AI Paper Atlas

Research map / Continual learning / Online and replay-based

Learning Probabilistic Prompt for Continual Learning

Hyekang Park, Sanghoon Lee, Geon Lee, Jongyoun Noh, BUMSUB HAM

ECCV 2026 · Continual learning › Online and replay-based · approach: Unified and unsupervised continual frameworks · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
model prompts as Gaussian distributions instead of deterministic vectors (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)FAA: 77.49 -> 78.73Improved
form a mixture of prompt distributions (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)FAA: 78.73 -> 79.98Improved
add distribution regularization loss L_DR (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)FAA: 79.98 -> 80.23Improved
add L_DR without mixture (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)CAA: 82.46 -> 83.80Improved
use single sampled prompt (N_s=1) (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)FAA: fail to convergeHurt
increase N_s from 5 to 30 (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)FAA: 33.55 -> 80.23Improved
increase N_s from 30 to 40 (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)FAA: 80.23 -> 79.96No clear effect
sample prompts instead of using mean mu_GM at training (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)FAA: 79.18 -> 80.23Improved
similarity-weighted aggregation instead of simple average (ablation)ProbPrompt · ImageNet-R 10-task (IN-1K pretrain)FAA: 79.06 -> 79.98Improved
keep only K most relevant distributions in mixture (ablation)ProbPrompt · ImageNet-R 5/10/20-taskFAA: 80.68-80.86 (5-task) vs 80.53No clear effect
ProbPrompt vs VQ-Prompt on fine-grained CUB-200ProbPrompt · CUB-200 10-taskFAA: 86.72 -> 87.52Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy