BIBLIOS

Tipo
Artigos em Conferência

Tipo de Documento
Artigo Completo

Título
One Arm to Rule Them All: Online Learning with Multi-armed Bandits for Low-Resource Conversational Agents

Participantes na publicação
Vânia Mendonça (Author)
Luísa Coheur (Author)
INSTITUTO SUPERIOR TÉCNICO
Alberto Sardinha (Author)
INSTITUTO SUPERIOR TÉCNICO

Resumo
In a low-resource scenario, the lack of annotated data can be an obstacle not only to train a robust system, but also to evaluate and compare different approaches before deploying the best one for a given setting. We propose to dynamically find the best approach for a given setting by taking advantage of feedback naturally present on the scenario in hand (when it exists). To this end, we present a novel application of online learning algorithms, where we frame the choice of the best approach as a multi-armed bandits problem. Our proof-of-concept is a retrieval-based conversational agent, in which the answer selection criteria available to the agent are the competing approaches (arms). In our experiment, an adversarial multi-armed bandits approach converges to the performance of the best criterion after just three interaction turns, which suggests the appropriateness of our approach in a low-resource conversational agent.


Evento
EPIA Conference on Artificial Intelligence

Identificadores da Publicação
ISSN - 0302-9743

Editora
Springer International Publishing


Identificadores do Documento
DOI - https://doi.org/10.1007/978-3-030-86230-5_49
URL - http://dx.doi.org/10.1007/978-3-030-86230-5_49

Keywords
Online learning Multi-armed bandits Conversational agents