Continual Learning in Transition

Source

Source name: arXiv
Source kind: paper
Authors / organization: Zhiyan Hou, Dan Zhang, Tao Feng, Liyuan Wang, Wei Li, Xiangzhao Hao, Hongyan An, Junfeng Fang, Haokai Ma, Zhaohui Xu, Haiyun Guo, Jinqiao Wang, Tat-Seng Chua
Venue: arXiv: cs.LG
Published: 2026-08-06
Collected: 2026-08-07
URL: http://arxiv.org/abs/2608.06216v1
PDF: https://arxiv.org/pdf/2608.06216v1
DOI: UNKNOWN

Plain-Language Brief

The text discusses a shift in continual learning (CL) away from just updating parameters and towards system-level adaptation. The paper explores three dimensions of this transition: When (pre-training, post-training, inference), How (off-policy, on-policy, beyond-gradient optimization), and Where (internal parameters vs external structural constraints). The authors survey various CL methods, trace the evolution of CL, and discuss challenges, implications, and future directions.

What the Document Says

Continual learning (CL) is transitioning from parameter-centric learning to system-level adaptation
The transition involves three dimensions: When, How, and Where learning occurs
When dimension includes pre-training, post-training, and inference stages
How dimension covers off-policy, on-policy, and beyond-gradient optimization mechanics
Where dimension refers to updates within internal parameters versus external structural constraints
The paper surveys various CL methods, traces the evolution of CL, and discusses challenges, implications, and future directions

Key Concepts Mentioned

Continual learning
parameter-centric learning
system-level adaptation
off-policy learning
on-policy learning
beyond-gradient optimization
pre-training
post-training
inference
internal parameters
external structural constraints

Claims Made by the Document

Continual learning is transitioning from parameter-centric to system-level adaptation

Evidence Used

The text states that emerging paradigms are reshaping the scope of CL beyond traditional model adaptation view, and mentions various developments such as on-policy learning, test-time training, and external harness components.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Classical continual learning (CL) has primarily focused on enabling models to update and retain knowledge through parameter-centric mechanisms, e.g., training strategies, architectural designs, and weight adaptation. However, emerging paradigms are reshaping the scope of CL beyond this traditional model adaptation view. For instance, on-policy learning broadens the space of update mechanisms; test-time training extends CL from the training phase to inference; and external harness components such as memory, skill libraries, and interaction protocols extend the evolutionary boundaries of model capabilities far beyond the static parameter space. Collectively, these developments indicate a transition from parameter-centric learning toward system-level adaptation. To characterize this transition, we examine the evolution of continual learning through three dimensions: When, How, and Where learning occurs. The How dimension encompasses off-policy, on-policy, and beyond-gradient optimization mechanics. The When dimension captures evolution across pre-training, post-training, and inference-time stages. The Where dimension delineates updates occurring within internal parameters versus external structural constraints. Anchored by this tri-axial framework, we systematically survey representative methods, trace the ongoing transition of continual learning, and discuss the key challenges, broader implications, and future directions arising from this paradigm shift.