Speedup Learning

  title={Speedup Learning},
  author={Alan Fern},
  booktitle={Encyclopedia of Machine Learning},
  • Alan Fern
  • Published in
    Encyclopedia of Machine…
  • Computer Science, Education
Speedup learning is a branch of machine learning that studies learning mechanisms for speeding up problem solvers based on problem solving experience. The input to a speedup learner typically consists of observations of prior problem-solving experience, which may include traces of the problem solver’s operations and/or solutions to solved problems. The output is knowledge that the problem solver can exploit to find solutions more quickly than before learning without seriously effecting solution… 

Learning to Speed Up Structured Output Prediction

This paper trains a speedup classifier that learns to mimic a black-box classifier under the learning-to-search approach and identifies inference situations where it can independently make correct judgments without input features.

Online Speedup Learning for Optimal Planning

A novel method is presented that reduces the cost of combining admissible heuristics for optimal planning, while maintaining its benefits, using an idealized search space model and an active online learning approach for learning a classifier, and employing the learned classifier to decide which heuristic to compute at each state.

What advice would I give a starting graduate student interested in robot learning ? Models ! Model-free !

  • Computer Science
  • 2020
This paper provides a personal survey and retrospective of my work on robot control and learning, and sees that model-free approaches to learning where policies or control laws are directly manipulated to change behavior also play a useful role, and that the authors should combine model-based and model- free approaches tolearning.

Select-and-Evaluate: A Learning Framework for Large-Scale Knowledge Graph Search

This work develops a learning framework for graph search called Select-and-Evaluate (SCALE), and shows that using the learned selection plans, it can significantly improve the computational-efficiency of graph search to achieve high accuracy.

ℋC-search for structured prediction in computer vision

This paper introduces a search operator suited to the vision domain that improves a candidate solution by probabilistically sampling likely object configurations in the scene from the hierarchical Berkeley segmentation, and complements this search operator by applying the DAgger algorithm to robustly train the search heuristic so it learns from its previous mistakes.

HC-Search: Learning Heuristics and Cost Functions for Structured Prediction

A new framework for structured prediction called HC-Search is introduced, which uses a search procedure guided by a learned heuristic H to uncover high quality candidate outputs and then uses a separate learned cost function C to select a final prediction among those outputs.

Learning to Speed Up Query Planning in Graph Databases

This paper presents a Learning to Plan (L2P) framework that is applicable to a large class of query reasoners that follow the Threshold Algorithm approach, and provides a concrete instantiation of the L2P framework for STAR, a state-of-the-art graph query reasoner.



A Formal Framework for Speedup Learning from Problems and Solutions

This work applies a formal framework for learning efficient problem solving from random problems and their solutions to two different representations of learned knowledge, namely control rules and macro-operators, and proves theorems that identify sufficient conditions for learning in each representation.

Learning while solving problems in best first search

It is shown how the knowledge acquisition phase can be integrated with the problem solving phase and a continuous online learning scheme that uses an "anytime" algorithm to learn continuously while solving problems is presented.

Explanation-Based Learning: A Problem Solving Perspective

Machine Learning Methods for Planning

Learning Declarative Control Rules for Constraint-BAsed Planning

This paper presents the first positive results on automatically acquiring high-level, declarative constraints using machine learning techniques, and shows that a new heuristic method for generating training examples together with a rule induction algorithm can learn useful control rules in a variety of domains.

Learning Action Strategies for Planning Domains

Towards Understanding and Harnessing the Potential of Clause Learning

The first precise characterization of clause learning as a proof system (CL) is presented, and it is shown that with a new learning scheme, CL can provide exponentially shorter proofs than many proper refinements of general resolution satisfying a natural property.

Learning Linear Ranking Functions for Beam Search with Application to Planning

Beam search is commonly used to help maintain tractability in large search spaces at the expense of completeness and optimality. Here we study supervised learning of linear ranking functions for

Learning Evaluation Functions for Global Optimization and Boolean Satisfiability

STAGE learns an evaluation function which predicts the outcome of a local search algorithm, such as hillclimbing or WALKSAT, as a function of state features along its search trajectories, and is used to bias future search trajectory toward better optima.

Quantitative Results Concerning the Utility of Explanation-based Learning