# Some Insights into Lifelong Reinforcement Learning Systems

**Authors:** Changjian Li  
**arXiv:** [2001.09608](https://arxiv.org/abs/2001.09608)  
**Date:** 2020-01-27  
**Summary page:** https://www.summarizepaper.com/en/paper/2001.09608v1/

## Key points

- Lifelong reinforcement learning systems continuously learn through trial-and-error interactions with the environment over their lifetime.
- Traditional reinforcement learning paradigms do not adequately model lifelong learning processes.
- Maximizing expected cumulative reward may not be suitable for lifelong reinforcement learning as it overlooks individual agent lifespans and internal reward mechanisms.
- There is a need to focus on understanding how agents learn within their own lifespans in order to develop more effective lifelong learning algorithms.

## Summary

The paper "Some Insights into Lifelong Reinforcement Learning Systems" by Changjian Li explores the concept of lifelong reinforcement learning systems. These systems continuously learn through trial-and-error interactions with the environment over their lifetime. However, traditional reinforcement learning paradigms fail to adequately model this type of learning. The author argues that maximizing expected cumulative reward may not be suitable for lifelong reinforcement learning. This is because it does not fully capture the complexities of lifelong learning processes and ignores individual agent lifespans and internal reward mechanisms. The paper highlights the need to shift towards understanding how agents learn within their own lifespans and opens up new avenues for research in developing more effective lifelong learning algorithms.