# Lifelong Learning Metrics

**Authors:** Alexander New, Megan Baker, Eric Nguyen, Gautam Vallabha  
**arXiv:** [2201.08278](https://arxiv.org/abs/2201.08278)  
**Date:** 2022-01-20  
**Summary page:** https://www.summarizepaper.com/en/paper/2201.08278v1/

## Key points

- The DARPA Lifelong Learning Machines (L2M) program aims to advance artificial intelligence (AI) systems by enabling them to continuously learn and improve.
- AI systems developed under the program can perform various functions, such as autonomous driving, real-time strategy, and drone simulation.
- Measuring system performance across different settings with diverse characteristics has been a challenge for the testing and evaluation team.
- A framework is provided in this document for evaluating AI systems' abilities to learn continuously while adapting to new tasks over time.
- The framework offers insights into how to design effective lifelong learning algorithms that can adapt to changing environments and improve performance over extended periods.
- This work could lead to significant advancements in AI systems capable of lifelong learning and adaptation.

## Summary

The DARPA Lifelong Learning Machines (L2M) program aims to advance artificial intelligence (AI) systems by enabling them to continuously learn and improve, using data from one task to enhance performance on another in a computationally sustainable manner. The program's performers have developed AI systems capable of performing various functions, such as autonomous driving, real-time strategy, and drone simulation. Measuring system performance across different settings with diverse characteristics has been a challenge for the testing and evaluation team. To address this issue, this document provides a framework for evaluating AI systems' abilities to learn continuously while adapting to new tasks over time. It offers insights into how to design effective lifelong learning algorithms that can adapt to changing environments and improve performance over extended periods. Ultimately, this work could lead to significant advancements in AI systems capable of lifelong learning and adaptation.