# Continual Learning with Deep Generative Replay

**Authors:** Hanul Shin, Jung Kwon Lee, Jaehong Kim, Jiwon Kim  
**arXiv:** [1705.08690](https://arxiv.org/abs/1705.08690)  
**Date:** 2017-05-24  
**Summary page:** https://www.summarizepaper.com/en/paper/1705.08690v1/

## Key points

- Artificial intelligence (AI) has been struggling with a problem called catastrophic forgetting
- Catastrophic forgetting hinders the training of comprehensive AI capable of solving multiple tasks
- A team of researchers led by Hanul Shin, Jung Kwon Lee, Jaehong Kim, and Jiwon Kim propose a novel framework called Deep Generative Replay to address this issue
- The framework consists of two models: a deep generative model (generator) and a task-solving model (solver)
- With only these two models, training data for previous tasks can be easily sampled and interleaved with those for new tasks
- The team tested their methods on several sequential learning settings involving image classification tasks
- Their approach significantly outperformed existing techniques
- This research could pave the way for more efficient and effective AI systems capable of continual learning without catastrophic forgetting.

## Summary

The field of artificial intelligence has been struggling with a problem called catastrophic forgetting, which hinders the training of comprehensive AI capable of solving multiple tasks. To address this issue, a team of researchers led by Hanul Shin, Jung Kwon Lee, Jaehong Kim, and Jiwon Kim propose a novel framework called Deep Generative Replay. This framework consists of two models: a deep generative model (generator) and a task-solving model (solver). With only these two models, training data for previous tasks can be easily sampled and interleaved with those for new tasks. The team tested their methods on several sequential learning settings involving image classification tasks and found that their approach significantly outperformed existing techniques. This research could pave the way for more efficient and effective AI systems capable of continual learning without catastrophic forgetting.