Research map / Continual learning / Online and replay-based
Online Task-Free Continual Learning with Dynamic Sparse Distributed Memory
Julien Pourcel, Ngoc-Son Vu, Robert M. French
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace SDM with dynamic DSDM (half the memory size) | SDM · Split MNIST (step=2) | last accuracy: 74.2 -> 94.0 | Improved |
| use L2 distance instead of city-block distance (ablation) | DSDM · Split MNIST | last accuracy | Improved |
| use DSDM instead of CV on CORE-50 (buffer 5k) | CV · CORE-50, buffer 5k | last accuracy: 43.1 -> 57.1 | Improved |
| use DSDM instead of CV on CORE-50 (buffer 1k) | CV · CORE-50, buffer 1k | average accuracy: 45.1 -> 43.9 | Hurt |
| use DSDM instead of CV on CIFAR-10 (buffer 5k) | CV · CIFAR-10 step=2, buffer 5k | last accuracy: 77.0 -> 76.0 | No clear effect |
| use DSDM instead of CV on CIFAR-10 (buffer 1k) | CV · CIFAR-10 step=2, buffer 1k | last accuracy: 62.9 -> 67.0 | Improved |
| increase DSDM buffer size from 1k to 5k | DSDM · CORE-50 | last accuracy: 43.3 -> 57.1 | Improved |
| know task boundaries (task-based instead of task-free) | MIR · CIFAR-10 | last accuracy: 42.8 -> 52.2 | Improved |
| know task boundaries (task-based instead of task-free) | GSS · CIFAR-10 | last accuracy: 38.4 -> 56.7 | Improved |
| replace ILUGAN with online DSDM | ILUGAN · CIFAR-100 step 2 | average accuracy: 58.0 -> 63.3 | Improved |
| replace ILUGAN with online DSDM | ILUGAN · CUB-200 step 5 | average accuracy: 54.9 -> 55.2 | Improved |
| use DSDM+ViT instead of Ensemble under Gaussian schedule | Ensemble · CIFAR-10 Gaussian schedule | last accuracy: 50.1 -> 84.9 | Improved |
Closest papers by meaning
- Online Task-Free Continual Learning via Dynamic Expansionable Memory Distribution (CVPR 2025 · similarity 0.89)
- Task-Free Dynamic Sparse Vision Transformer for Continual Learning (AAAI 2024 · similarity 0.88)
- Dynamic Expansion Diffusion Learning for Lifelong Generative Modelling (AAAI 2025 · similarity 0.88)
- Federated Continual Learning Goes Online: Uncertainty-Aware Memory Management for Vision Tasks and Beyond (ICLR 2025 · similarity 0.87)
- Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation (ICLR 2024 · similarity 0.87)
- Task-Free Continual Generation and Representation Learning via Dynamic Expansionable Memory Cluster (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
