Results 1 
3 of
3
Learning From Previous Proof Experience: A Survey
, 1999
"... We present an overview of various learning techniques used in automated theorem provers. We characterize the main problems arising in this context and classify the solutions to these problems from published approaches. We analyze the suitability of several combinations of solutions for different app ..."
Abstract

Cited by 4 (0 self)
 Add to MetaCart
We present an overview of various learning techniques used in automated theorem provers. We characterize the main problems arising in this context and classify the solutions to these problems from published approaches. We analyze the suitability of several combinations of solutions for different approaches to theorem proving and place these combinations in a spectrum ranging from provers using very specialized learning approaches to optimally adapt to a small class of proof problems, to provers that learn more general kinds of knowledge, resulting in systems that are less efficient in special cases but show improved performance for a wide range of problems. Finally, we suggest combinations of solutions for various proof philosophies.
Evolving Term Features: First Steps
, 1999
"... Features play a central role in many areas of artificial intelligence and in particular in machine learning. Mostly, features are provided by the user. This is inevitable if the objects that are to be represented with features are not accessible for the computer. Terms are a type of data structure t ..."
Abstract
 Add to MetaCart
(Show Context)
Features play a central role in many areas of artificial intelligence and in particular in machine learning. Mostly, features are provided by the user. This is inevitable if the objects that are to be represented with features are not accessible for the computer. Terms are a type of data structure that is fundamental for a variety of problems such as, for instance, automated theorem proving. In this report we propose to automate the generation of term features using term patterns. The search for a suitable set of features is conducted by a genetic algorithm. 1 Introduction In artificial intelligence (AI), in particular in machine learning, features play a central role. A feature (or attribute) captures a certain aspect of objects under consideration by representing such an aspect with a numerical, categorical, or Boolean feature value. A set of carefully chosen featuresa feature vectorprovides a reasonably abstract view of an object, and presents information in a way that facil...
Finding Simple Proofs In Logic Calculi
, 1998
"... The design of searchguiding heuristics for theorem provers centers on minimizing the time required to find any proof. Mathematicians, however, are also interested in simple proofs. Relevant simplicity criteria like proof length, for instance, hardly play a role in the design of heuristics. In th ..."
Abstract
 Add to MetaCart
(Show Context)
The design of searchguiding heuristics for theorem provers centers on minimizing the time required to find any proof. Mathematicians, however, are also interested in simple proofs. Relevant simplicity criteria like proof length, for instance, hardly play a role in the design of heuristics. In this report we present heuristics designed to find simple proofs and empirically evaluate their performance in the area of logic calculi. The experiments demonstrate that significantly simpler proofs are found without incurring increased search effort in many cases. As a matter of fact, the search for simpler proofs very often succeeds faster than a search guided by a "standard" heuristic based on counting symbols. 1 Introduction Problems related to an inference rule called condensed detachment (CD)also known as substitution and detachmenthave piqued the attention of mathematicians [15, 6] and researchers in the field of automated deduction [11, 17, 9, 13, 18, 2, 16, 4] alike. CD i...