AI Paper Atlas

Research map / Continual learning / Online and replay-based

PAC-Bayes bounds for cumulative loss in Continual Learning

Lior Friedman, Ron Meir

ICLR 2026 · Continual learning › Online and replay-based · approach: Unified and unsupervised continual frameworks

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
use VI (optimizes the bound) instead of EWCPAC-Bayes bound · Split-MNISTbound gap (CuL vs bound): EWC 0.9 ± 0.0 vs 4.2 ± 0.1; VI 17.6 ± 0.4 vs 19.4 ± 0.3Mixed
compare bound at final task versus averaged over tasksPAC-Bayes bound · Split-MNIST VIbound: 19.4 ± 0.3 -> 7.5 ± 0.5Improved
compare bound at final task versus averaged over tasksPAC-Bayes bound · Permuted-MNIST EWCbound: 10.6 ± 0.2 -> 5.1 ± 0.2Improved
compare bound at final task versus averaged over tasksPAC-Bayes bound · Split-CIFAR10 VIbound: 52.2 ± 0.2 -> 52.5 ± 0.2Improved
add experience replay to SGDSGD · Split-ImageNetforgetting: 7.8 ± 0.1 -> 2.8 ± 0.1Improved
add experience replay to SGDSGD · Split-ImageNetcumulative error: 34.8 ± 0.3 -> 55.7 ± 0.7Hurt
use EWC instead of SGDSGD · Split-ImageNetforgetting: 7.8 ± 0.1 -> 5.7 ± 0.4Improved
use over-parametrized deep model instead of under-parametrizedoracle bound · Swap tasks regressioncumulative error: 0.11 ± 0.01 -> 0.01 ± 0.00Improved
alternate tasks instead of one swap in linear over-parametrized regressionoracle bound · linear regressioncumulative error: 10.3 ± 0.2 -> 280.1 ± 5.4Hurt
gradual task change instead of swap in linear regressionoracle bound · linear regressioncumulative error: 10.3 ± 0.2 -> 4.7 ± 0.2Improved

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