Learning Dynamic Belief Graphs to Generalize on Text-Based Games

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F20%3A10432150" target="_blank" >RIV/00216208:11320/20:10432150 - isvavai.cz</a>

Result on the web

<a href="https://proceedings.neurips.cc/paper/2020/hash/1fc30b9d4319760b04fab735fbfed9a9-Abstract.html" target="_blank" >https://proceedings.neurips.cc/paper/2020/hash/1fc30b9d4319760b04fab735fbfed9a9-Abstract.html</a>

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Learning Dynamic Belief Graphs to Generalize on Text-Based Games

Original language description

Playing text-based games requires skills in processing natural language and sequential decision making. Achieving human-level performance on text-based games remains an open challenge, and prior research has largely relied on hand-crafted structured representations and heuristics. In this work, we investigate how an agent can plan and generalize in text-based games using graph-structured representations learned end-to-end from raw text. We propose a novel graph-aided transformer agent (GATA) that infers and updates latent belief graphs during planning to enable effective action selection by capturing the underlying game dynamics. GATA is trained using a combination of reinforcement and self-supervised learning. Our work demonstrates that the learned graph-based representations help agents converge to better policies than their text-only counterparts and facilitate effective generalization across game configurations. Experiments on 500+ unique games from the TextWorld suite show that our best agent outperforms text-based baselines by an average of 24.2%.

Czech name

—

Czech description

—

Type

D - Article in proceedings

CEP classification

—

OECD FORD branch

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

Project

—

Continuities

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Publication year

2020

Confidentiality

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

Article name in the collection

Advances in Neural Information Processing Systems 33

ISBN

978-1-71382-954-6

ISSN

1049-5258

e-ISSN

—

Number of pages

13

Pages from-to

3045-3057

Publisher name

Curran Associates, Inc.

Place of publication

Neuveden

Event location

Virtuální

Event date

Dec 6, 2020

Type of event by nationality

WRD - Celosvětová akce

UT code for WoS article

—