine learning concerned with accumulating
knowledge in dynamic environments. So far, CL
research has mainly focused on incremental clas-
sification tasks, where models learn to classify
new categories while retaining knowledge of pre-
viously learned ones. Here, we argue that main-
taining such a focus limits both theoretical devel-
opment and practical applicability of CL methods.
Through a detailed analysis of concrete examples
— including multi-target classification, robotics
with constrained output spaces, learning in con-
tinuous task domains, and higher-level concept
memorization — we demonstrate how current CL
approaches often fail when applied beyond stan-
dard classification. We identify three fundamen-
tal challenges: (C1) the nature of continuity in
learning problems, (C2) the choice of appropri-
ate spaces and metrics for measuring similarity,
and (C3) the role of learning objectives beyond
classification. For each challenge, we provide spe-
cific recommendations to help move the field for-
ward, including formalizing temporal dynamics
through distribution processes, developing princi-
pled approaches for continuous task spaces, and
incorporating density estimation and generative
objectives. In so doing, this position paper aims to
broaden the scope of CL research while strength-
ening its theoretical foundations, making it more
applicable to real-world problems.
