Crowd science for hybrid AI applications

Podrobná bibliografie
Parent link:Service-Oriented System Engineering (SOSE): Proceedings 15th IEEE International conference, 23-26 August 2021, Virtual, Oxford. [P. 172-175].— , 2021
Hlavní autor: Taran Е. А. Ekaterina Aleksandrovna
Korporativní autor: Национальный исследовательский Томский политехнический университет Школа базовой инженерной подготовки Отделение социально-гуманитарных наук
Další autoři: Malanina V. A. Veronika Anatolievna, Casati F. Fabio
Shrnutí:Title screen
Most AI applications are hybrid, that is, employ machines to make inferences but can fall back on humans when the algorithm is not confident enough. This is true for a wide class of applications ranging from self-driving cars to decision making and process automation in enterprise AI. In this WIP paper we present our vision and progress towards an AI and crowd service that trains, assess and refines ML systems intended to be used in hybrid context. We specifically focus on crowdsourcing as a mean to assist ML algorithm development, and on the different ways in which crowd and machine can interact before, during and after the training process in a synergic way that goes well beyond the 'traditional' application of crowd workers to provide data labels for ML training.
Режим доступа: по договору с организацией-держателем ресурса
Jazyk:angličtina
Vydáno: 2021
Témata:
On-line přístup:https://doi.org/10.1109/SOSE52839.2021.00027
Médium: Elektronický zdroj Kapitola
KOHA link:https://koha.lib.tpu.ru/cgi-bin/koha/opac-detail.pl?biblionumber=666777
Popis
Shrnutí:Title screen
Most AI applications are hybrid, that is, employ machines to make inferences but can fall back on humans when the algorithm is not confident enough. This is true for a wide class of applications ranging from self-driving cars to decision making and process automation in enterprise AI. In this WIP paper we present our vision and progress towards an AI and crowd service that trains, assess and refines ML systems intended to be used in hybrid context. We specifically focus on crowdsourcing as a mean to assist ML algorithm development, and on the different ways in which crowd and machine can interact before, during and after the training process in a synergic way that goes well beyond the 'traditional' application of crowd workers to provide data labels for ML training.
Режим доступа: по договору с организацией-держателем ресурса
DOI:10.1109/SOSE52839.2021.00027