Latent Properties of Lifelong Learning Systems

TL;DR

This paper introduces an explainable surrogate-modeling approach to estimate latent properties of lifelong learning algorithms, addressing confounding factors in existing metrics and validated through synthetic and real data experiments.

Contribution

It presents a novel, algorithm-agnostic method for analyzing lifelong learning systems, improving understanding of their intrinsic properties.

Findings

Effective estimation of latent properties demonstrated on synthetic data.

Analysis of real lifelong learning approaches validates the surrogate model.

Addresses confounding effects of task and scenario structure in metrics.

Abstract

Creating artificial intelligence (AI) systems capable of demonstrating lifelong learning is a fundamental challenge, and many approaches and metrics have been proposed to analyze algorithmic properties. However, for existing lifelong learning metrics, algorithmic contributions are confounded by task and scenario structure. To mitigate this issue, we introduce an algorithm-agnostic explainable surrogate-modeling approach to estimate latent properties of lifelong learning algorithms. We validate the approach for estimating these properties via experiments on synthetic data. To validate the structure of the surrogate model, we analyze real performance data from a collection of popular lifelong learning approaches and baselines adapted for lifelong classification and lifelong reinforcement learning.