Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online Learning with Improving Agents: Multiclass, Budgeted Agents and Bandit Learners

Sajad Ashkezari, Shai Ben-David

TL;DR

This work extensively extends previously published results by providing combinatorial dimensions that characterize online learnability in this model, by analyzing the multiclass setup, learnability in a bandit feedback setup, modeling agents'cost for making improvements and more.

Abstract

We investigate the recently introduced model of learning with improvements, where agents are allowed to make small changes to their feature values to be warranted a more desirable label. We extensively extend previously published results by providing combinatorial dimensions that characterize online learnability in this model, by analyzing the multiclass setup, learnability in a bandit feedback setup, modeling agents' cost for making improvements and more.

Online Learning with Improving Agents: Multiclass, Budgeted Agents and Bandit Learners

TL;DR

This work extensively extends previously published results by providing combinatorial dimensions that characterize online learnability in this model, by analyzing the multiclass setup, learnability in a bandit feedback setup, modeling agents'cost for making improvements and more.

Abstract

We investigate the recently introduced model of learning with improvements, where agents are allowed to make small changes to their feature values to be warranted a more desirable label. We extensively extend previously published results by providing combinatorial dimensions that characterize online learnability in this model, by analyzing the multiclass setup, learnability in a bandit feedback setup, modeling agents' cost for making improvements and more.
Paper Structure (9 sections, 8 theorems, 1 equation, 3 algorithms)

This paper contains 9 sections, 8 theorems, 1 equation, 3 algorithms.

Table of Contents

  1. Introduction
  2. Notation and Setup
  3. Binary Classes
  4. Full Feedback Multiclass - Unweighted Graph
  5. Bandit Feedback - Unweighted Graph
  6. Price of Bandit Feedback
  7. Full Feedback - Weighted Graph
  8. Conclusion
  9. Acknowledgments

Key Result

Theorem 5

The optimal number of mistakes in the realizable online learning with improvement setting for deterministic learners is $\mathrm{ILdim}(\mathcal{H})$.

Theorems & Definitions (23)

  • Definition 3: Improvement Littlestone Tree ($\mathrm{ILT}$)
  • Definition 4: Improvement Littlestone Dimension $\mathrm{ILdim}$
  • Theorem 5
  • proof
  • Definition 6: Multiclass ILT
  • Definition 7: Multiclass $\mathrm{ILdim}$
  • Lemma 8
  • proof
  • Lemma 9
  • proof
  • ...and 13 more