Towards Learning by Interacting in Cognitive Systems Christensen
Towards Learning by Interacting in Cognitive Systems, Christensen et al. Course: Robots Learning from Humans Kim, Sah-Moo SUALAB & Department of Mathematics Chung-Ang University
Contents n Introduction n Social Learning is Powerful! n Ingredients for Social Learning : 1) Joint Attention n Ingredients for Social Learning : 2) Interaction Management n Conclusion 2
Introduction n Why Learning by Interacting? Learning by Interacting is an agreement that embodiment is a foundation for any kind of interactive learning n What about former intelligence? Mostly, offline learning ☞ Learning ∩ Recognition = Empty! n Where is Social Learning? © 2015, SNU CSE Biointelligence Lab. , http: //bi. snu. ac. kr 3
Social Learning is Powerful! n In Social Learning, particular information facilitates the learning process n And increases its effectivity. n An example on Social Learning? Observation on Parent-Infant model © 2015, SNU CSE Biointelligence Lab. , http: //bi. snu. ac. kr 4
Joint Attention n “Perceptual Magnet Theory” n Kuhl against theory n Question : How can joint attention help to learn things that an infant would not learn in a non-interactive situation? Synchrony and Acoustic packaging help to pay attention n © 2015, SNU CSE Biointelligence Lab. , http: //bi. snu. ac. kr 5
Modeling Joint Attention n Mostly, Bottom-up n On the other hand, purely model-driven jointattention n Multimodality again! Attentional processes need to able to take multi-modal information But synchronization and the handling of different processing rates are non-trivial © 2015, SNU CSE Biointelligence Lab. , http: //bi. snu. ac. kr 6
Interaction Management n What’s Interaction Management? If it fails, then an awkward situation may happen! n ‘Jiggle ‘ and ‘Jiggle-stop’ n Contingent stimulation © 2015, SNU CSE Biointelligence Lab. , http: //bi. snu. ac. kr 7
Modeling IM n Bi-directional interaction Not only focus on the same object with the partner, But also be able to exchange information about it n Grounding process Grounding is a mechanism that may serve as a general mechanism for learning higher-level linguistic interaction through non-linguistic, physical interaction © 2015, SNU CSE Biointelligence Lab. , http: //bi. snu. ac. kr 8
Pre-requisites for Modeling IM n Ability to process multi-modal information in a way allowing to synchronize the information at different levels in order to draw conclusions about co-occurring events n Possibility to develop top-down strategies from bottom-up data analysis n Overall system has to be highly integrated © 2015, SNU CSE Biointelligence Lab. , http: //bi. snu. ac. kr 9
Conclusion n Learning should be modeled as a continuous process driven by social and interactive cues n Complex perception-action system needs to be developed n Remember 2 things : Joint Attention and Interaction Management © 2015, SNU CSE Biointelligence Lab. , http: //bi. snu. ac. kr 10
- Slides: 10