Combining Top-Down and Bottom-Up Approaches for Automated Discovery of Typed Programs

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F67985807%3A_____%2F17%3A00490832" target="_blank" >RIV/67985807:_____/17:00490832 - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1109/SSCI.2017.8285209" target="_blank" >http://dx.doi.org/10.1109/SSCI.2017.8285209</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1109/SSCI.2017.8285209" target="_blank" >10.1109/SSCI.2017.8285209</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Combining Top-Down and Bottom-Up Approaches for Automated Discovery of Typed Programs

Popis výsledku v původním jazyce

Automated program discovery has been mainly approached via Genetic Programming, which represents the search space of programs implicitly by a collection of individuals. In this paper, we propose a way of representing program search space explicitly, with a top-down hierarchy of semi-constructed programs. Together with a bottom-up generation procedure, we maintain an overview of the overall search space structure, while being able to quickly get samples from the fringe. Moreover, having a type system with parametric polymorphism allows us to limit the state space size using type level programming while keeping a complete control over program sizes. Our approach can naturally be used for searching the space of programs using tree search methods. We back this claim by a simple experiment utilizing a Monte-Carlo Tree Search algorithm, though other search methods (such as A*) might be used as well.

Název v anglickém jazyce

Combining Top-Down and Bottom-Up Approaches for Automated Discovery of Typed Programs

Popis výsledku anglicky

Automated program discovery has been mainly approached via Genetic Programming, which represents the search space of programs implicitly by a collection of individuals. In this paper, we propose a way of representing program search space explicitly, with a top-down hierarchy of semi-constructed programs. Together with a bottom-up generation procedure, we maintain an overview of the overall search space structure, while being able to quickly get samples from the fringe. Moreover, having a type system with parametric polymorphism allows us to limit the state space size using type level programming while keeping a complete control over program sizes. Our approach can naturally be used for searching the space of programs using tree search methods. We back this claim by a simple experiment utilizing a Monte-Carlo Tree Search algorithm, though other search methods (such as A*) might be used as well.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

<a href="/cs/project/GA15-19877S" target="_blank" >GA15-19877S: Automatické modelování znalostí a plánů pro autonomní roboty</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2017

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

SSCI 2017 IEEE Symposium Series on Computational Intelligence

ISBN

978-1-5386-2726-6

ISSN

—

e-ISSN

—

Počet stran výsledku

8

Strana od-do

938-945

Název nakladatele

IEEE

Místo vydání

New York

Místo konání akce

Honolulu

Datum konání akce

27. 11. 2017

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

000428251401003