%0 Conference Proceedings %T How do People Train a Machine? Strategies and (Mis)Understandings %+ Extreme Situated Interaction (EX-SITU) %+ Laboratoire Interdisciplinaire des Sciences du Numérique (LISN) %+ Institut des Systèmes Intelligents et de Robotique (ISIR) %+ Architectures et Modèles pour l'Interaction (AMI) %+ Sciences et Technologies de la Musique et du Son (STMS) %A Sanchez, Téo %A Caramiaux, Baptiste %A Françoise, Jules %A Bevilacqua, Frédéric %A Mackay, Wendy %Z This research was supported by the ELEMENT project (ANR-18- CE33-0002) from the French National Research Agency and the CNRS-funded project INTACT under the PEPS programme. We want to acknowledge and thank everyone who was involved in each stage of the research, in particular the anonymous reviewers and the participants of the study. We especially want to express our sincere gratitude to Matteo Merzagora, Aude Ghilbert, Paul Boniface and Arnaud Malher from the association TRACES and the Projet Siscode (Horizon 2020 Research and Innovation programme, grant agreement N° 788217), whose collaboration made this study possible. Thanks to Gianni Franchi for his useful thoughts on uncertainty in Deep Neural Networks. %< avec comité de lecture %( ACM on Human-Computer Interaction, Volume 5, Issue CSCW1, April 2021, Article No.: 162, pp 1–26 %B CSCW 2021 - The 24th ACM Conference on Computer-Supported Cooperative Work and Social Computing %C Virtual, United States %8 2021-10-23 %D 2021 %R 10.1145/3449236 %K Interactive Machine Learning %K Human-AI Interaction %K Sketch recognition %K Human-centered analysis %Z Computer Science [cs]/Human-Computer Interaction [cs.HC] %Z Computer Science [cs]/Machine Learning [cs.LG]Conference papers %X Machine learning systems became pervasive in modern interactive technology but provide users with little, if any, agency with respect to how their models are trained from data. In this paper, we are interested in the way novices handle learning algorithms, what they understand from their behavior and what strategy they may use to "make it work". We developed a web-based sketch recognition algorithm based on Deep Neural Network (DNN), called Marcelle-Sketch, that end-users can train incrementally. We present an experimental study that investigate people's strategies and (mis)understandings in a realistic algorithm-teaching task. Our study involved 12 participants who performed individual teaching sessions using a think-aloud protocol. Our results show that participants adopted heterogeneous strategies in which variability affected the model performances. We highlighted the importance of sketch sequencing, particularly at the early stage of the teaching task. We also found that users' understanding is facilitated by simple operations on drawings, while confusions are caused by certain inherent properties of DNN. From these findings, we propose implications for design of IML systems dedicated to novices and discuss the socio-cultural aspect of this research. %G English %2 https://inria.hal.science/hal-03182950/document %2 https://inria.hal.science/hal-03182950/file/V5cscw162-sanchezA.pdf %L hal-03182950 %U https://inria.hal.science/hal-03182950 %~ CNRS %~ INRIA %~ INRIA-SACLAY %~ IRCAM %~ OPENAIRE %~ INRIA_TEST %~ ISIR %~ TESTALAIN1 %~ STMS %~ CENTRALESUPELEC %~ INRIA2 %~ HCI-SORBONNE %~ UNIV-PARIS-SACLAY %~ SORBONNE-UNIVERSITE %~ SORBONNE-UNIV %~ SU-SCIENCES %~ TEST-HALCNRS %~ UNIVERSITE-PARIS-SACLAY %~ SU-TI %~ ANR %~ ISIR_INTERACTIONS_MULTI_ECHELLES %~ LISN %~ GS-ENGINEERING %~ GS-COMPUTER-SCIENCE %~ LISN-EX-SITU %~ LISN-AMI %~ ALLIANCE-SU