Social Robots for (Second) Language Learning in (Migrant) Primary School Children

International Journal of Social Robotics, Год журнала: 2021, Номер 14(3), С. 827 - 843

Опубликована: Окт. 11, 2021

Abstract Especially these days, innovation and support from technology to relieve pressure in education is highly urgent. This study tested the potential advantage of a social robot over tablet (second) language learning on performance, engagement, enjoyment. Shortages primary call for new solutions. Previous studies combined robots with tablets, compensate robot’s limitations, however, this applied direct human–robot interaction. Primary school children ( N = 63, aged 4–6) participated 3-wave field experiment story-telling exercises, either semi-autonomous (without tablet, using WOz) or tablet. Results showed increased gains time when training robot, compared Children who trained were more engaged task enjoyed it more. Robot’s behavioral style (social neutral) hardly differed overall, seems vary high versus low educational abilities. While need sophistication before being implemented schools, our shows as tutors learning.

Язык: Английский

---

Personalized Estimation of Engagement From Videos Using Active Learning With Deep Reinforcement Learning

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Год журнала: 2019, Номер unknown, С. 217 - 226

Опубликована: Июнь 1, 2019

Perceiving users' engagement accurately is important for technologies that need to respond learners in a natural and intelligent way. In this paper, we address the problem of automated estimation from videos child-robot interactions recorded unconstrained environments (kindergartens). This challenging due diverse person-specific styles expressions through facial body gestures, as well because illumination changes, partial occlusion, changing background classroom each child active. To tackle these difficult challenges, propose novel deep reinforcement learning architecture active video data. The key our approach personalized policy enables model decide whether estimate child's level (low, medium, high) or, when uncertain, query human label. Queried are labeled by expert an offline manner, used personalize classifier target over time. We show on database 43 children involved robot-assisted activities (8 sessions 3 months), combined human-AI can easily adapt its interpretations using only handful videos, while being robust many complex influences results large improvements non-personalized traditional methods.

Язык: Английский

---

Multi-modal Active Learning From Human Data: A Deep Reinforcement Learning Approach

INTERNATIONAL CONFERENCE ON MULTIMODAL INTERACTION, Год журнала: 2019, Номер unknown, С. 6 - 15

Опубликована: Окт. 14, 2019

Human behavior expression and experience are inherently multimodal, characterized by vast individual contextual heterogeneity. To achieve meaningful human-computer human-robot interactions, multi-modal models of the user's states (e.g., engagement) therefore needed. Most existing works that try to build classifiers for assume data train fully labeled. Nevertheless, labeling is costly tedious, also prone subjective interpretations human coders. This even more pronounced when some users expressive with their facial expressions, voice). Thus, building can accurately estimate during an interaction challenging. tackle this, we propose a novel active learning (AL) approach uses notion deep reinforcement (RL) find optimal policy selection data, needed target (modality-specific) models. We investigate different strategies fusion, show proposed model-level fusion coupled RL outperforms feature-level modality-specific models, naïve AL such as random sampling, standard heuristics uncertainty sampling. benefits this on task engagement estimation from real-world child-robot interactions autism therapy. Importantly, be used efficiently personalize user using small amount actively selected data.

Язык: Английский

---

Robot-based play-drama intervention may improve the narrative abilities of Chinese-speaking preschoolers with autism spectrum disorder

Research in Developmental Disabilities, Год журнала: 2019, Номер 95, С. 103515 - 103515

Опубликована: Окт. 24, 2019

Язык: Английский

---

A reinforcement learning based algorithm for personalization of digital, just-in-time, adaptive interventions

Artificial Intelligence in Medicine, Год журнала: 2021, Номер 115, С. 102062 - 102062

Опубликована: Апрель 2, 2021

Язык: Английский

---

Education Data Science: Past, Present, Future

AERA Open, Год журнала: 2021, Номер 7

Опубликована: Янв. 1, 2021

This AERA Open special topic concerns the large emerging research area of education data science (EDS). In a narrow sense, EDS applies statistics and computational techniques to educational phenomena questions. broader it is an umbrella for fleet new being used identify forms data, measures, descriptives, predictions, experiments in education. Not only are old questions analyzed ways but also based on novel discoveries from techniques. overview defines field discusses 12 articles that illustrate AERA-angle EDS. Our relates variety promises poses as well areas where scholars could successfully focus going forward.

Язык: Английский

---

MultiMediate '23: Engagement Estimation and Bodily Behaviour Recognition in Social Interactions

Опубликована: Окт. 26, 2023

Automatic analysis of human behaviour is a fundamental prerequisite for the creation machines that can effectively interact with- and support humans in social interactions. In MultiMediate'23, we address two key tasks first time controlled challenge: engagement estimation bodily recognition This paper describes MultiMediate'23 challenge presents novel sets annotations both tasks. For collected on NOvice eXpert Interaction (NOXI) database. recognition, annotated test recordings MPIIGroupInteraction corpus with BBSI annotation scheme. addition, present baseline results

Язык: Английский

---

Collaborative Storytelling with Large-scale Neural Language Models

Опубликована: Окт. 16, 2020

Storytelling plays a central role in human socializing and entertainment. However, much of the research on automatic storytelling generation assumes that stories will be generated by an agent without any interaction. In this paper, we introduce task collaborative storytelling, where artificial intelligence person collaborate to create unique story taking turns adding it. We present system which works with storyteller generating new utterances based so far. constructed tuning publicly-available large scale language model dataset writing prompts their accompanying fictional works. identify sufficiently human-like important technical issue propose sample-and-rank approach improve utterance quality. Quantitative evaluation shows our outperforms baseline, qualitative system's capabilities.

Язык: Английский

---

Robotics in Healthcare

Intelligent systems reference library, Год журнала: 2021, Номер unknown, С. 281 - 306

Опубликована: Ноя. 26, 2021

Язык: Английский

---

Designing for Caregiving: Integrating Robotic Assistance in Senior Living Communities

Designing Interactive Systems Conference, Год журнала: 2022, Номер unknown

Опубликована: Июнь 12, 2022

Robots hold significant promise to assist with providing care an aging population and help overcome increasing caregiver demands. Although a large body of research has explored robotic assistance for individuals disabilities age-related challenges, this past work focuses primarily on building capabilities not yet fully considered how these could be used by professional caregivers. To better understand the workflows practices caregivers who support populations determine can integrated into their work, we conducted field study using ethnographic co-design methods in senior living community. From our results, created set design opportunities assistance, which organized three different parts: supporting workflows, adapting resident abilities, feedback all stakeholders interaction.

Язык: Английский

---