Lifelong learning in costly feature spaces

DOI10.1016/J.TCS.2019.11.010zbMATH Open1436.68292arXiv1706.10271OpenAlexW2985865902MaRDI QIDQ2290680FDOQ2290680

Authors: Maria-Florina Balcan, Vaishnavh Nagarajan, Avrim Blum 

Publication date: 29 January 2020

Published in: Theoretical Computer Science (Search for Journal in Brave)

Abstract: An important long-term goal in machine learning systems is to build learning agents that, like humans, can learn many tasks over their lifetime, and moreover use information from these tasks to improve their ability to do so efficiently. In this work, our goal is to provide new theoretical insights into the potential of this paradigm. In particular, we propose a lifelong learning framework that adheres to a novel notion of resource efficiency that is critical in many real-world domains where feature evaluations are costly. That is, our learner aims to reuse information from previously learned related tasks to learn future tasks in a feature-efficient manner. Furthermore, we consider novel combinatorial ways in which learning tasks can relate. Specifically, we design lifelong learning algorithms for two structurally different and widely used families of target functions: decision trees/lists and monomials/polynomials. We also provide strong feature-efficiency guarantees for these algorithms; in fact, we show that in order to learn future targets, we need only slightly more feature evaluations per training example than what is needed to predict on an arbitrary example using those targets. We also provide algorithms with guarantees in an agnostic model where not all the targets are related to each other. Finally, we also provide lower bounds on the performance of a lifelong learner in these models, which are in fact tight under some conditions.

Full work available at URL: https://arxiv.org/abs/1706.10271

Recommendations

• Lifelong learning in costly feature spaces
• Towards knowledgeable supervised lifelong learning systems
• Multi-task and lifelong learning of kernels
• PAC-Bayesian lifelong learning for multi-armed bandits
• A Bayesian/information theoretic model of learning to learn via multiple task sampling

zbMATH Keywords

lifelong learningrepresentation learningcostly features

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05)

Cites Work

• Title not available (Why is that?)
• Lifelong planning \(\text{A}^*\)
• Multi-task learning for classification with Dirichlet process priors
• Title not available (Why is that?)
• Title not available (Why is that?)
• Data Mining with Decision Trees
• A Bayesian/information theoretic model of learning to learn via multiple task sampling
• Title not available (Why is that?)
• Learning sparse polynomial functions

Cited In (6)

• Title not available (Why is that?)
• PAC-Bayesian lifelong learning for multi-armed bandits
• Deep Bayesian unsupervised lifelong learning
• Towards knowledgeable supervised lifelong learning systems
• Lifelong learning in costly feature spaces
• Adversarial Feature Alignment: Avoid Catastrophic Forgetting in Incremental Task Lifelong Learning

This page was built for publication: Lifelong learning in costly feature spaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2290680)