Research map / Continual learning
Online and replay-based: research map
693 papers on Online and replay-based in Continual learning, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 2 clusters and 8 approaches. The busiest year so far is 2026.
What it is
A model that keeps learning from a stream of new data tends to forget what it learned before ("catastrophic forgetting"). This topic studies how to keep old skills while adding new ones, often by storing a small memory of past examples and replaying them, or by protecting the weights that old tasks depend on.
Terms you will see on this page:
- Catastrophic forgetting
- the sharp drop in accuracy on earlier tasks after training on new ones.
- Replay buffer
- a small store of past examples mixed into training on new data.
- Plasticity
- the ability to keep learning new things; the opposite worry to forgetting.
Start reading with the most cited papers here:
- PackNet: Adding Multiple Tasks to a Single Network by Iterative Pruning (CVPR 2018 · 1,240 citations)
- Continual Learning Through Synaptic Intelligence (ICML 2017 · 1,007 citations)
Within Continual learning, its share shrank from 46.4% in 2023–24 to 34.2% in 2025–26 (163 → 197 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Online and replay-based in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Plasticity-preserving continual reinforcement learning rising: 5% → 14% of the topic (8 → 27 papers)
- Theoretical analysis of forgetting rising: 22% → 28% of the topic (36 → 55 papers)
- Replay buffer construction and update falling: 7% → 3% of the topic (12 → 5 papers)
- Dynamic expansion and isolation falling: 16% → 13% of the topic (26 → 26 papers)
Papers from 2025–2026 that are already cited most:
- Rethinking and Improving Student Learning and Forgetting Processes for Attention based Knowledge Tracing Models (AAAI 2025 · 17 citations)
Approaches and key papers
Forgetting, plasticity and growing networks · 546 papers
Keywords: weight, transfer, growing
Approaches in this cluster:
- Theoretical analysis of forgetting (165 papers)
Studies why and when networks forget, using linear models, teacher-student setups and complexity results. - Unified and unsupervised continual frameworks (160 papers)
Proposes general frameworks, benchmarks and architecture-growing methods for task-agnostic and unsupervised continual learning. - Gradient projection and plasticity preservation (92 papers)
Constrains updates to subspaces or preserves network trainability to limit forgetting. - Dynamic expansion and isolation (83 papers)
Grows or partitions networks and generators to keep task knowledge separate. - Plasticity-preserving continual reinforcement learning (46 papers)
Keeps policies adaptable using architectural design, policy subspaces and composition across tasks.
Most cited and most cited since 2024:
- PackNet: Adding Multiple Tasks to a Single Network by Iterative Pruning (CVPR 2018 · 1,240 citations)
- Continual Learning Through Synaptic Intelligence (ICML 2017 · 1,007 citations)
- Towards Continual Knowledge Graph Embedding via Incremental Distillation (AAAI 2024 · 42 citations)
- Towards Efficient Replay in Federated Incremental Learning (CVPR 2024 · 34 citations)
Online learning with replay · 147 papers
Keywords: online continual, replay, experience
Approaches in this cluster:
- Memory-based and Bayesian online learning (61 papers)
Handles blurry or task-free streams with expandable memory, Bayesian updates and meta-learning. - Plug-and-play online training enhancements (50 papers)
Adds distillation, collaborative learners or state space modules to improve plasticity of online learners. - Replay buffer construction and update (36 papers)
Improves what is stored and replayed, and how, to reduce representation drift online.
Most cited and most cited since 2024:
- Experience Replay for Continual Learning (NeurIPS 2019 · 368 citations)
- Rainbow Memory: Continual Learning With a Memory of Diverse Samples (CVPR 2021 · 347 citations)
- Adaptive VIO: Deep Visual-Inertial Odometry with Online Continual Learning (CVPR 2024 · 27 citations)
- Pre-trained Online Contrastive Learning for Insurance Fraud Detection (AAAI 2024 · 22 citations)
Related topics in Continual learning
- Domain and prompt-based (138)
- Class-incremental (450)
- Continual fine-tuning of LLMs (315)
