# Marginal Replay vs Conditional Replay for Continual Learning

arXiv: https://arxiv.org/abs/1810.12069 · HTML: https://mlsanity.com/paper/1810.12069/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 29 Oct 2018.

## Built on by (most cited first, 10 of 10)

- [Understanding Continual Learning Settings with Data Distribution Drift Analysis](https://mlsanity.com/paper/2104.01678/) (2021)
- [Continual Learning on Noisy Data Streams via Self-Purified Replay](https://mlsanity.com/paper/2110.07735/) (2021)
- [CRIL: Continual Robot Imitation Learning via Generative and Prediction Model](https://mlsanity.com/paper/2106.09422/) (2021)
- [Sequoia: A Software Framework to Unify Continual Learning Research](https://mlsanity.com/paper/2108.01005/) (2021)
- [Continual Learning in Deep Networks: an Analysis of the Last Layer](https://mlsanity.com/paper/2106.01834/) (2021)
- [Beyond Supervised Continual Learning: a Review](https://mlsanity.com/paper/2208.14307/) (2022)
- [Navigating Memory Construction by Global Pseudo-Task Simulation for Continual Learning](https://mlsanity.com/paper/2210.08442/) (2022)
- [Latent-Insensitive autoencoders for Anomaly Detection](https://mlsanity.com/paper/2110.13101/) (2021)
- [Adiabatic replay for continual learning](https://mlsanity.com/paper/2303.13157/) (2023)
- [Overcoming Domain Drift in Online Continual Learning](https://mlsanity.com/paper/2405.09133/) (2024)

## Abstract (arXiv)

We present a new replay-based method of continual classification learning that we term "conditional replay" which generates samples and labels together by sampling from a distribution conditioned on the class. We compare conditional replay to another replay-based continual learning paradigm (which we term "marginal replay") that generates samples independently of their class and assigns labels in a separate step. The main improvement in conditional replay is that labels for generated samples need not be inferred, which reduces the margin for error in complex continual classification learning tasks. We demonstrate the effectiveness of this approach using novel and standard benchmarks constructed from MNIST and FashionMNIST data, and compare to the regularization-based textitelastic weight consolidation (EWC) method.